Changelog (development)

×

v.81.71 (2026-05-17)

• CN matrix — drop the WGS-vs-WES per-line provenance tag. The values themselves come uniformly from DepMap OmicsCNGene.csv (24Q4 release), which is already a per-line hybrid by DepMap’s own construction. The earlier “WGS” / “WES” pill was set by looking up each line in the newer OmicsCNGeneWGS 25Q3 file — one release ahead of the values — which created false precision (a line whose 24Q4 value was still WES-derived could be flagged as WGS because DepMap had WGS’d it since). Cleaner story: one source, one sentence. Affects both Correlate V2 (small italic “wes” tag on CN sort-value pills removed) and Green Listed (per-column WGS / WES badges in the CN results table removed).

v.81.70 (2026-05-17)

• CN matrix — gene list now cross-referenced against the greenlisted human CRISPR libraries (Brunello, Gattinara, GeCKO v2, Jacquere, VBC, MinLibCas9, TKOv3, Yusa v1), expanded through the human synonym table so an alias hit counts. Anything not reachable by any sgRNA in any of those libraries is dropped — in practice this removes residual pseudogenes (MARK2P17, NPM1P28, ...), Ig/TR gene segments (IGHJ2, TRBC1), MTND* mitochondrial pseudogenes, RN7SL / RNA5SP / SNRPEP small-RNA pseudogenes, and obscure LINC entries with no library coverage. Real genes including all the previously-recovered ones (TP53, TP63, TP73, USP1, USP7, BMP1–BMP7, AKAP1–AKAP12, CASP3, MARK2 etc.) are kept. Gene count 32 174 → 20 794 (close to the realistic ~20 k protein-coding count), binary 92 → 59 MB.

v.81.69 (2026-05-17)

• CN matrix — bug fix: TP53 was missing. Cause: an over-broad pseudogene regex (P\d+$) that was meant to drop ribosomal pseudogenes like RPL32P26 / RPS12P2 also matched any gene symbol ending with P + digits: TP53, TP63, TP73, USP1–USP54, BMP1–BMP15, AKAP1–AKAP14, GAPDHP* etc. Replaced with narrow patterns anchored to specific known pseudogene prefixes (RPL, RPS, HNRNP, HBA / HBB, EEF1, TUBB, GAPDH). Gene count 24 567 → 32 174 (recovered 7 607 real genes). Binary 70 → 92 MB.

v.81.68 (2026-05-17)

• CN matrix — widened gene coverage. Filter relaxed from “WGS-set only” to “all parseable symbols minus noise patterns”. Drops obvious clutter (LOC IDs, RNU / SNORA / SNORD small nuclear RNAs, antisense / divergent-transcript suffixes, pseudogenes ending in P + digits) but keeps protein-coding + named lncRNAs + miRNA hosts + olfactory receptors + Ig/TCR segments. Gene count 18 660 → 24 567. Binary 53 → 70 MB.
• New CN-based collections (20). Per-gene focal-amplification filters added for AR, BCL2, EGFR, KIT, KRAS, MITF, PDGFRA, SOX2. New “Focal deletions” category with per-gene filters for CDKN2A, CDKN2B, RB1, PTEN, BRCA1, BRCA2, ATM, ARID1A, SMARCA4, STK11, KEAP1, BAP1 plus an “Any focal deletion” catch-all. Each collection filters lines with the gene at amp tier (CN ≥ 3) or deletion tier (CN ≤ 0.5) on the curated panel. Distinct from the integrated TSG functional-loss collections (which combine CN + mutation + expression); these are CN-only deep deletions.

v.81.67 (2026-05-17)

• CN matrix upgraded to hybrid WGS-primary / WES-fallback. Source: DepMap OmicsCNGene.csv (24Q4) — DepMap’s own merged file that combines WGS and WES inferences. Cell-line coverage jumps from 1095 (WGS-only) to 1929 (WGS + WES). Lines that were never WGS’d — Jurkat, K562, many older haematopoietic lines — now appear in the CN sort. Per-line provenance preserved: lines also in OmicsCNGeneWGS.csv (25Q3) are tagged WGS; lines absent get a small wes caveat next to the CN tier pill, hover for the explanation. Binary blob grew from 32 → 55 MB to accommodate the extra ~834 cell lines.

v.81.66 (2026-05-16)

• URL-hash deep link — loading the app with #cell (or #cellbrowser) in the URL opens the cell-line browser modal automatically. So correlate.cmm.se/#cell jumps straight to the browser, useful for linking from sibling apps (Greenlisted, MouseCLB, lab pages). Also responds to in-app hash changes via hashchange.

v.81.65 (2026-05-13)

• CN values in the cell-line list now show a biological tier label instead of just a 2-decimal number. Tiers: deep del < 0.3, het loss 0.3–0.7, WT 0.7–1.3, low gain 1.3–2.0, gain 2.0–3.0, amp 3.0–5.0, strong amp ≥ 5.0. Each tier carries its own coloured pill (red for losses, gray for WT, blue for gains, darker / saturated blue for amp / strong amp). The raw number is kept as a small de-emphasised suffix and in the tooltip for anyone who wants the underlying value. Previously the raw “1.48” / “1.77” type values read as sequencing noise rather than biology — this maps each line to a discrete biological category. Caption row updated to list the tier cutoffs.

v.81.64 (2026-05-13)

• Fix — CLB list caption (“Values shown: ...”) now spans the full width of the row-major grid instead of occupying the first cell. Previously it consumed grid position 1, pushing the first cell line into position 2 of the top row. Caption updated to reflect matrix-vs-panel scope dynamically (full DepMap matrix when loaded, curated panel only otherwise).

v.81.63 (2026-05-13)

• CN sort opened to any gene. The full DepMap OmicsCNGeneWGS matrix (19366 genes × 1095 cell lines) is now processed into web_data/cn.bin.gz (32 MB) with the same gene-major int16 + scaleFactor encoding as geneEffects.bin.gz. App lazy-loads the matrix on first CN-sort use; the curated-panel data still serves as the pre-load fallback so MYC / ERBB2 / CDKN2A / etc. work instantly while the matrix is downloading. Picker now lists the curated panel at the top (with event counts) and any matrix gene below by substring match — pick any gene for CRISPR-targeting reference (high CN often means more guides needed for biallelic knockout).

v.81.62 (2026-05-13)

• Drug-response colours aligned between the cell-line list and the sort-drug dropdown. The list previously used AUC < 0.6 = green (anything that “kills many cells”), so a row with AUC = 0.5 showed green even though the dropdown counted it as “partly sensitive” (amber). Now both use the same thresholds: green = very-sensitive (AUC < 0.3), amber = partly-sensitive (0.3 – 0.6), light gray = modest, very light gray = little effect.
• Broadly-cytotoxic flag in the drug-sort dropdown. Compounds where ≥ 40 % of the cohort scores very-sensitive get a broadly cytotoxic % badge. Romidepsin (HDAC1/2) is very-sensitive in 304 / 386 (79 %) lines, Gemcitabine (dNTP depletion) in 274 / 389 (70 %) — broadly toxic chemotherapy signatures, not targeted-therapy candidates. Sort order now puts targeted hits (Trametinib / Dasatinib / Dabrafenib etc.) above broadly-cytotoxic ones with the same raw v-count, so the top of the dropdown surfaces actually selective therapies first. Header copy updated to explain the threshold.

v.81.61 (2026-05-13)

• Wiki drug-response section now shows a per-compound AUC histogram, mirroring the genome-section pattern. Each “Standout sensitive” and “Standout resistant” row gets a thin Plotly histogram of the AUC distribution across all PRISM-tested cell lines (grey bars), with a red dashed line at this cell line’s AUC. Lets you see at a glance whether a low AUC is part of a long tail (broad killer) or genuinely isolated (selective hit), the same way the Ploidy / Aneuploidy / CIN histograms work.

v.81.60 (2026-05-13)

• New sort option: “Sort: Copy number of gene”. Pick a gene from the curated cancer-relevant CN panel (33 amp + 14 del genes from clinical_cn.json — MYC, ERBB2, CDK4, MDM2, EGFR, CDKN2A, BRCA1/2, etc.) and the cell-line list sorts by that gene’s copy number. Picker shows per-gene event counts (e.g. “MYC amp — 87 lines”) so you can see at a glance which genes have the most flagged lines. Per-row CN value is coloured blue for amplification (CN ≥ 3) and red for deletion (CN ≤ 0.5). Lines without a significant CN event for the selected gene show “—” at the end of the list. Scope note: the curated panel covers ~47 clinically actionable cancer genes; sorting by arbitrary genes (any of the 25 k in the genome) would need processing the full DepMap OmicsCNGene matrix into a binary blob like geneEffects.bin.gz — queued as a future addition.

v.81.59 (2026-05-13)

• CLB detail card — Immunology section restyle. The previous layout placed the badge (“MSI-high” / “Class-I lost”) and the explanation sentence in an inline flex row, squeezing the sentence into whatever horizontal space was left next to the badge — awkward in the 280 px detail pane. The two now stack vertically (badge above, sentence below), matching the layout of the other sections (Hotspot Mutations, Functional loss, Genome) and giving the caption full width to wrap naturally. Section header gained a flag count and a ? help icon explaining that these are cell-intrinsic inferences (what a CD8+ T cell could see if it got there), not tumour-microenvironment readouts — most human cancer cell lines can only be grown in immunocompromised hosts in vivo.

v.81.58 (2026-05-13)

• Fix — hovering a functional-loss or focal-CN gene chip in the CLB detail card now shows a single combined tooltip instead of two overlapping popups. The chips previously had both a custom gene tooltip (NCBI summary) and a native browser title attribute (why the gene is highlighted); the browser tooltip overlay would partially cover the custom one. Now the “why this gene is flagged” line is rendered inside the custom tooltip as a red-bordered context block above the gene info.

v.81.57 (2026-05-13)

• Wiki “Strong drug-target candidates” now requires PRISM corroboration before naming a drug class. Previously a collection-based call (e.g. A375 flagged as CDK4/6-dependent from CDKN2A loss + CRISPR signal) was added unconditionally, so the summary could claim “CDK4/6 inhibitors” while the drug-response section below showed no CDK4/6 inhibitor in the top-sensitive list. Now, for any drug class PRISM tested (CDK4/6, MEK, BRAF, EGFR, PI3K/AKT/mTOR, ABL), the claim is emitted only if at least one matching compound is sensitive (z < −1σ) in this cell line. Classes PRISM doesn’t test at all (e.g. KRAS-G12C inhibitors) are still emitted so legitimate biology isn’t suppressed.
• “Useful as a model for” softened to “May be useful as a model for”. The inference is built from mutation + dependency layers without prospective validation; the modal verb is more honest about the claim.
• “Context-aware cross-checks” renamed to “Targeted therapies matched to this cell line’s alterations” and given a one-line explanation: “For each driver alteration detected in this cell line, the PRISM compounds with a matching target are listed below, sorted by sensitivity in this line (most-sensitive first).” Previously the section header gave no hint of what cross-check meant.

v.81.56 (2026-05-13)

• CLB list flow changed from CSS multi-column (column-major fill) to CSS grid (row-major fill). The list now reads left-to-right then wraps to the next row, so when sorted descending the highest value is in the upper-left and the actual lowest sits at the bottom-right end of the list. Previously the bottom of column 1 was a mid-range value, which made “the lowest is not at the end” a real visual gotcha.
• Lines that don’t have a value for the current sort axis (GE / Expr / drug / hotspot count / etc.) now show a dim “—” in the value column instead of being blank, so the cluster of no-data lines at the end of the list is visually distinct from data-having lines that just happen to have small values.
• Drug-response wiki cohort counts now use the GE-cohort intersection (same basis as the sort-drug dropdown), so the two parts of the app show matching numbers. Previously the wiki counted all PRISM keys (e.g. Romidepsin 370 V / 78 P of 461) while the dropdown counted GE-cohort overlap (304 V / 71 P of 386); both numbers were technically correct but the discrepancy was confusing.

v.81.55 (2026-05-12)

• Drug-response dropdown now refreshes its per-compound very-/partly-sensitive counts when the visible cell-line subset changes. Previously, toggling a collection (e.g. excluding TP53 functional loss) updated the list of cell lines but the open dropdown kept showing counts from the previous subset — so the numbers next to each drug didn’t add up to what was actually shown. The dropdown was already scoring against _clbVisibleCellLines; it just wasn’t re-rendered on filter change. Now renderCellLineList() triggers a re-render of the open drug dropdown.

v.81.54 (2026-05-12)

• Sort input now clears when the sort basis changes. Previously a gene name left in the box from “Sort: Expression” would carry over to “Sort: Drug response” (and vice versa), where it would either match nothing or be mistaken for the other kind of token. The suggestion dropdown is also closed on the switch.

v.81.53 (2026-05-12)

• Fix — audit pass on the GE-vs-expression index mismatch from v.81.52. Same root cause was reaching three more places: (1) showGeneEffectByTissue — the main per-tissue expression boxplot was reading expressionData[gi * nCellLines_GE + i] and pairing the result with metadata.cellLines[i]; with GE = 1186 and Expr = 1699 lines in different orders, that often straddled into a neighbouring gene’s row. (2) CSV export — the Expr_GENE columns were resolved through the GE index map. (3) _computeCollectionMemberships — the NE, EMT, and expression-based TNBC / HR+ / HER2+ breast subtype calls were using the GE stride against expressionData, so the cohort scoring was reading the wrong lines’ values. All three now resolve each cell line through expressionMetadata.cellLines with nExprCL stride. Also fixed: mem.tnbc = new Set() in the breast block was wiping the v.81.46 Lehmann-annotated belt-and-braces adds — now preserved.

v.81.52 (2026-05-12)

• Fix — inline Expr value next to a cell line in the list was reading from the wrong slot when sorting by expression. The GE matrix and the expression matrix have different cell-line orderings (and may have different cohort sizes), but renderCellLineList() was using the GE index map and GE nCellLines for the expressionData[gi * nCL + ci] lookup regardless of mode. Result: the value displayed for a given line was actually some other line’s expression. The PD-L1-high collection itself was correct (it already used the expression matrix indices), so the discrepancy showed up as SNU423 — correctly flagged as PD-L1-high — appearing with the inline value of a low-CD274 line. Now uses expressionMetadata.cellLines / nCellLines when sorting by expression.

v.81.51 (2026-05-12)

• Fix — gene autocomplete suggestions flashed and disappeared in GE / Expression sort mode before the user could click them. updateSortControls() (debounced 200 ms after each keystroke) was hiding the suggestion dropdown for any non-drug mode, but the dropdown is now also used for the gene autocomplete from v.81.47. Now keeps the dropdown visible for any sort mode that uses it (drug, GE, expression).

v.81.50 (2026-05-12)

• Fix — pressing Reset and then picking “Sort: Drug response” (or GE / Expression) didn’t show the input box. Cause: the reset handler still set display:none on the inner clbSortGene input from before v.81.35, but the input is now inside a clbSortGeneWrap wrapper. When the user later switched modes, the wrapper became visible but the input was still stuck on display:none and stayed invisible. Fix: reset now hides the wrapper, not the input; clears any stale inline display on the input itself; and dismisses the drug-picker dropdown for good measure.

v.81.49 (2026-05-12)

• Collections — the “?” inclusion-criteria modal now matches the live panel. (1) Same order and category-note explainers as the panel (CATEGORY_ORDER and CATEGORY_NOTES factored in), so both surfaces read as one document. Previously the modal followed catalog insertion order which was arbitrary. (2) Section headers re-styled to the same green-accent banner (14 px, 2 px green underline) used by the panel and the Wiki, so visual hierarchy is consistent everywhere. (3) Inclusion-source badge on every entry — a small uppercase pill (MUTATION, EXPRESSION, CN AMP, FUSION, MSISENSOR2 SCORE, LEHMANN PUBLICATION, INTEGRATED LOF, etc.) tells the user at a glance which DepMap data layer or publication panel each collection was built from. Intro paragraph explains the badge. (4) TNBC dual-inclusion rule documented. The Triple-negative breast (approximate) description now explicitly states both inclusion paths: (a) expression-based receptor-status surrogate (ESR1 / PGR below breast median + ERBB2 not in top 20 %), OR (b) the line carries a Lehmann TNBC subtype annotation. The (a)+(b) union guarantees Lehmann ⊆ TNBC by construction — the earlier discrepancy where the sub-types appeared to contain more lines than their parent set is resolved and explained.

v.81.48 (2026-05-12)

• Collections panel — category headlines made larger and clearer. Bumped from 10 px to 13 px, weight 600 → 700, and the bottom rule changed from a faint 1 px grey hairline to a 2 px green accent so the major sections of the collection panel separate visually at a glance. More top margin (16 px) and a touch of padding give each section room to breathe. Italic category-note line under each header bumped from 10 px to 11 px to match.

v.81.47 (2026-05-12)

• Two fixes prompted by user inspection of the PD-L1 high collection. (1) PD-L1 (CD274) threshold corrected. Previous z > +1.0 cutoff picked up lines with absolute log₂(TPM+1) of 0.4–4.4 because CD274 has a strongly-left-skewed cohort distribution (most cell lines barely express PD-L1 at all). A z above the mean caught lines that are still essentially silent in absolute terms. Now requires BOTH top-quintile absolute expression AND log₂(TPM+1) ≥ 3 (≈ TPM > 7, the “clearly expressed” floor used elsewhere in the app for lineage markers). Description updated to spell out the dual-threshold rationale. (2) Gene-symbol autocomplete on the sort input. When the user picks Sort: GE for gene(s) or Sort: Expression of gene(s) and focuses the gene-name input, a suggestion dropdown now opens (same UX as the drug-response dropdown). Typing “CD27” surfaces CD27, CD274, CD276, etc.; cancer-pathway-panel genes float to the top of the suggestion list (marked with a green dot) so the relevant matches show first. Exact prefix matches outrank middle-substring matches. Click a suggestion to fill the input and trigger the sort; outside-click closes the dropdown.

v.81.46 (2026-05-12)

• Two data-integrity fixes that the user flagged in the collections panel. (1) TNBC consistency. Every cell line with a Lehmann TNBC annotation IS TNBC by the source papers’ definition, but the existing Triple-negative breast collection used expression-based criteria (ESR1 below breast median + PGR below breast median + ERBB2 not in top 20 %) that sometimes missed Lehmann-annotated lines, producing the impossible situation where the Lehmann sub-types appeared to contain more lines than their parent TNBC collection. Now: any line with a Lehmann tnbcType4 / tnbcType6 annotation is also added to the tnbc set, so Lehmann ⊆ TNBC by construction. Runs unconditionally (not gated on expression load). (2) MMR-deficient vs MSI-high. The card’s “MSI-high” badge uses DepMap’s MSIsensor2 readout (functional MSI), while the existing MMR-deficient (likely MSI-high) collection used “any damaging mutation in MLH1 / MSH2 / MSH6 / PMS2 / EPCAM” (genetic signal). These overlap but aren’t identical — sporadic MMR-deficiency by MLH1 promoter hypermethylation is invisible to the mutation matrix, and heterozygous mutations don’t always abolish MMR. So the existing collection includes MSS lines and misses methylation-driven MSI-high lines. Fix: renamed the old collection to “MMR-mutated (Lynch-panel damaging mutation)” to make the signal honest, and added a separate “MSI-high (functional; MSIsensor2 ≥ 20)” collection that matches the card badge. Descriptions on both spell out the overlap-but-not-identical relationship.

v.81.45 (2026-05-12)

• Collections — substantive reorganisation and expansion to match how cancer doctors and researchers actually talk. (1) Renamed and expanded “Oncogene hotspot mutations (mutation only)” → “Driver oncogene mutations”. The old name was opaque and the category had only 2 entries. Now contains KRAS-mut, BRAF-mut, EGFR-mut, PIK3CA-mut (the canonical four every oncologist asks about) plus the pre-existing CTNNB1 and IDH entries. Category note clarifies it’s mutation-only and points at the stricter “Oncogene addiction” for the CRISPR-confirmed subset. (2) Consolidated four singleton DNA-repair categories + Mutation burden into one “DNA repair / damage response” category containing MMR-deficient, HR-deficient/BRCAness, POLE/POLD1 ultramutated, ATM-deficient, and Hypermutated. Previously these were five separate categories with one entry each, which made the panel feel over-segmented. (3) Three new focal-amp collections: MET amp (NSCLC actionable; capmatinib / tepotinib), FGFR1 / FGFR2 / FGFR3 amp (urothelial, gastric, breast, lung-SCC; erdafitinib / pemigatinib / futibatinib), CCNE1 amp alone (pulled out from the G1/S bundle — CCNE1 is a CDK4/6-inhibitor resistance background and a CDK2 / WEE1 / PKMYT1-responder background). (4) Four new disease-defining fusions: SS18-SSX1/2+ (synovial sarcoma), PAX3-FOXO1+ (alveolar RMS), TMPRSS2-ERG+ (prostate), NPM1-ALK+ (ALCL). (5) New immunology collection: PD-L1 (CD274) high expression (z > +1 vs cohort) — canonical clinical checkpoint-immunotherapy biomarker. Description spells out the IHC-vs-mRNA caveat and the cell-intrinsic IFN-pathway cross-link.

v.81.44 (2026-05-12)

• Drug list — clinical indication is now explicitly labelled wherever it appears. Previously the cancer-type list (CRC, SCLC, pancreatic, …) trailed the per-compound row as bare italic text after the σ block, with no label, so it read as a random out-of-context label instead of “the diseases this drug is used for”. Now every place a compound is rendered carries an explicit “Used in:” prefix on the indication string: (1) Standout-sensitive / -resistant lists in the Wiki Drug-response section. (2) Context-aware cross-checks block (the same section). (3) The drug-response autocomplete dropdown in the cell-line browser sort — the dropdown rows now split target/MoA on one subtitle line and the clinical indication on its own labelled line, instead of joining all three with middle dots. Drug-response intro paragraph also names every column explicitly so the user knows what to expect before reading any single compound.

v.81.43 (2026-05-12)

• Collections panel — multi-column layout. Panel max-width bumped from 420 px to min(1100px, calc(100vw - 80px)) and max-height to min(640px, calc(100vh - 160px)), so on a wide monitor it spreads out instead of being a tall narrow scroll. Within each category, entries are laid out in a responsive CSS grid (repeat(auto-fill, minmax(300px, 1fr))) that auto-flows into 2 or 3 columns when there’s room, and collapses to a single column on narrow viewports. Category headers and their italic explainer lines stay full-width so the section structure is still clear. Long collection labels truncate with ellipsis rather than wrapping or pushing the n=count off-screen.

v.81.42 (2026-05-12)

• Drug-response sort dropdown — compounds are now ranked by potency in the currently-visible cell-line subset, not alphabetically. Two upgrades: (1) Per-compound very-sensitive / partly-sensitive counts are computed against the visible subset (i.e. respecting all active filters — tissue / collection / hotspot / fusion / CN / oncoprint), not the full PRISM cohort. So when the user has filtered down to e.g. paediatric KRAS-mutant pancreatic lines, the counts and ranking reflect that cohort. (2) Sort order: descending very-sensitive count, then descending partly-sensitive count, then alphabetic by name. The most potent drugs for the current selection bubble to the top, instead of being hidden alphabetically among 89 compounds. Each row shows the in-subset summary inline: 17 v · 28 p / 84 means “17 of 84 lines in this selection are very-sensitive, 28 partly-sensitive.” Compounds with zero potency in the subset are dimmed but still clickable. Header strip restates the bucket thresholds and the selection size.

v.81.41 (2026-05-12)

• New Patient age at diagnosis collection category with five clinical buckets, computed from the numeric age field in the DepMap Model table (not the coarse ageCategory which is just Adult / Pediatric / Unknown / Fetus). Infant (≤ 1 yr) — congenital and very-early-onset cancers (infant ALL, neuroblastoma stage MS). Paediatric (≤ 14 yr) — standard paediatric-oncology definition; enriched for driver fusions, chromatin / polycomb mutations, embryonal histologies. AYA (15–39 yr) — NCI Adolescent & Young Adult bucket; Ewing, osteosarcoma, testicular GCT, T-ALL, Hodgkin lymphoma. Adult (40–64 yr) — the canonical adult-onset solid-tumour bulk (CRC, breast, lung, pancreatic, prostate). Elderly (≥ 65 yr) — MDS / secondary AML / late prostate / GBM-in-elderly; often distinct mutational backgrounds. Sits in a new “Patient demographics” slot between Disease subtypes and Immunology in the collection ordering. Combine with other oncoprint-style filters for queries like “AYA AND EWSR1-FLI1+”, “Paediatric AND TP53-loss”, or “Elderly AND POLE-deficient”.

v.81.40 (2026-05-11)

• CRISPR dependencies section — explicit fallback message for cell lines with no usable CRISPR data (e.g. 538MGBA). Previously the intro paragraph rendered and then nothing followed (because the z-scored dependency rows produced empty strings), so the user saw a silent void between intro and Source line. Now a red callout reads: “No usable CRISPR gene-effect data for this cell line. The cell line is indexed in the cohort but its CRISPR row contains no measurable values (NaN-only across the matrix), so the z-scored dependency view cannot be built. This happens for a handful of lines per DepMap release — usually those screened too recently to be in this quarter’s CRISPRGeneEffect file, or dropped during QC. Other Wiki sections still show available data for this line.”

v.81.39 (2026-05-11)

• Drug-response section — each compound now shows cohort-wide sensitivity counts. Below the per-line viability / z-score row, a small grey line spells out: “PRISM cohort: N very-sensitive (AUC < 0.3) · M partly-sensitive (AUC 0.3–0.6) of K tested.” Lets the user tell apart a niche-hit compound that kills only a handful of lines from a broadly active drug that kills hundreds. The cohort counts also appear inline in the context-aware cross-checks section in a compact form. Computed once per compound at first Wiki open and cached on the drug-response object for the session.

v.81.38 (2026-05-11)

• Executive-summary drug recommendations are now variant-aware. The previous “KRAS-G12C inhibitors (if G12C)” hedge is gone — we already have the specific codon in the inferred-hotspot data, so the summary now names the drug that actually applies. PANC0504 (KRAS G12D, typical pancreatic) will now show “KRAS-G12D-targeting strategies (MRTX1133 investigational; MEK-inhibitor combinations for p.G12D)” rather than a G12C hedge. Variant-aware mappings: KRAS G12C → sotorasib/adagrasib; KRAS G12D → MRTX1133/MEK combos; KRAS G12R → MEK combos only (no allele-specific inhibitor); other KRAS codons → named explicitly. BRAF V600 → vemurafenib/dabrafenib + trametinib; non-V600 BRAF → flagged as vemurafenib-resistant. EGFR T790M → osimertinib (T790M-specific); L858R / exon-19-del → osimertinib first-line + erlotinib/gefitinib alternatives. IDH1 R132 → ivosidenib / vorasidenib (glioma); IDH2 → enasidenib. Both the “Useful as a model for” and the “Strong drug-target candidates” sentences use the same logic.

v.81.37 (2026-05-11)

• Collections panel categories reordered into a logical research-workflow flow rather than insertion order. New order: (1) Cancer drivers — Disease-defining fusions › Tumor-suppressor functional loss › Focal amplifications › Oncogene hotspot mutations (mutation only) › Oncogene addiction (mutation × CRISPR dependency). Drivers come first because they’re the most common starting point. (2) Disease subtypes — Breast — receptor subtype followed immediately by TNBC molecular subtypes (Lehmann), so the breast-cancer subtype groupings sit together. (3) Phenotype / immune state — Immunology. (4) DNA repair / mutation burden — mismatch repair › homologous recombination › polymerase proofreading › ATM/ATR axis › Mutation burden. (5) Genome state — ploidy / instability (WGD, aneuploidy, LoH). (6) Chromatin / oxidative / pathway — SWI/SNF › NRF2-mutation › Pathway-activity expression signatures › Expression signature (NE / EMT). Category notes (italic grey one-liners under each header) extended to Disease-defining fusions, Oncogene hotspot mutations, and TNBC molecular subtypes so jargon-heavy category names are self-explanatory.

v.81.36 (2026-05-11)

• Wiki — POLE / POLD1 ultramutators (e.g. JHUEM7) now get their own flag and a passenger-noise caveat. (1) New flag pill when POLE or POLD1 is damaged AND damaging-mutation count > 200: “POLE / POLD1 proofreading-deficient (ultramutator)”. Tooltip explains: replication errors not corrected → > 100 mut/Mb point-mutation burden, but the genome stays structurally calm (low CIN / aneuploidy / LoH, no WGD, microsatellite stable). Distinct from MMR-deficiency, which is the other major hypermutator route. (2) Passenger-noise caveat on the existing “Very high mutation burden” flag (> 1000 damaging) and “High mutation burden” (> 500): at hypermutator burdens, single damaging mutations in any driver gene panel are statistically likely to be passenger events, so flags like HRD / TP53 should be cross-checked against the integrated functional-loss view rather than taken at face value. (3) Genome signatures intro now states explicitly that the displayed metrics are structural / chromosomal (WGD, aneuploidy, CIN, LoH, MSI) and not point-mutation burden — so a structurally-calm genome at very high TMB (the JHUEM7 picture) reads as a real biological state, not an annotation error.

v.81.35 (2026-05-11)

• Drug-response sort — proper dropdown panel. The previous native <datalist> approach only revealed compound suggestions when the user typed or clicked the input’s tiny caret, which the user couldn’t see was even there. Replaced with a custom dropdown panel that opens on focus (just click into the input), shows all ~89 PRISM compounds with target / MoA / indication subtitles, and live-filters as the user types. Mirrors the existing CN / fusion-filter dropdown pattern. Clicking an option fills the input and triggers the sort. Outside-click closes the panel.

v.81.34 (2026-05-11)

• Drug-response sort — PRISM compound autocomplete. The input field for the “Sort: Drug response (PRISM AUC)” mode is now wired to an HTML <datalist> populated from drugResponse.compounds — the user gets a typeahead dropdown of every PRISM compound name (sorted A–Z) with the target / MoA / indication shown as a secondary subtitle. Placeholder updated to “type a PRISM compound (autocomplete)” so the affordance is obvious.
• Thirteen new collections targeting gaps in the catalog. (1) TNBC molecular subtypes (Lehmann) — four entries (BL1, BL2, M, LAR) for the ~22 TNBC overlap cell lines annotated in lehmannTnbc. Each entry summarises the panel-defined phenotype (DDR for BL1, growth-factor signalling for BL2, EMT/stem for M, AR signalling for LAR) and predicted therapy. (2) Disease-defining fusions — four entries: BCR-ABL1+ (CML/Ph+ ALL), EWSR1-FLI1+ (Ewing sarcoma), EML4-ALK+ (ALK-rearranged NSCLC), PML-RARA+ (APL). Members are cell lines where the curated clinical-fusion panel calls that specific fusion. (3) Additional DNA-repair contexts. POLE / POLD1 proofreading-deficient (ultramutated; distinct from MMR — usually > 100 mut/Mb and ICB-responsive), ATM-deficient (HRD-adjacent, separate from the BRCA-centric set). (4) Oncogene hotspots not covered by addiction. CTNNB1 hotspot (β-catenin / Wnt activator; mutually exclusive with APC loss), IDH1 / IDH2 hotspot (oncometabolite 2-HG producers; mutant-IDH-inhibitor cohort). (5) Expression-based DDR predictor. SLFN11-high (z > +1) — the single strongest expression-based predictor of platinum / topo-I / PARPi / alkylator response.

v.81.33 (2026-05-11)

• Executive summary — new “Strong drug-target candidates” sentence. Surfaces specific drug names where the evidence is robust: (a) any oncogene-addiction collection fires for this cell line (mutation + functional CRISPR dependency, the strongest combined signal) — KRAS-G12C inhibitors / BRAF+MEK / EGFR TKIs / ABL TKIs / PI3K-α / CDK4/6 inhibitors named directly; (b) PRISM-screen compounds with z < −1.5σ sensitivity — top 2 named with target and z. Capped at four entries. The earlier “Useful as a model for” sentence frames biology; this one names concrete therapies likely to work.
• Collections panel — categories with jargon-heavy names get inline explainers. Under each affected category header, a small italic grey line spells out the gist. Examples: Oncogene addiction (mutation × CRISPR dependency) → “Cancer cells that have become functionally dependent on a driver gene — knocking it out (or blocking it with a drug) kills the cell. Stricter than the mutation-only collections above.” Pathway-activity expression signatures → “Each set marks lines where the pathway is transcriptionally active (mean z > +0.75 across a curated gene panel). Reflects what the cell is doing, not just what is mutated.” Tumor-suppressor functional loss and Focal amplifications also get one-line gists.

v.81.32 (2026-05-11)

• Fifteen new collections lifting Wiki-only views into cohort-level filters. (a) Oncogene addiction — new category “Oncogene addiction (mutation × CRISPR dependency)”. Six entries combine the driver-mutation call with the actual CRISPR knock-out read-out (GE < −0.5): KRAS-addicted, BRAF-addicted, EGFR-dependent, CDK4/6-dependent (RB1-intact + CDK4/6 essential), BCR-ABL-addicted, PI3K-active and dependent. Stricter than the existing mutation-only collections — picks lines where losing the driver actually kills the cell, not just lines that carry the mutation. (b) Pathway-activity expression signatures — new category. Eight entries reuse the same curated panels the Wiki uses (MYC targets, E2F / S-phase, G2/M, IFN response, TGF-β, hypoxia / HIF, NRF2 signature, stem / lineage plasticity). Membership = mean z > +0.75 across the panel genes. Computed lazily after expression loads, mirroring the existing NE / EMT / breast-subtype collections. (c) High LoH (≥ 0.3) added to the existing Genome — ploidy / instability category for HRD-adjacent stratification.

v.81.31 (2026-05-11)

• STR authentication block now links to the source data. New “Validate against the source” sub-block above the reference STR profile table with a direct Cellosaurus link for this cell line’s RRID (provenance, parent / derivative relationships, contamination flags, misidentification history are all there), plus ATCC search and DSMZ search links for buying commercial authentication services. The reference STR profile is also explicitly attributed to Cellosaurus (SIB / ExPASy) in a footnote line so the user knows where to validate.
• Executive summary gets a “Useful as a model for” sentence. One additional synthesized sentence at the end of the summary banner mapping the cell line’s driver pattern to the biology / therapy questions it’s well-suited to study. Examples: BRAF V600E → “BRAF/MEK-inhibitor response (vemurafenib / dabrafenib / trametinib)”; BCR-ABL1 → “ABL TKI (imatinib / dasatinib / nilotinib) response”; EML4-ALK → “ALK-inhibitor (crizotinib / alectinib / lorlatinib) response”; CDKN2A loss → “CDK4/6-inhibitor (palbociclib) response”; BRCA1/2 loss → “PARP-inhibitor response”; MSI-high → “checkpoint-immunotherapy response background”; ERBB2 amp → “HER2-targeted-therapy response”; MDM2 amp with WT TP53 → “MDM2-inhibitor (nutlin) response”; Class-I lost / reduced → “cancer immune-escape via MHC-I loss”. Capped at three angles to keep the summary tight.

v.81.30 (2026-05-11)

• Drug response — clearer σ / z-score explanation. The intro paragraph now defines σ explicitly (standard deviation of the AUC for that drug across all PRISM cell lines), explains what “−1.4σ below average” concretely means (this line kills 1.4 SDs harder than typical for this drug), and gives interpretation thresholds: |z| > 1σ = noteworthy, |z| > 2σ = strong outlier. AUC scale also restated at the top so the user doesn’t have to remember 0 = killed / 1 = no effect.
• New “Sort: Drug response” in the cell line browser. Pick the option, type a compound name (or substring; multiple compounds averaged) in the input field — e.g. PALBOCICLIB, IRINOTECAN, PALBO, ABEMA. List sorts ascending by AUC by default (most-sensitive lines at top) with the AUC value shown inline on each row, colour-coded by sensitivity (green < 0.6, amber 0.6–0.85, grey > 0.85). Lines not in the PRISM screen sink to the bottom regardless of direction.
• Wiki palette sweep — calmer, fewer hues. The Wiki was previously using purple (druggable targets), teal (FACS markers), bright orange (pathway-activity signatures), sky blue (lineage markers), and several blue shades (typical-pattern, hallmark-checklist) on top of the existing green/red/grey base. All consolidated into four accent colours: green for active / dependencies / sensitivities / druggable / FACS, red for loss / resistance / immune escape, indigo for typical patterns / lineage / cross-checks, grey for neutral / other / off. Section headers keep their existing green strip styling; pathway-status synthesis colours (per-panel, dynamic) keep their amber/green/red since those convey the call.

v.81.29 (2026-05-11)

• Wiki Expression profile — new “Potential FACS markers” block. Intersects expression with a curated ~100-gene panel of well-known cell-surface antigens: CD molecules (CD3/4/8/14/19/20/22/24/27/33/44/45/56/68/127/133, …), receptor tyrosine kinases (EGFR, HER2-4, FGFR1-4, MET, AXL, KIT, IGF1R, NTRK, ALK, …), immune checkpoints (PD-L1, PD-L2, PD-1, CTLA4, LAG3, TIM-3, TIGIT, B7-H3, VISTA), ADC / bispecific / clinical targets (TROP2, FOLR1, MSLN, NECTIN4, CEACAM5/6, MUC1/16, DLL3, LIV1, STEAP1/2, PSMA, NaPi2b, claudins), adhesion molecules (EPCAM, CDH1/2, ICAM1, integrins), death receptors (FAS, DR4, DR5), Notch family, and tumour-associated antigens (CA9, GPC3, …). Threshold TPM > 4 (a bit stricter than druggable targets, since FACS needs antibody-detectable protein); top 12 shown, sorted by uniqueness vs cohort (z-score). Each entry shows raw TPM and z, so the user can pick markers that are both bright AND distinctive to this line as wet-lab handles for FACS, antibody assays, or candidate targeting.

v.81.28 (2026-05-11)

• Wiki executive summary — new synthesized opening paragraph at the very top of every cell-line Wiki, so the headline biology is visible before scrolling. Driver hierarchy: clinical fusion › canonical oncogene hotspot › canonical TSG functional loss. Genomic flags (WGD, MSI, Class-I status, hypermutation) follow as a final sentence. Examples: “JR is a Rhabdomyosarcoma cell line. Driven by PAX3-FOXO1 driver fusion plus FGFR3 p.K650, with functional loss of TP53, RB1. Focal amplification of MYCN, MDM2. Genome / phenotype: whole-genome doubled; Class-I antigen presentation reduced.” Falls back to a gentle “no canonical driver detected” note when nothing in the integrated panels triggers.
• HLA / polymorphic-locus hotspots demoted from the default view. Hotspot mutation calls in HLA-A/B/C / MIC / KIR genes are usually allelic divergence from the GRCh38 reference, not somatic events — surfacing them as “hotspot mutations” falsely implied they were cancer drivers. They’re now hidden behind the existing “show all” toggle in the cell-line detail card’s Hotspot Mutations list, and filtered out of the Wiki’s “Top hotspot hits” line. Still recoverable via the toggle for users who want to inspect them; the ⚠ polymorphic-locus marker still flags them when shown.

v.81.27 (2026-05-11)

• Wiki Expression profile section reworked from raw top-expression to uniquely-high vs cohort + pathway-activity signatures. The previous “Most-expressed genes” row was always dominated by mitochondrial and ribosomal genes (MT-ATP8, GAPDH, EEF1A1, MT-CO2 …) that are high in every cell line and tell you nothing about this specific one. (1) Top uniquely high expression — now z-ranked against the full cohort. Each gene shows both its raw TPM and its z-score, e.g. FGFR3 (TPM 6.4, z +2.8). Pan-housekeeping genes drop out naturally because they have low SD across the cohort. (2) Pathway-activity signatures — new block. Nine curated gene panels (MYC targets, E2F / S-phase replication, G2/M progression, IFN response, EMT, TGF-β response, hypoxia / HIF target, NRF2 / oxidative-stress, stem / lineage plasticity). For each panel we compute the mean z-score over the panel genes in this cell line; panels with mean z > +0.75 are surfaced as “turned ON” (with the pathway note in a tooltip) and mean z < −0.75 as “turned OFF”. (3) Lineage markers and druggable targets kept and now annotated with z-scores so you can see which are uniquely high here vs typical for the cohort, not just absolutely high.

v.81.26 (2026-05-11)

• Wiki — bug fix: hovering a gene name in the Wiki body showed the help cursor but no actual MyGene.info tooltip popped up. Cause: openCellLineWiki rendered the body HTML but never called attachGeneTooltips on it, so the mouseenter handlers that fetch and display the gene-info tooltip were never bound. One-line fix: call this.attachGeneTooltips(body) after the Wiki body is composed and the modal is shown.

v.81.25 (2026-05-11)

• Wiki Genome signatures rows — every numeric value now carries a plain-English descriptor so the user doesn’t have to know what a “high” or “low” value looks like. Ploidy 3.43 now reads as “near-tetraploid”, Aneuploidy 23 / 39 as “medium”, CIN 0.14 as “low”, LoH 0.46 as “high”, MSI 2.9 as “microsatellite stable — well below the MSI-high cutoff of 20”. Each row also restates the relevant threshold inline (e.g. “high tier ≥ 25 of 39” for Aneuploidy, “high tier ≥ 0.3, often seen in WGD-positive or HRD lines” for LoH) so the cutoff is visible without scrolling to the cohort histogram below. Thresholds are the same ones already used elsewhere in the app for collections and the card view.

v.81.24 (2026-05-11)

• Wiki — section headers (Cancer classification, Patient & sample origin, Genome signatures, Driver mutations, etc.) made much larger and visually distinct so the major sections separate cleanly. Header bumped 13 → 17 px, bold, set on a light-green tinted strip with a green left-border accent and a small right-curve, mirroring the “section” treatment in the live app. Outer section bottom margin nudged 18 → 24 px for extra breathing room. The downloadable standalone-HTML version of the Wiki picks up the same header styling (h3 rule added to its embedded CSS) so exported reports also read as cleanly-sectioned documents.

v.81.23 (2026-05-11)

• Wiki — CRISPR dependencies section now ranks by z-score vs the full cohort instead of raw gene-effect, so the top entries are uniquely essential to this cell line (downstream of its driver mutations, lineage-specific factors, etc.) instead of the pan-essentials (RAN, PCNA, RPL17 …) that every cell line shares. Pan-essentials are additionally filtered out against the DepMap common-essentials list as a safety net. Each gene is now annotated with both its raw GE and its z-score, e.g. FOXA1 (GE −1.20, z −3.2). The matching “growth-promoting on knockout” row is also z-ranked. The 💊 pill meaning is now spelled out inline in the section header: approved or clinical-stage drug targets this gene. Druggable / pathway dependency interpretation panels also switched to the uniquely-essential set (z < −1.5). Computation: per-gene mean and SD across the full cohort, cached on first Wiki open. Section caption rewritten to make this framing explicit.

v.81.22 (2026-05-11)

• Wiki — Driver mutations section condensed and corrected. (1) Self-contradiction fixed. For an RB1-lost / CDK4/6-dependent cell line (e.g. JMSU1), the pathway bullet previously said “CDK4/6 inhibitors likely active” while the synthesis underneath said “CDK4/6 inhibitors typically ineffective”. Bullet now appends the caveat “but RB1 is lost — RB1 sits downstream of CDK4/6, so clinical CDK4/6 inhibitors are typically not effective even when the cell is CDK4/6-dependent by gene-effect”. (2) “Essential” language calibrated. Gene-effect −0.33 is no longer described as “essential” — CRISPR thresholds across all pathway-status panels (RB/CDK4-6, p53, RAS/MAPK, PI3K/AKT, BCR-ABL, EGFR) are now: < −0.5 = “strongly reduces growth / essential”, −0.5 to −0.2 = “moderately reduces growth / partial dependency”, ≥ −0.2 = neutral. The dependency caption now notes the scale explicitly: 0 = no effect, −0.5 = selectively essential threshold, −1 ≈ typical strongly-essential gene. (3) Condensed. Dropped the “TP53 is mutated” and “G1/S cell-cycle brake lost” flag cards (covered better by the p53 and Cell cycle pathway-status panels); remaining flags (MMR / HRD / very-high TMB) are now compact inline pills with the narrative on hover instead of bordered title-plus-body cards. Three count rows (damaging, hotspot-mutated, top hits) combined into one line. Each pathway-scan entry collapsed from four rows (title, note, gene list, divider) to a single line. Pathway status now appears before Pathway scan so the interpretive view comes first.

v.81.21 (2026-05-11)

• Wiki — remove WGD / MSI-high from the “Other alterations in this cell line” block under Expected drivers for this subtype. They are genome-wide phenotypes, not gene-specific driver alterations, and both are already prominent in the Genome signatures section directly above (full WGD row, MSI score row, plus the cohort histograms). Surfacing them twice in adjacent sections was duplicative and confusing. The drivers block now stays focused on what it’s for: gene-specific alterations — fusions, functional-loss calls, and hotspot mutations.

v.81.20 (2026-05-11)

• Wiki “Expected drivers for this subtype” section reorganized into three clearly-separated blocks so the expected-vs-found split is readable at a glance. (1) Typical alteration pattern — the curated KB description for the Oncotree subtype (blue panel). (2) Hallmark-gene checklist — one row per expected gene with a ✓ when present in this cell line (plus a short per-row hit summary: hotspot p.V600E, damaging, functional loss, fusion, etc.) or ✗ when absent. Header shows “N of M present”. (3) Other alterations in this cell line — driver fusions, functional-loss calls, and hotspot mutations on cancer-panel genes that are not in the hallmark list, plus genome-level flags (MSI-high, WGD). This makes the “hallmark expected but missing” vs “present but unexpected” structure visible without scanning a wall of text.

v.81.19 (2026-05-11)

• Wiki — new “Where this line sits in the cohort” sub-block under the Genome signatures section. Four small Plotly panels rendered FACS-style: Ploidy, Aneuploidy, CIN as continuous histograms, WGD as a binary two-bar count. This cell line’s value is marked with a dashed red line (continuous) or a red bar (WGD), with its numeric value labelled above the line. Per-panel PNG / SVG export via Plotly’s built-in modebar (hover top-right of each plot). Cohort CSV export button next to the heading dumps ModelID, Name, Ploidy, WGD, Aneuploidy, CIN, plus an IsInspected flag for the currently-open cell line. Skipped silently for lines with no global-signatures entry.

v.81.18 (2026-05-11)

• Cell-line list sex symbol — when the annotated sex and the expression-inferred sex directly conflict (e.g. TE6: annotation Male, expression Female), the line now shows “?” with a conflict tooltip instead of the annotated symbol. Neither signal is trustworthy on its own when they disagree, and silently picking the annotation hides the conflict from the user. The detail card still shows both Sex (annotation) and Sex (expression) on separate rows with the existing amber-disagreement style, so the raw signals stay available for users who want to dig in. The sex filter dropdown is unchanged — selecting “Male (annotation)” still includes TE6 since that is its annotation.

v.81.17 (2026-05-11)

• New collections category Focal amplifications. Five entries covering the gain-of-function side of focal CN events (the loss side is already represented by the per-TSG functional-loss collections, since DepMap’s integrated LoF call already picks up deep deletions). MYC family amp (MYC / MYCN / MYCL — pathway active across lineages), ERBB2 amp (HER2-driven; lineage-agnostic, distinct from the breast-only expression-based HER2+ collection), MDM2 amp (p53 functionally suppressed without a TP53 mutation — nice contrast to the TP53-loss set), G1/S amp (CDK4 / CDK6 / CCND1 / CCNE1 — G1/S deregulation; contrast to RB1-loss / CDKN2A-loss), and Any focal amplification as a generic catch-all. Membership pulls from the curated clinical CN panel (clinicalCn.byCellLine.amplifications).

v.81.16 (2026-05-11)

• New collections category Tumor-suppressor functional loss. One filter entry per gene in the standard 9-gene TSG panel — TP53, PTEN, RB1, CDKN2A, APC, NF1, SMAD4, STK11/LKB1, VHL — plus aggregates “Any TSG functional loss (≥ 1 of the 9)” and “Multiple TSG losses (≥ 2)”. Each gene-specific entry includes the pathway implication (e.g. “TP53 functional loss → p53 checkpoint lost”, “APC loss → WNT pathway active”). Membership pulls from DepMap’s integrated functional-loss call (low CN, damaging mutation at high allele frequency, or near-zero expression) so deep deletions and silenced loci that the damaging-mutation matrix misses are still captured. Combine with the existing include / exclude oncoprint-style controls to ask things like “TP53-loss AND not RB1-loss” in one shot.

v.81.15 (2026-05-11)

• Collections panel — fixed strange-looking text in tooltips. HTML entities (—, −, ≥, ’, –) used in the catalog description strings were rendering literally in some tooltip contexts. Swapped them out for the Unicode characters directly (— − ≥ – etc.) so they show as glyphs everywhere, regardless of whether the text ends up inside a title="" attribute, a textContent assignment, or innerHTML body.
• Immunogenicity prediction added as two new collections under Immunology, replacing the hot/cold framing. Likely immunogenic = ≥ 200 damaging coding mutations and class-I antigen presentation appears intact (no B2M damaging, HLA-A/B/C expression not strongly reduced, HLA CN not lost vs B2M). Likely non-immunogenic = either class-I reduced/lost (immune escape) or ≤ 50 damaging mutations with otherwise intact MHC-I (too few neoantigens). Lines that don't fit either profile are unlabelled. The catalog description spells out the inputs and the cell-intrinsic-proxy caveat (real-tumour immunogenicity also depends on the TME, which isn't observable in cell lines).

v.81.14 (2026-05-11)

• Collections filter — oncoprint-style include/exclude. The single-collection dropdown is replaced with a “Collections [N]” button that opens a multi-select panel. Each collection has two toggle buttons: + = require (cell line must be IN this set), − = exclude (cell line must NOT be in this set). Multiple collections combine with AND, so e.g. you can ask for “MSI-high” AND not “WGD-positive” in one shot. Active states are rendered as removable chips in the active-filter strip; clicking × on a chip clears just that state. Button label and chips stay in sync with the panel, and Reset clears all of them.

v.81.13 (2026-05-11)

• Cell-line browser — iterative cleanup step 3. (1) Tooltips rewritten in plain language with term definitions. The “?” tooltips next to Genome, Functional loss, Focal CN events and Clinically relevant fusions previously read like method-section snippets. They now define every term inline (Ploidy, WGD, Aneuploidy, CIN; the three-signal LoF / focal-CN / fusion criteria) before going into the caveats. (2) Three new collections. “Class-I antigen presentation reduced/lost” (Immunology) groups the immune-escape candidates surfaced by the per-cell-line flag. “WGD-positive” and “High aneuploidy” (Genome — ploidy / instability) let the user pick or contrast genomically unstable lines. (3) Three new sort options: Ploidy, Aneuploidy, CIN. The numeric value appears inline on each row alongside a short unit tag (pl 2.29, aneup 12, CIN 0.14). Lines missing the metric are pushed to the bottom regardless of sort direction. Caption line above the list spells out the unit so the numbers read unambiguously.

v.81.12 (2026-05-11)

• Cell-line detail card — iterative cleanup step 2. Damaging mutations list now defaults to only the cancer-panel-relevant subset (genes that appear in any of the curated cancer pathway panels). The long passenger tail (ACTN1, ADGRG6, BZW2, etc. on most lines) sits behind a “show all N” toggle. When zero cancer-panel hits, the section shows “None in cancer-panel set” with the toggle still available. Hotspot and fusion lists unchanged (they’re already curated). The new flags section renamed “Phenotype flags” → “Immunology” since both MSI-high and Class-I reduced are immunology-relevant signals (checkpoint-inhibitor response prediction and immune evasion, respectively).

v.81.11 (2026-05-11)

• Cell-line detail card — calmer, more readable. Step 1 of an iterative cleanup. The MSI and Class-I antigen-presentation badges no longer sit next to the cell-line ID where they read as the headline feature; they now live in a new “Phenotype flags” section with a one-line plain-English caption next to each chip so the user doesn’t need to hover to know what they mean. Genome row rewritten from a single cryptic line (“Ploidy 2.29 · Aneup. 4 · CIN 0.14”) into four indented rows with descriptors (“Ploidy 2.29 (near-diploid)”, “Aneuploidy 4 / 39 (low)”, “CIN 0.14 (low)”, “WGD yes” when positive). Focal-CN chips no longer use the ↑↓ glyphs (they read as “going up / down” rather than “amplified / deleted”); the gene name now sits next to a small “amp” / “del” / “strong amp” / “deep del” word. Palette muted across the card: amps use dark-blue text on a light-blue tint, deletions and LoF use dark-red text on a light-red tint; fusion tier pills (high/med/low) keep the semantic colour but on neutral grey borders; the Lehmann TNBC chip is now grey instead of saturated purple. No data removed.

v.81.10 (2026-05-11)

• Network legend zoom-sync — bug fix: clicking the vis-network home/reset button (above the “+” in the lower-right nav cluster) reset the view but the legend kept reading “at current zoom 2.44×”. Cause: vis-network only fires zoom on wheel/pinch input, not on the home button or programmatic fit(). Switched the listener to afterDrawing with a scale-change diff (cheap; only re-renders the legend HTML when getScale() actually changed) so the legend tracks every kind of view change — mousewheel, pinch, nav buttons, double-click fit, and stabilization-done autofit.

v.81.9 (2026-05-11)

• Network — edge-thickness legend now scales with the canvas zoom, so the legend lines visually match the edges actually being drawn (avoids the misrepresentation where the legend stayed at base size while zoomed-in edges looked much thicker). A small “(at current zoom N.NN×)” hint appears next to the legend heading when zoom ≠ 1. Hint and scaled widths are screen-only — PNG/SVG exports continue to render the legend at the canonical base-scale mapping with no zoom annotation, so exported figures remain publication-ready. Line widths clamped to [0.5×, 3×] in the legend panel to avoid overflow at extreme zoom.

v.81.8 (2026-05-11)

• Mutation analysis “Compare by Tissue / Additional Hotspot / Fusion” inline table redesigned to make subset counts obvious. New leading N (in subset) column shows how many cells share the row’s tissue / hotspot / fusion (e.g. how many cells carry GNAS mutation); a vertical separator divides it from the right-hand block, which now sits under a colspan header naming the main stratification gene (e.g. “TP53 stratification (within subset)”) so it’s clear that N(WT) / N(Mut) are the main-gene intersection within that subset. Added a “Hide rows with N < x” filter (default 5) so trivial subsets (e.g. 3-cell hotspots) are suppressed; the filter shows how many rows are hidden. Sort key map extended to 7 columns to match the new layout.

v.81.7 (2026-05-11)

• Export-chart dialog layout fix — Width, Height and Lock-aspect were wrapping in three rows because the labels plus inputs didn't fit in the 420 px modal. Modal widened to 500 px and the three controls are now in a single flex-wrap:nowrap row, so Height number sits on the same row as the Height label.

v.81.6 (2026-04-27)

• Curated focal CN events integrated end-to-end. New scripts/clinical_cn_anchor.json + scripts/process_clinical_cn.py extract clinically actionable amplifications and TSG deletions from OmicsCNGeneWGS.csv. Output web_data/clinical_cn.json (~125 KB; 527 cell lines with at least one focal CN event).
  • Amplification panel (~32 oncogenes): ERBB2, MYC, MYCN, MYCL, EGFR, MDM2, MDM4, CDK4, CDK6, CCND1/2/3, CCNE1, FGFR1/2/3, MET, KIT, PDGFRA, KRAS, AR, ESR1, MITF, SOX2, NKX2-1, BCL2, AURKA/B, NTRK1, NUTM1, TERT, IGF1R. Threshold: relative CN ≥ 3.0 (high-level amp), ≥ 5.0 (strong amp).
  • Extended-deletion panel (~14 TSGs): BAP1, SMARCA4, SMARCB1, ARID1A, STK11, BRCA1/2, SETD2, ATM, CDKN2B, KEAP1, FAT1, FBXW7, TET2. Deliberately excludes the 8 TSGs already in DepMap inferred-LoF (TP53, RB1, PTEN, NF1, CDKN2A, VHL, MTAP, APC) to avoid duplication.
  • Cell-line browser detail panel: new "Focal CN events" section with red ▲ chips (amplifications) and blue ▼ chips (deletions). Bold border for strong / deep tier; thin border for suggestive. Hover shows CN value plus context (e.g. "MYC amplification (CN = 8.4, strong). Pan-cancer driver. Burkitt lymphoma, neuroblastoma, breast, colon, etc.").
  • AI .json.gz export: per-cell-line cellLineMetadata.<cl>.cnEvents = { amplifications: [{gene, cn, tier}], deletions: [...] }. Schema bumped to 3.2 with documentation and threshold definitions.
  • New CLB toolbar CN filter — same multi-line dropdown pattern as the fusion filter. Lists every (gene, kind) tuple in the panel with live n= scaled to the active cohort filters. Selecting "▲ MYC amp (n=65)" filters the cell-line list to the 65 MYC-amplified lines; selecting "▼ BAP1 del" similarly. Sorted with amplifications first, deletions after.
  • Top amp counts (sanity check): MYC n=65, CCND1 n=61, KRAS n=27, MYCN n=27, ERBB2 n=25, CDK4 n=24, MDM2 n=23, CCNE1 n=15, EGFR n=15, MET n=11. Top extended-deletion counts: CDKN2B n=295 (chr9p21 co-deletion with CDKN2A; expected high), FAT1 n=50, STK11 n=22, BAP1 n=15, ATM n=14, SMARCB1 n=14.
• Reference Data card on main page updated to list the new "Focal CN events" annotation alongside HLA CN.
• Deferred (next pass): adding CN-state filtering to the main parameter panel (correlation analysis), so users can stratify "MYC-amplified vs not" cohorts. Bigger UI change touching multiple analysis paths; keep as a separate session.

v.81.5 (2026-04-27)

• HLA copy-number ingested + class-I inference activated. Ran scripts/process_hla_cn.py against the downloaded OmicsCNGeneWGS.csv (~400 MB DepMap WGS CN matrix); output web_data/hla_cn.json (190 KB) covers 830 cell lines for HLA-A/B/C, B2M, and antigen-processing-machinery genes. Two corrections during ingestion:
  • The CSV format is DepMap's relative copy number (1.0 = diploid), not log2 ratio or absolute — format detection rewritten to recognise this from the B2M reference distribution.
  • Short-read WGS systematically under-maps HLA-C in the polymorphic region (cohort median ~0.22, not real loss). The class-I CN inference now uses HLA-A/B normalised to B2M per cell line (B2M sits on chr15, clean diploid reference). Ratio < 0.5 = strong loss, < 0.7 = suggestive loss. Catches HLA-specific copy loss while controlling for sample-level mappability and the HLA-C reference bias.
• Lehmann TNBC subtype annotation. The Oncotree subtypes (Breast Invasive Ductal Carcinoma, Triple-Negative Breast Cancer, etc.) don't capture the Lehmann TNBC molecular axis (BL1, BL2, IM, M, MSL, LAR) and our previous user-flagged: "the cell-line browser TNBC labels don't match Lehmann's classifications." Added curated subtype assignments from Lehmann JCI 2011 (PMID 21633166) plus the 2016 reclassification (PMID 27310713; collapsed to four classes — BL1, BL2, M, LAR — with IM and MSL re-attributed to immune / stromal infiltrate). New scripts/lehmann_tnbc_anchor.json + scripts/process_lehmann_tnbc.py; output web_data/lehmann_tnbc.json covers 22 of 29 panel cell lines (the other 7 aren't in DepMap 25Q3). Surfaced as a small purple chip next to the Subtype row in the cell-line browser detail panel (e.g. "Lehmann BL1" or "Lehmann MSL → M" when 2016 reclassified). Hover for full definitions of the six classes. Also embedded per cell line in the AI .json.gz export as cellLineMetadata.<cl>.lehmannTnbc.
• Reference Data card on main page updated to list HLA CN and Lehmann TNBC under the curated-annotations bucket.
• schemaVersion 3.0 → 3.1.

v.81.4 (2026-04-27)

• Class-I antigen-presentation inference per cell line. Three integrated signals — none alone is conclusive, hence the integration:
  1. B2M damaging mutation — β2-microglobulin truncation knocks out the entire class-I antigen-presentation system regardless of HLA status. Strongest single signal.
  2. HLA-A/B/C expression — z-score vs the full cohort. Very low expression of all three implies functional loss (transcriptional silencing OR LOH; we can't distinguish without phased reads).
  3. HLA-A/B/C copy number — from web_data/hla_cn.json (extracted from DepMap's WGS CN matrix). log2-ratio < −0.4 in all three suggests one-copy loss.
  Status tier: likely_lost (B2M damaging hit, OR strong reduction in two independent signals), reduced (one signal), intact (data available, no signals), unknown (data missing). Surfaces as a small red / amber chip next to the cell-line ID in the browser detail panel, with a hover tooltip listing the specific evidence. Also embedded in the AI .json.gz export per cell line as cellLineMetadata.<cl>.class1AntigenPresentation (only emitted when status is non-trivial). Schema bumped to 3.0.
• Important caveat: this is a functional inference, not allele-specific HLA-LOH detection. Real LOH calls would need phased reads (LOHHLA) on raw BAMs, which DepMap doesn't redistribute. The integration here catches the same class of immune-escape-candidate cell lines for biology purposes — tooltip and AI-export field both spell that out.
• New scripts/process_hla_cn.py — takes a downloaded OmicsCNGeneWGS.csv (~400 MB), extracts copy number for HLA-A/B/C, B2M, antigen-processing-machinery genes (TAP1/2, TAPBP, ERAP1/2, NLRC5), and class-II HLA. Output: web_data/hla_cn.json (~50 KB). The CN file isn't bundled (gitignored); download it once from DepMap, run the script, commit the resulting JSON.

v.81.3 (2026-04-27)

• HLA / polymorphic-locus caveat. The HLA region (chr 6p21) is the most polymorphic in the human genome — HLA-A alone has >7,000 known alleles — and gnomAD-based germline filtering leaks here. So "hotspot" calls in HLA / MIC / KIR genes typically reflect allelic divergence from the GRCh38 reference, not somatic events. Real cancer-relevant HLA loss exists (immune escape via class-I LoF + LOH, especially in MSI-high tumours) but should be inferred from expression loss + LOH, not the raw hotspot matrix.

  New _POLYMORPHIC_LOCI set covers the classical class-I HLA (A/B/C), non-classical (E/F/G), class-II (DPA/DPB/DQA/DQB/DRA/DRB1-5/DM/DO), MIC (A/B), and the KIR family. Wherever a hotspot or damaging mutation is rendered, polymorphic-locus genes now get a small ⚠ marker with a hover tooltip explaining the caveat:

  • Cell-line browser detail panel — hotspot / damaging / fusion gene lists. Polymorphic genes also sorted to the end of those lists so they don't crowd the always-visible top-10 (still surfaced when "Show all" is expanded).
  • Mutation analysis results table — gene name now carries a ⚠ when polymorphic.
  • AI .json.gz export — cellLineMetadata.<cl>.mutations.<gene>.caveat: "polymorphic_locus" per-cell-line flag, with the schema description telling the LLM to treat such calls as suspect unless paired with expression loss + LOH.

v.81.2 (2026-04-27)

• AI .json.gz export — precomputed extras for the remaining view types that were lagging behind gene_effect / scatter (Priority 5 of the recent brief). Schema bumped to 2.8.
  • Correlation view: new extras.tissueStratifiedCorrelations — top 20 correlation pairs by |r|, each broken out into per-tissue r values for tissues with n ≥ 10 cell lines. Catches the most common artifact in cross-cohort correlation analysis: a cohort-wide r that's actually lineage segregation rather than biology. If a pair's overall r is 0.5 but the per-tissue r is near zero in every tissue, the LLM can flag it as lineage-driven before reading anything into the relationship.
  • Cluster view: new extras.clusterAnnotations — per-cluster pathway overlap with the wiki cancer pathway KB (pathway must share ≥2 genes with the cluster) plus a CORUM co-member count. Tells the LLM whether a cluster is biologically coherent (e.g. "cluster 3 has 5/8 members in PI3K/AKT/mTOR") or a grab-bag without obvious functional unity.
  • Gate-comparison view already shipped most of the relevant precomputed extras (differentialGeneEffect, differentialExpression, tissueEnrichment, subtissueEnrichment, mutationEnrichment) via _gateCompareResults; left unchanged.

v.81.1 (2026-04-27)

• AI .json.gz export — three additions from a Claude Code brief targeting the scatter view (Priorities 1–2 of the brief). Schema bumped to 2.7.
  • New topExpressionCorrelates field — top 30 expression-vs-expression Pearson correlates of the focal gene. Parallel to topCoessentials (GE-vs-GE) and topCorrelates (GE-vs-expression), but answers a different axis: which other genes' expression covaries with the focal gene's expression in this cohort? Most useful for scatter / expression-correlate views where the focal axis is expression. Polarity description in dataStructure: positive r = co-regulated (often shared transcriptional program / phenotype state / lineage marker); negative r = anti-correlated. Same n-gate as the other correlate blocks (n ≥ max(50, 0.6 × cohortSize)). Suppressed when focal-gene expression SD < 0.05.
  • Variance warning extended to expression axis. Was previously gated only on the GE axis (|mean| < 0.3, SD < 0.15, frac essential < 5%). Now also fires when focal-gene expression has SD < 0.5 OR mean < 0.5 (log2-TPM+1, gene silent or stably expressed everywhere). Warning structure became { geneEffect: ..., expression: ... } so both can fire independently. Catches the case where a "what explains the variability in expression of X" question is asked about a silent / housekeeping gene.
  • New extras.pairCorrelation for scatter views — Pearson + Spearman + n + two-sided p between the two scatter axes, computed over the filtered cohort. Reviewer flagged that context.correlation was null in scatter exports; the most basic precomputed thing for a scatter view, missing entirely. Now precomputed at export time over whichever combination of GE / expression / dual-gene the user picked.
• Deferred from this round: applying the same precomputed-extras discipline to mutation_analysis / gate_comparison / correlation / cluster / expression_correlate view types (Priority 5 of the brief). Bigger scope; ship after the scatter improvements are tested.

v.81.0 (2026-04-27)

• AI .json.gz export — the precomputed-summary discipline (topCoessentials / topCorrelates / focalGeneTissueSummary / focalGeneMutationSummary / pathway-partner whitelist) was silently scoped to the gene-effect view because the focal-gene resolver only checked context.gene. Scatter / mutation / expression-correlate exports were quietly emitting empty extras while the _analysisInstructions still told the LLM to use those fields. Fixed: the resolver now falls back through context.gene → context.hotspotGene → context.targetGene → context.gene1, so every view that has a single focal gene now produces all the same precomputed summaries (caught when an Opus 4.7 review of a FAM167A scatter export found none of the extras present).
• New extras.focalGeneVarianceWarning — emitted when the focal gene sits in screen noise (|mean GE| < 0.3, SD < 0.15, fewer than 5% of cohort lines essential). Stops the LLM from confidently inventing biology from noise-driven correlations — FAM167A in 74 Skin lines (mean=−0.05, SD=0.11, 0/74 essential) is the canonical case where this warning fires. Step 0 of _analysisInstructions now tells the LLM to read this warning first and recommend stratification before pattern-hunting.
• Scatter view description rewritten. Was producing tautological text like "Scatter of FAM167A vs FAM167A filtered by Skin" when both axes were the same gene with different types (GE vs expression). Now: "Scatter of FAM167A gene effect (x) vs FAM167A expression (y), filtered by Skin." — explicit about which axis carries which data type.
• _analysisInstructions revised: Step 1 now says the precomputed scans are present "when the export includes them (gene-effect / scatter / mutation / expression-correlate views with a single focal gene)" rather than asserting they're always there. Multi-gene views (correlation / cluster / gate-comparison) are explicitly told to recompute. Eliminates the silent failure where an LLM following the instructions for an unsupported view kept hitting empty fields.
• schemaVersion 2.5 → 2.6.

v.80.9 (2026-04-27)

• "Reference Data" card — was eating the entire upper-left of the homepage after v.80.8 added all the new sources to the visible bullet lists. Collapsed back down to three named-source one-liners (DepMap 25Q3 · CORUM + Reactome · in-app curated KBs) with a "▾ show files & details" toggle that expands the full per-bucket file / KB list when needed. The Cell Lines / Gene Effect / Correlation / Open / Reset buttons are now visible without scrolling again.

v.80.8 (2026-04-27)

• AI .json.gz export — Reactome integration. Layered as a fourth tier in the focal-gene partner lookup, after CORUM. Catches pathway / signalling-cascade relationships CORUM doesn't cover — e.g. for IL4R, CORUM gives the receptor complex (IL13, IL13RA1, IL2RG, IL4) while Reactome adds the JAK/STAT cascade (STAT1, STAT5A, STAT5B, STAT6, plus PIK3CA and SOCS members from the broader signalling neighbourhood). New scripts/process_reactome.py fetches the GMT (~290 KB) directly from reactome.org via its public download URL (no auth, no manual intervention), filters to specific pathways (size 5–100 genes — drops broad parents like "Signal Transduction" with 2,600+ members), caps per-gene partners at 120 most-co-occurring. Output: web_data/reactome_partners.json (~6 MB; covers ~10,000 human genes — about half the protein-coding genome). Schema bumped to 2.5.
• "Reference Data" card on the main page restructured to show all data sources in three buckets: Cell-line data (DepMap files, expanded to include the new CSVs we ingest — OmicsInferredMolecularSubtypes, OmicsGlobalSignatures, OmicsFusionFilteredSupplementary, PRISM Repurposing), Pathway / complex memberships (CORUM + Reactome + in-app curated KBs), Curated annotations (clinical fusions, subtype hallmarks, STR profiles). Clearer attribution for CORUM and Reactome alongside the existing DepMap acknowledgement note.

v.80.7 (2026-04-27)

• AI .json.gz export — CORUM integration. The pathway-partner expression carve-out is no longer hand-curated for ~9 complexes; it now layers three sources so any focal gene gets meaningful partners:
  1. Hand-curated complexes (NEDD8/CRL, Proteasome, Hippo, MYC, TP53, BRCA, mTOR, BCL2, splicing) — tight, biologist-vetted high-value sets.
  2. CORUM (manually curated mammalian protein complexes, ~3,000 human genes covered) — auto-populates partners for arbitrary focal genes (SMARCA4 → BAF subunits, MCM4 → MCM2-7 helicase, RUVBL1 → INO80/SRCAP, etc.).
  3. Wiki cancer pathways — signalling-cascade groupings CORUM doesn't cover (RAS/MAPK, RTK family, p53-MDM2-CDKN2A axis).
  New scripts/process_corum.py reads a downloaded CORUM TSV (mips.helmholtz-muenchen.de/corum/, "Human" complexes) and writes web_data/corum_partners.json. Schema bumped to 2.4; dataStructure.expression documents the layered lookup. CORUM data file is optional — if not present, the export falls back to the hand-curated + wiki-pathway layers and still works.

v.80.6 (2026-04-27)

• AI .json.gz export — bug fixes after a third Opus 4.7 review of v.80.5. Two precomputed summaries had real bugs that were actively misleading the LLM (worse than not having the summaries at all per the reviewer). Schema bumped to 2.3.
  • topCoessentials n-gate. The list was entirely noise — every entry computed on n=23 or 24 cell lines because ~65 GE genes have screen-batch-only partial coverage and there was no minimum-n filter. Now gated at n >= max(50, 0.6 × cohortSize). The real biology (POMP, UBE2M, UBA3, SKP1 for NEDD8) should now surface as the actual top hits.
  • topCoessentials partner genes added back to the geneEffect matrix if they were dropped by the variance filter, so the LLM can verify the correlation by recomputing from the file (in v.80.5 the partner genes weren't in the matrix at all, making the values unreproducible).
  • Same n-gate applied to topCorrelates for consistency.
  • Both correlate sets suppressed when focal-gene SD < 0.05 — near-zero focal variance produces only noise correlations, so emit empty rather than 30 spurious entries.
  • focalGeneMutationSummary rewritten. Was ranking by |Δ| which floated meaningless n=3 entries (BCL6 with n_mut=3, p=0) above canonical drivers. Now split into two:
    • coreDrivers: TP53, KRAS, NRAS, BRAF, EGFR, PIK3CA, PTEN, APC, RB1, CDKN2A, NF1, NF2, STK11, SMAD4, MYC, IDH1, ARID1A, SMARCA4, KMT2D — always shown (n_mut ≥ 5) so "did TP53 status matter?" answers itself even when Δ is small.
    • topByEffect: top 20 from the extended panel (n_mut ≥ 10), ranked by |t| not |Δ| so small-n large-effect noise no longer outranks well-powered drivers.
    Each row now has a callType field (hotspot / damaging / hotspot+damaging). p-value approximation fixed — using Abramowitz & Stegun 26.2.17 normal-CDF (was a worse approximation that returned p=0 for |t|>~7, the literal printable bug the reviewer caught).
  • extras._method block added with concrete formulas for every precomputed summary (mirrored at dataStructure._method for cases where extras is omitted). Reviewer wanted it next to the values it documents.

Deferred: generalising the pathway-partner whitelist away from hand-curated complexes to CORUM / Reactome / STRING (separate task with its own data file).

v.80.5 (2026-04-27)

• AI .json.gz export — second round after another Opus 4.7 review of the v.80.4 NEDD8 export. The reviewer's ranked five all done. Schema bumped to 2.2.
  • New topCoessentials field — top 30 GE-vs-GE Pearson correlates of the focal gene, parallel to the existing topCorrelates (which is GE-vs-expression). Positive r means co-essentiality buffering (lines that depend more on partner depend less on focal); negative r means same-pathway co-essentiality. Reviewer flagged this as the more biologically central question for many gene-effect views (the NEDD8 example: the proteasome / cullin / Hippo co-essentiality story was the most informative finding and required a hand-rolled GE-vs-GE scan).
  • Pathway-partner whitelist for the focal gene's expression. Curated complex / machinery memberships for common analysis targets bypass the variance filter so housekeeping-ish pathway components are present in the expression matrix. Initial coverage: Neddylation/CRL (NEDD8 + UBA3, NAE1, UBE2M, UBE2F, RBX1/2, SKP1, CUL1-5, CAND1, COPS5/6, DCUN1D1/5), Proteasome (PSMA/B/C/D/POMP/ADRM1/USP14), Hippo/YAP-TAZ, MYC network, TP53/MDM2, BRCA/HR, mTOR/PI3K, BCL2 family, mRNA splicing core. Easy to extend in _FOCAL_GENE_COMPLEX_PARTNERS. Plus the existing wiki cancer pathways still feed into the always-include set when the focal gene is in one.
  • New extras.focalGeneMutationSummary — per common driver gene (~70 curated), Welch's t comparing focal-gene GE in mutated vs WT cell lines. Top 30 by |delta|. Saves the LLM from running per-gene t-tests by hand (the reviewer had to do TP53/KRAS/BRAF/etc. manually for the NEDD8 question).
  • bySubtype gated at n ≥ 5 (was 3). At n=3 the SD/z is noise and tempts the LLM to over-interpret. Tissue-level (byTissue) stays at n ≥ 3 since tissue groups are typically larger.
  • topCorrelates polarity guidance now in _analysisInstructions step 1. Was only in dataStructure before, so an LLM following only the analysis steps could misread sign. Step 1 now explicitly says: "topCoessentials (GE-vs-GE) … positive r = co-essentiality buffering within a complex, negative r = same-pathway co-essentiality. topCorrelates (GE-vs-expression) … positive r = high partner expression covaries with WEAKER focal-gene essentiality. Read both fields before running your own correlations." Step 3 also nudges the LLM to use extras before scanning the matrix.
  • Top-level nTotal field added (= length of cellLineOrder) so the LLM knows the cohort denominator without having to count.
  • dataStructure._method block added documenting the z-score formula, the topCorrelates / topCoessentials Pearson method, and the focalGeneMutationSummary Welch's t methodology — reviewer flagged that without provenance, citing numbers required guesswork.

v.80.4 (2026-04-27)

• AI .json.gz export — round of fixes after a real Opus 4.7 review of a NEDD8 export. The reviewer caught real gaps: NEDD8's own expression had been dropped by the variance filter, the neddylation machinery (UBA3, NAE1, UBE2M) wasn't in the expression matrix despite being the most informative genes for the question, and the AI had to hand-roll tissue rollups to discover that the Bone signal was entirely Ewing sarcoma. Schema bumped to 2.1.
  • Focal gene always included in both matrices regardless of variance threshold. For a NEDD8 GE export, the NEDD8 row of geneEffect and expression are guaranteed present.
  • Genes surfaced in topCorrelates are always included in the expression matrix. The expression-correlate scan runs before the expression matrix is filtered, and any gene that turns up in the top 30 is added to the always-include set. So the neddylation E1/E2/CRL components that drive co-essentiality with NEDD8 are now present in expression even if they're stable housekeeping-ish genes that the variance filter would otherwise drop.
  • New extras.focalGeneTissueSummary — precomputed per-tissue and per-subtype mean / sd / n / z-vs-overall for the focal gene's GE, sorted by mean. The reviewer had to discover by hand that "Bone = Ewing sarcoma", "Eye includes 5 RPE1 non-cancerous lines", "PNS = neuroblastoma"; this surfaces all of that automatically.
  • Schema docs updated — topCorrelates now has an explicit description (Pearson, focal-gene GE vs partner expression, with a note that n can vary per gene because expression coverage isn't uniform). The reviewer was confused enough by the missing description to recompute everything from scratch.
  • Step 2 confirmation gate softened: the analysis instructions now explicitly tell the LLM to skip the "ask the user to confirm" step when the question is specific and self-contained, only triggering it for open-ended questions like "explain the variability". Prevents the friction the reviewer flagged.

v.80.3 (2026-04-27)

• AI .json.gz export — fixes after a real export from the Gene Effect browser on all 1186 cell lines came out at 64 MB. Variance threshold alone wasn't tight enough at large n: ~9000 genes still passed |GE|>0.5 in 1186 cells, dominating file size. Added a per-matrix gene cap on top of the variance filter — after the threshold, genes are sorted by relevance (max-abs for GE, sd for expression) and the top N kept:
  • ≤100 cells: no cap
  • 100–500: cap 12 000
  • 500–1000: cap 6 000
  • 1000–1500: cap 3 500 (your case — should now be ~20–25 MB)
  • >1500: cap 2 500
• AI export file shape unified across all sources. v.80.1 had results-only sources (correlations / clusters / expression correlates) skip the matrices entirely, which produced very different files between e.g. the Gene Effect browser (64 MB with full matrices) and the Correlation browser (~2 MB metadata-only). With the gene cap, every source now ships the same shape — capped GE matrix + capped expression matrix + cell-line metadata + source-specific extras. Comparing files between sources should feel consistent now.
• AI dialog tier description updated to reflect the new "variance filter + cap" rules.

v.80.2 (2026-04-27)

• AI export dialog unified across all 8 source views (gene effect / scatter / mutation analysis / gate compare / correlation / cluster / expression correlate / GE-modal). The four sources that previously used a bare browser prompt() for the question (mutation, scatter, plus the four "Export for AI" menu items in correlation / cluster / gate / expression correlate panels) now open the same fixed-position modal as the GE-modal export. Every export path now shows: source name, cell-line cohort size, exactly what's getting written into the file (matrices vs results-only, with the active tier filters spelled out), the question textarea, and an "After download" suggestion pointing at claude.ai for the actual analysis (the embedded aiInstructions are tuned for it; ChatGPT / Gemini also accept .json.gz). Dialog dismisses on close button, outside click, or after a successful export.

v.80.1 (2026-04-27)

• AI .json.gz export — size fixes after a real export from the correlation analysis on all 1186 cell lines came out at 74 MB (claude.ai's attachment limit is ~30 MB).
  • Results-only sources skip the matrices. For Correlation / Cluster / Expression correlate exports, the analytical result is the result table itself (correlation pairs / cluster genes / per-gene r-values), which is already shipped in extras. Including the full GE matrix on top was dead weight — the AI doesn't need to re-derive correlations to interpret them. With matrices skipped, a 1186-cell correlation export should now be a few MB rather than 74 MB. Cell-line metadata, mutations, clinical fusions, inferred subtypes and global signatures are still included so the AI can reason about the cohort.
  • Tighter matrix tiers for the data-heavy sources (gene effect / scatter / mutation analysis / gates). Split the old 500–1500 band into 500–1000 (unchanged: GE>0.3, expr mean>2.0 OR sd>0.5) and 1000–1500 (new: GE>0.5, expr mean>2.5 OR sd>0.8). Top tier (>1500) tightened to expr mean>3.0 OR sd>1.0. Mutations / fusions / LoF / signatures still always full at every tier.
  • Tier-info text in the AI dialog updated to show which path applies (results-only vs matrix-bearing) and the new bands.

v.80.0 (2026-04-27)

• AI JSON export — major refactor (Phases 1+2+3 from the colleague-Claude review brief). All eight "Analyze with AI" buttons (gene-effect, scatter, mutation analysis, gate compare, correlation, cluster, expression correlate, plus the existing GE-modal export) now produce the same gzipped schema. Source-specific result tables (gate enrichment, correlation pairs, cluster genes, expression correlate r values, differential GE for mutation analysis) live under a single extras field; the context field tells the AI which view it came from.
• Phase 1 — size + tiering: the 400-cell-line hard cap is gone. New tier table: ≤100 full / 100–500 GE>0.2 + expr mean>1.0 OR sd>0.5 / 500–1500 GE>0.3 + expr mean>2.0 OR sd>0.5 / >1500 GE>0.5 + expr mean>2.0 OR sd>1.0. Mutations, clinical fusions, inferred subtypes, and global signatures are always full (they're tiny relative to the matrices and most useful at large n where stratification power matters). Expression filter switched from AND to OR so tissue-restricted markers (low overall mean, high SD — e.g. AFP in liver lines, MITF in melanoma) survive instead of being silently dropped. GE values now rounded to 3 decimals (was 4) — ~25% gzipped-size reduction with no analytical loss. Snapshot PNG dropped from gzipped exports (the AI was told to ignore it anyway, it inflated parse cost ~30% for a typical scatter, and the context field already gives the model everything it needs to restate the view).
• Phase 2 — schema cleanup: per-cell-line mutations unified from hotspotMutations: {gene:level} + damagingMutations: {gene:level} into a single mutations: {gene: {hotspot: 0|1|2, damaging: bool}}. New per-cell-line layers: clinicalFusions (curated driver fusion calls with tier), inferred (DepMap inferred subtypes — specificVariants like KRAS p.G12D, namedFusions, functionalLoss covering CDKN2A/RB1/PTEN/etc., msi binary), signatures (ploidy, wgd, cin, lohFraction, msiScore, aneuploidy from PureCN + MSIsensor2 + Ben-David). Added top-level schemaVersion: "2.0". Null fields (topCorrelates, cellLineGroups, extras, question) are now omitted when not applicable rather than emitted as null with documentation that an LLM had to disambiguate "not applicable" from "computation failed".
• Phase 3 — unified export paths: the five _buildXxxAIExport builders (mutation / gates / correlations / clusters / exprCorrelates) and the separate uncompressed JSON output are gone. exportForAI() is now a thin shim that prompts for an optional question, stashes it in the shared aiQuestion textarea, and routes through exportFullAIAnalysis. _getAICellLines extended to handle all eight sources: GE / scatter / mutation use existing logic; gates returns gateA ∪ gateB with stratification under cellLineGroups; correlations / clusters re-derive the main-panel filtered cohort via getFilteredCellLineIndices; exprCorrelates uses _exprCorrelateContext.subgroupIndices. Source-specific context blocks added per source so the AI can describe what the user was looking at regardless of view.

v.79.2 (2026-04-27)

• Cache-bust the app.js script tag (was ?v=20260224v1 from a February release; bumped to ?v=v79_1). Without this, browsers happily kept the cached old app.js while loading the fresh index.html, producing weird state where some buttons worked (their handlers hadn't changed) but the cell-line browser didn't open (the cached old JS referenced DOM IDs that no longer exist in the new HTML). Will re-bump on every JS release going forward.

v.79.1 (2026-04-27)

• CLB fusion filter dropdown rebuilt as a custom multi-line component — Safari truncates native datalist options to the input width, which was hiding the n= and clipping fusion names mid-word ("★ KMT2A-", "★ EWSR1-", and rendering several option rows visibly empty). The new dropdown shows two lines per entry: bold name with n=, then disease context for clinical fusions ("Ewing sarcoma", "CML / Ph+ B-ALL", …) or "any fusion involving this gene" for the gene-based long tail. Auto-completes as you type, capped at 200 visible items, click to commit a clean canonical name to the input.
• CLB toolbar buttons grouped with vertical separators: selection (Select Visible · Deselect All · Copy Names) | inspect (Inspect Gene Effects · Inspect Correlations · Open Saved Inspect — all green-tinted, run analyses on the selected cell lines) | export (Export Genes CSV · GE matrix CSV · All matrices CSV — now purple-tinted, raw-data exports). The three exports remain because they have non-overlapping scopes (typed gene panel vs genome-wide GE only vs all matrices); the wiki download (per-cell-line HTML) is a different artifact entirely.
• Cell-line wiki "Expected drivers" section reworked for clarity. Was confusing when the Oncotree subtype label itself contained a fusion name — users were reading "Chronic Myeloid Leukemia, BCR-ABL1+" as a positive call about the cell line rather than the disease classification. Now two clearly separated blocks:
  • Observed in this cell line (green) — pulls from every integrated layer: specific driver variants (KRAS p.G12D), driver fusions with tier badges, functional-loss tumour suppressors, top hotspot-mutated genes, MSI-high flag, WGD flag.
  • Typical for this disease subtype (blue, was the old block) — the curated knowledge-base description, plus the intersection with what's actually mutated in the line.
  Section header explains that the subtype label is a disease classification, not a per-cell-line call. When no curated KB entry exists for the subtype, the observed block still gives actionable info instead of just an "uncovered" message.
• Subtype knowledge base extended with 9 new entries: Chronic Myelogenous Leukemia (BCR-ABL1+), Chronic Myeloid Leukemia (BCR-ABL1+), Acute Promyelocytic Leukemia with PML-RARA, Diffuse Glioma, Cutaneous Squamous Cell Carcinoma, Esophageal Adenocarcinoma, Esophageal Squamous Cell Carcinoma, Stomach Adenocarcinoma, Endometrial Carcinoma.

v.79.0 (2026-04-27)

• Cell line Wiki restructured for clarity. Sections regrouped into four buckets: Identity (Cancer classification · Patient & sample origin · Sex), Genome state (NEW — Genome signatures), Driver landscape (Expected drivers for this subtype · Driver mutations · Functional loss NEW · Fusion landscape), Functional behaviour (CRISPR dependencies · Expression profile · Drug response), and Lab use (STR authentication · External resources). Headers shortened, each section opens with one short sentence saying what it contains, then the data, then a small grey "Source:" footer.
• Genome signatures section — Ploidy, WGD (in red when positive), Aneuploidy score (Ben-David 0–39), CIN, LoH fraction, and MSI score (MSI-high lines flagged red). One-liner per metric explaining what it means. WGD-positive lines (~58% of the panel) and MSI-high lines (~6%) are systematically different in many comparisons, so this is interpretive context that affects how every section below should be read.
• Functional loss section — new red-bordered chips for tumor suppressors that are functionally inactivated by integrated CN + mutation + expression evidence (RB1, TP53, PTEN, NF1, CDKN2A, VHL, MTAP, APC). Catches deletion-driven losses (CDKN2A homo-del in 445 lines panel-wide, RB1 deep deletion) that the damaging-mutation matrix alone is blind to.
• Driver mutations — the top-hotspots line now shows the specific variant in green next to the gene symbol (e.g. KRAS (p.G12D)) when DepMap's inferred-subtype pipeline calls a specific codon (KRAS p.G12D/G12C, BRAF V600E, EGFR L858R, JAK2 V617F, EGFR exon-19 del, plus generic group-level calls).
• Fusion landscape — clinically relevant fusions (BCR-ABL1 high, EWSR1-FLI1 high, …) now appear at the top of the section with green/amber/red tier chips and the ⚠ atypical badge where applicable, before the noisy raw partner list. Source footer points at the curated list provenance and the per-cell-line orthogonal validation.

v.78.1 (2026-04-27)

• CLB fusion filter dropdown — the n= count was getting truncated by Safari's narrow datalist subtitle row (showed "★ EWSR1-" rather than "★ EWSR1-FLI1 (n=14)"). Moved the full label (★ marker, fusion / gene name, n= count) into the option's value attribute so it renders as the only line. Filter lookup gained a small helper that strips the ★ prefix and the (n=N) suffix to recover the underlying name, applied everywhere the value is used (cell-line filter, GE-modal forwarding, active-filter badges, CSV export labels). Free-typing a canonical name like "BCR-ABL1" still works, with or without the (n=11) suffix.
• Inline ? tooltips next to the new section headers in the cell line detail panel: Clinically relevant fusions (recaps the orthogonal-evidence tier rules), Functional loss (explains the integrated CN/mutation/expression call and why it catches deletion-driven losses the damaging-mutation matrix misses), and Genome signatures (defines Ploidy / WGD / Aneuploidy / CIN). Hover the ? for the explanation without leaving the detail panel.
• DepMap genome-wide signatures integrated. New scripts/process_global_signatures.py + web_data/global_signatures.json (~80 KB) processes OmicsGlobalSignatures.csv — Ploidy (PureCN), WGD (whole-genome doubling, 692/1186 lines ≈58%), CIN, LoH fraction, MSI score (continuous), and Aneuploidy (Ben-David 2021 score). Each cell line now shows a one-line "Genome" row in the detail panel with the most useful four (Ploidy / WGD / Aneuploidy / CIN), each with a hover tooltip. WGD displayed in red when positive.

v.78.0 (2026-04-27)

• DepMap inferred molecular subtypes integrated. New scripts/process_inferred_subtypes.py + web_data/inferred_subtypes.json (~140 KB) processes OmicsInferredMolecularSubtypes.csv — per-cell-line specific hotspot variants (KRAS p.G12D, BRAF p.V600E, etc.), 11 named driver fusions, integrated LoF for 8 tumor suppressors (RB1, TP53, PTEN, NF1, CDKN2A, VHL, MTAP, APC), and MSI status. Cross-validation against our own clinical_fusions.json shows perfect agreement on BCR-ABL1 (12/12), EWSR1-FLI1 (14/14), EWSR1-ERG (3/3), ETV6-RUNX1, TCF3-PBX1, PAX-FOXO1.
• Cell-line browser detail panel surfaces three new things:
  • Specific hotspot variants in green next to the gene symbol — e.g. "KRAS (p.G12D)" instead of just "KRAS". Most-specific call wins (p.G12D > generic p.G12).
  • New Functional loss section showing integrated LoF calls as red-bordered chips. Most importantly catches deletion-driven losses (CDKN2A homo-del in 445 lines, RB1 deep deletion, etc.) that the damaging-mutation matrix alone is blind to — LoF combines CN < 0.3, likely-LoF mutation with AF > 0.5, or expression < 0.1 log-TPM.
  • Small red MSI badge next to the cell-line ID when MSIsensor2 score ≥ 20 (74 lines).
• CLB ? help modal updated with sections explaining the new hotspot-variant suffixes, Functional loss tier and what data sources feed it, and the MSI badge threshold.

v.77.1 (2026-04-27)

• Clinically relevant fusions — renamed (was "Clinical Fusions" which sounded off). Tier rules refined to add soft-pass logic: a strong signal in one channel + a directionally supportive signal in the other (in correct direction at half-magnitude) now reaches high, not just both-strong. This fixes the canonical EWSR1-FLI1 lines like A-673 (exprZ=1.9 strong, depZ=−0.42 supportive) which previously landed at medium only because FLI1 is broadly used in many lineages so absolute dependency z-scores stay small even when the call is genuinely fusion-driven. Atypical-with-no-support cases (e.g. BCR-ABL1 in an Ovary line, depZ=−0.08) still land at low because the supportive threshold is not crossed. Run now: 57 high / 23 medium / 4 low (was 51 / 29 / 4).
• Cell-line browser fusion filter dropdown now lists the clinically relevant fusion pairs first (★ BCR-ABL1 (n=11), ★ EWSR1-FLI1 (n=14), …) with live n= scaled to the active cohort filters (tissue / subtype / hotspot / etc.), followed by a separator and the existing gene-based long tail. Selecting a pair filters the cell-line list to the exact curated validated set; selecting a gene keeps the original behaviour (any line with any fusion of that gene). Filter and dropdown both honour the dual lookup so the n= counts and the resulting cell-line list always agree.
• CLB ? help modal — new section explaining what clinically relevant fusions mean, what the high/medium/low tiers are based on, why the ⚠ atypical badge appears, and what data source the calls come from.

v.77.0 (2026-04-27)

• Clinical Fusions — new curated driver-fusion classifier. Built from a 51-fusion anchor list (BCR-ABL1, EWSR1-FLI1, EML4-ALK, PML-RARA, TMPRSS2-ERG, etc. plus 7 KMT2A partners), each entry validated per cell line with three orthogonal signals: lineage match, partner expression z-score vs lineage-matched negatives, and partner dependency (CRISPR gene effect) z-score. Tier rules consider only the signals the curator marked as informative (e.g. BCR-ABL1 doesn't elevate ABL1 mRNA, so expression isn't checked — the call rests on the strong ABL1 dependency). Lineage is a modifier not a gate: a solid call in an atypical tissue (e.g. EML4-ALK in a Bowel line with ALK exprZ=51) keeps its tier and gets an "atypical" badge instead of being hidden. Cell-line browser detail panel now shows a Clinical Fusions block with green/amber/red tier chips, and the long noisy raw fusion list (formerly the only source) is collapsed under "Show all fusions" so it doesn't dominate. New scripts/clinical_fusions_anchor.json + scripts/process_clinical_fusions.py — pipeline reads the higher-quality OmicsFusionFilteredSupplementary.csv (Arriba-annotated), filters to confidence: high and drops read-through transcripts. Output: web_data/clinical_fusions.json. First run produced 84 calls across 24 fusions: 51 high, 29 medium, 4 low; 8 atypical-lineage calls (e.g. EWSR1-FLI1 in a CNS/Brain line, EML4-ALK in a Bowel line) preserved with their tier and a flag.

v.76.6 (2026-04-24)

• CLB cell-line list — when sorted by GE / Expression / mutation count, the inline number is now prefixed with a small grey unit tag (GE 0.45, Expr 4.12, hs 2), carries a tooltip that spells out the metric + gene(s) + direction ("negative = essential" / "higher = more expressed"), and GE values are tinted red when essential (< −0.5) or amber (< −0.2). A one-line caption above the list repeats what the numbers represent and which gene(s) they average.

v.76.5 (2026-04-24)

• p-value added to the plot heading for the remaining two gate-comparison views — the scatter/inspect gate compare (showGateGenePlot) and the GE-inspect gate compare (showGEGateGenePlot). v.76.1 only covered the UMAP/PCA variant; the other two still showed Δ without p, which is what the latest screenshot pointed out. Mutation uses the chi² approximation already used in the gate tables; GE and expression use Welch's t-test.

v.76.4 (2026-04-24)

• AI JSON export — rewrote aiInstructions so the assistant produces a concise one-line chart description followed by the question, instead of trying to render the embedded base64 snapshot. The previous wording encouraged pasting a Markdown image tag with the full data URI; some assistants took that literally and spewed a screen of base64 gibberish. The new instructions explicitly say the snapshot is metadata, not something to display, and give a filled-in sentence template derived from context.

v.76.3 (2026-04-24)

• AI JSON export — the embedded aiInstructions now tells the assistant to open its reply with two explicit paragraphs: (1) "You were looking at this chart:" + the image + a plain-English sentence describing what the chart shows (derived from context: analysis type, cohort size, applied filters); (2) "You asked:" + the verbatim question, skipped if empty. Analysis only starts after those two paragraphs. Previously the instructions just said to show the image and restate the question inline, without guaranteeing the AI would describe what the figure contained.

v.76.2 (2026-04-24)

• CLB gate-comparison Tissue / Subtissue tables now include a p-value column (2×2 χ² test of tissue presence vs gate assignment, same approximation the Mutations tab uses). The header is clickable to sort by significance and values below 0.05 are bold-weighted. Previously only mutations / diff-GE / expression tabs exposed a p-value.

v.76.1 (2026-04-24)

• CLB UMAP / PCA gate comparison — clicking a gene in the Mutations / Diff GE / Expression tabs now shows the Welch t-test (or chi² for mutations) p-value in the plot heading alongside the Δ already there, so significance is visible without switching back to the table.
• Fixed a silent crash in the gate-comparison GE plot (gateA.length / gateB.length were referenced but never defined in scope, throwing a ReferenceError before the title was rendered). Now reads this._umapGateA / this._umapGateB.

v.76.0 (2026-04-24)

• Inspect modal Genes (X/Y) row — the controls now wrap inside the box instead of pushing Find correlates outside it. Button widths reduced, the swap button sits between the X and Y fields, and the row uses flex-wrap so narrow modals stack cleanly.

v.75.9 (2026-04-24)

• CLB PCA loadings now appear in gene-color mode too, not only when colouring by tissue / subtype. The loadings sidecar describes the PCA axes (which don't change when you recolour), so it stays useful alongside any colour scheme.
• Long gene symbols in the loadings sidecar are no longer cropped. The SVG widens based on the longest label (measured with canvas.measureText), the 182×182 axis square stays centred, and both anchor positions (start on the right, end on the left) have room to spell out the full symbol.

v.75.8 (2026-04-24)

• Inspect / correlation browser — Update is now the primary (filled) button, the same size as Find correlates, so the main action is visually obvious.
• New ⇄ swap button next to the gene inputs exchanges the X and Y axes (both gene symbols and the GE / Expression types) and re-renders in one click.

v.75.7 (2026-04-24)

• CLB PCA / UMAP — the data area of the plot (not just the outer canvas) is now square. Canvas height is derived from the width minus the fixed margins, so the scatter box you see between the axes is a true square at any size. The H input is auto-synced to the computed value.

v.75.6 (2026-04-24)

• CLB PCA / UMAP — plot area is now the same across tissue / subtype / gene-color / cluster-highlight views. The right margin is fixed at 170 px and the legend (or colorbar) is pinned at x=1.02, so toggling the coloring mode no longer resizes the main plot.
• Gene-color colorbar title is placed above the bar (horizontal) instead of rotated on the right, and the tick font size is reduced, so long GENE_SYMBOL / Gene Effect titles no longer clip at the right edge of the figure.
• The Color by dropdown (tissue / subtype) is now live: changing it drops any active gene-color overlay or cluster highlight and re-renders with the new grouping. Previously you had to press Clear on the gene-color row first.
• Click-to-label and the label layer now work in gene-color mode too; previously clicking a point in gene-color view opened the cell-line detail but didn't place a label on the plot.

v.75.5 (2026-04-24)

• CLB PCA / UMAP — clicking a point now labels that cell line on the plot (same behaviour as the correlation scatter). Click again to remove the label, or use the Clear labels button that appears once at least one label is placed.
• New Cluster stats button opens a floating panel ranking tissues (or subtypes, via the Group by selector) by mean silhouette score in the current 2D projection. Higher = tighter cluster; the bar visualises the score on a −1 … +1 scale with the zero mark. Click any row to highlight its cells in green and render everyone else in gray; click the same row (or Clear highlight) to return to the tissue coloring.

v.75.4 (2026-04-24)

• CLB PCA / UMAP with a custom gene list now z-scores each gene (unit variance) before the projection. Without this, a single gene with a larger dynamic range (e.g. a strong dependency at ±3 alongside a partner at ±0.5) dominated PC1/PC2 and the plot showed a perfect gradient along that gene with the rest contributing noise. Top-variance (default, no list) keeps the old unscaled behaviour since those genes already have comparable magnitudes.

v.75.3 (2026-04-24)

• CLB Dimensionality Reduction now defaults to PCA instead of UMAP.
• New Genes input next to the method selector. If you enter a list (comma / semicolon / whitespace separated), the projection is computed from those genes only; leave it empty to keep the previous top-variance behaviour. Unknown symbols are skipped silently.
• When a custom gene list is used, a Highlight gene dropdown appears next to the Color by gene row. Pick any gene from your list to color points by that gene's GE (or Expression, when the analysis is run on expression data), without having to type the symbol back into the color-by box.

v.75.2 (2026-04-23)

• Find correlates — r now matches the scatter r for the same pair. Two bugs were feeding the discrepancy: (1) the X-mean and X-variance were fixed over the whole filtered set, but correlation with a candidate Y was computed on the cells where that Y was valid — a classical mismatch when Y had NaNs that X didn't. Fixed by recomputing Pearson pair-completely for each candidate. (2) The starting cell-line set was _inspectFilteredCellLineIds, which also excludes cells where the previously displayed Y gene was NaN. After clicking a suggestion the scatter rebuilt with a different cell set. Fixed by taking the filter set directly from DOM with no gene-validity pre-filter.

v.75.1 (2026-04-23)

• Correlation browser — tissue / mutation / fusion filter dropdowns are now populated and visible as soon as the modal opens (previously you had to enter genes and click Update first). Any selection you make is preserved when the plot is built.
• Find correlates — now refreshes the filtered cell-line set at click time and falls back to computing it straight from the DOM filters when the plot hasn't been built yet. Previously it could operate on all cell lines even though you had a tissue filter selected.

v.75.0 (2026-04-23)

• Scatter / inspect .json.gz (AI export) finally honours the inspect-modal filters. The cell-line set is pulled from _inspectFilteredCellLineIds (tissue, subtype, mutation, fusion, custom CL list all applied). The context block records each of those filters so the AI can see the cohort definition.
• Mutation .json.gz context now also records excluded tissues, additional hotspot, oncoprint gates, and the custom cell-line list count.
• Scatter and mutation .json.gz buttons now prompt for an optional question; previously only the gene-effect view exposed a textarea.
• Every AI export embeds a PNG snapshot of the current chart (inspect scatter / GE) and an aiInstructions field telling the assistant to open its reply with "You were looking at this chart:" and render the image back before answering.

v.74.9 (2026-04-23)

• Find correlates — Search field filters both lists by gene-symbol substring.
• Find correlates — column headers are now clickable to sort. Gene toggles alphabetical asc/desc; r toggles signed desc/asc. Default sort is still |r| descending.

v.74.8 (2026-04-23)

• Find correlates — clicking a result now preserves the inspect modal's filters (tissue / subtype / mutation / fusion / custom-CL list). Previously, selecting a suggested gene rebuilt the scatter from the main parameter filters and dropped the inspect-local ones, so the r shown on the new plot was computed on a different cell-line set and didn't match the value in the correlate list. The r now matches.

v.74.7 (2026-04-23)

• Find correlates modal now has an |r| ≥ threshold (default 0.3) that filters both the GE and Expression lists live. Each list shows its own count (and "showing top 500" when the full set exceeds the display cap).
• An Enrichr button next to each list sends the currently-thresholded correlates to Enrichr for pathway enrichment (GO:BP / MF / CC, KEGG, Reactome).

v.74.6 (2026-04-23)

• CLB Export Genes CSV: dropped the Sex_expression column. The CSV now carries a single Sex column (annotated sex) alongside Tissue and Subtype.

v.74.5 (2026-04-23)

• Correlation / cluster Export for AI now captures the filter state applied when the analysis ran — lineage, sub-lineage, excluded tissues, hotspot, fusion, oncoprint gates, selection label, and custom cell-line list — so the AI can see the cohort definition, not just the results.
• Every Export for AI button now prompts for an optional question; the text is written into the JSON as question.
• Every Export for AI export embeds a PNG snapshot of the current view as a data URI under snapshot, plus an aiInstructions field telling the assistant to open its reply with "You were looking at this chart:" and render the image back, then restate the question, then answer from the data.

v.74.4 (2026-04-23)

• CLB ? help text updated with the new sort options: GE for gene(s) and Expression of gene(s). Explains that multiple gene symbols are averaged per cell line (comma / semicolon / newline / whitespace separators).
• New Export Genes CSV button in the CLB toolbar. Prompts for one or more gene symbols and exports a tidy CSV with one row per selected cell line (or per visible cell line if nothing is selected). Columns: CellLineID, CellLineName, Tissue, Subtype, Sex_annotation, Sex_expression, GE_<gene>, Expr_<gene> — two data columns per gene.
• New Find correlates button in the correlation / inspect scatter modal. Takes the X-axis gene (GE or Expression, as selected), applies the current inspect filters (tissue / hotspot / fusion / custom-CL), and returns two lists side-by-side: top GE correlates and top Expression correlates (|r|-sorted, n=100 each). Click any gene to set it as the Y axis with the matching data type and redraw the scatter.

v.74.3 (2026-04-23)

• Gene Effect analysis modal can now plot expression instead of gene effect. New GE / Expression toggle next to the gene input; switching re-analyses the current gene against the expression matrix (log2-TPM+1) and redraws the By Tissue box plot. The label, axis title, and hover read Expression in expression mode. Switching from the mutation-inspect layout to Expression falls back to the standalone view (0/1/2 stratification is tied to GE).

v.74.2 (2026-04-23)

• Cell Line Browser sort gains two features:
  • Sort by Expression (in addition to Gene Effect). New option in the Sort dropdown; defaults to descending so highest-expressing lines surface first. Falls back to a name sort until expression data is loaded.
  • Multi-gene input — the Gene box now accepts one or more symbols (comma, semicolon, newline, or whitespace separated). Multiple genes are averaged per cell line, so you can rank cells by combined expression / dependency of a set. Useful for finding cell lines with the strongest signature across a pathway or marker panel.

v.74.1 (2026-04-23)

• Network PNG / SVG exports now match the on-screen layout: the filter banner sits in its own header strip above the network rather than being drawn over the upper nodes. Both formats reserve a header area sized to the banner font, paint a light-grey strip with a thin separator (matching the on-screen header), centre the filter text in it, and shift the network and legend down by the strip’s height so nothing overlaps. Transparent-background mode keeps just the centred banner text without the grey strip.

v.74.0 (2026-04-23)

• Network filter banner moved out of the canvas overlay and into a dedicated header strip above the network plot. The earlier overlay kept landing on top of nodes whatever pan was applied because vis.js’s physics + fit re-centred the layout. With a strip outside the canvas, the network has the full plot area and the banner is unambiguously above it. PNG / SVG exports still draw the filter text inline at the top of the figure (no change to the export look).

v.73.9 (2026-04-23)

• Correlation scatter highlights are no longer an unreadable blob when many cell lines are selected. Two-tier behaviour: when up to 12 cells are highlighted, each is labelled with just its name (lineage dropped from the label, since it’s already in the hover); when more than 12, the per-point labels are dropped entirely — the orange markers stay, hover still shows each name, and a small caption pinned in the upper-left says “● N cell lines highlighted (hover to see names)”.

v.73.8 (2026-04-23)

• Curated-collection descriptions rewritten to be honest about what each set captures and what it doesn’t. Each entry now has a structured Inclusion / Mechanism / Caveat block with the explicit gene panel, the biology, and the limitations (e.g. SWI/SNF-deficient now scoped specifically to the BAF/PBAF complex subunits, with a caveat that the broader chromatin machinery — histone writers/readers/erasers, DNMTs, etc. — is not captured). Categories also tightened: “Chromatin” → “Chromatin remodelling — SWI/SNF complex”; “DNA repair” split into “mismatch repair” and “homologous recombination”; “Breast subtype” → “Breast — receptor subtype” with explicit “approximate” tags reminding the reader that transcript-based receptor calls are not clinical IHC/FISH.

v.73.7 (2026-04-23)

• Network filter banner pan: the previous offset was being reset by a fit() call inside the stabilization callback. Moved the pan into the same callback, after the fit, so the top nodes really do drop below the banner now.
• Clicking a correlation row from the CLB inspect-correlations modal now also pre-populates the scatter’s highlight list with the selected cell-line names — so the cells driving the result show up labelled in red on the correlation scatter, set against the rest of the dataset.

v.73.6 (2026-04-23)

• Network filter banner cleaned up. When the analysis was launched from a CLB selection, the trailing “n=1186” (the full dataset size) used to appear after the “Selection (58 cell lines)” tag, contradicting it. Suppressed when a selection label is present, since the count is already in the label.
• Banner now sits a bit lower with an opaque white background and a subtle shadow so it doesn’t blend into the network. The network view also pans down ~40 px after the initial physics settle, giving the top nodes breathing room below the banner.

v.73.5 (2026-04-23)

• Selection-Inspect (GE / Correlations) results can now be saved to a JSON file and re-opened later — no more losing a 5–10 minute correlation run if you wanted to look at a different pair afterwards. Save button appears in the modal header once results are computed; Open Saved Inspect button next to the Inspect Correlations button on the CLB toolbar pops a file picker, restores the result table, and re-wires all interactivity (sort, cutoff, Network, Enrichr).
• Clicking a correlation row in Inspect Correlations used to close the inspect modal AND open the correlation popup behind the Cell Line Browser, leaving it invisible. Two fixes: the inspect modal now stays open (it's still useful), and the correlation popup z-index was raised above the CLB so it surfaces on top.

v.73.4 (2026-04-23)

• Curated collections dropdown moved to its own row below the main CLB toolbar with a ? info button that opens a reference modal listing each collection’s inclusion criteria + current dataset count. Adding / editing collections in the catalog automatically updates both the dropdown and the reference modal.
• Sex filter split into By annotation (Male / Female / Unknown) and By expression (Male / Female / Unknown) option groups. Each gets a live count under the current filter intersection. Previously “Unknown” collapsed the two axes together, hiding that many cell lines are annotation-called but expression-unconfirmed (or vice versa).
• Inspect Gene Effects overhauled:
  • Row count now configurable (default 200, up to 2000 per side) via a Top N input.
  • Sortable columns — click any header to sort ascending / descending. Headers and numeric columns are centred; Gene column stays left-aligned.
  • Per-side |Δ| cutoff input filters the table instantly. Right side defaults to |Δ| ≥ 0.1.
  • Per-side Network button — prompts for r cutoff, then builds a correlation network from the currently-displayed genes restricted to the selected cell lines. Same flow as the Inspect Correlations network button.

v.73.3 (2026-04-23)

• Inspect Correlations: refactored the inner loop. The previous implementation did a full first pass to build the top-r-in-selected list, then a second pass that recomputed the same values while also computing r-in-rest. Now it’s a single pass that emits both top lists in one go. Baseline changed from selected vs rest to selected vs all cells; the all-cells centred-vector matrix is cached on the instance so subsequent inspect runs skip that setup entirely. Warning updated to be honest about cost: up to 5–10 minutes on a typical laptop.
• Clicking a gene pair in the inspect-correlations results used to require an analysis to have been run already (the old path tried to find the pair in this.results.correlations and silently failed when it wasn’t there). Now it opens the correlation inspect directly with a synthetic correlation stub — always works.
• Clicking a gene in Inspect Gene Effects now highlights the selected cell lines in the By-Tissue box plot in red with a larger marker and a dark outline, so you can see where the selected lines fall within each tissue. The subtitle picks up a small “● N selected cell lines highlighted” note. Highlight clears when you close the GE modal.

v.73.2 (2026-04-23)

• CLB dropdowns now show n=X per option with live scaling to all active filters — the Sex dropdown and Curated collections dropdown are now joined to the tissue / subtype / hotspot / fusion / custom-CL filter state (previously only tissue / subtype / hotspot / fusion had live counts). Counts exclude the dropdown’s own selection when computing its labels so switching is predictable.
• Inspect Correlations: each column gets an |r| ≥ / |Δ| ≥ cutoff input and a Network button. Typing in the cutoff filters the table instantly (no recompute). Clicking Network prompts for a cutoff, pulls the genes from pairs above the cutoff, populates the main gene list + the current selection as a custom cell-line filter, and runs the analysis — yielding a correlation network of those gene pairs restricted to the selected cell lines.
• Removed the now-redundant Transparent background checkbox from the Network export sidebar. The publication-quality export dialog asks about background when you click Export, so the separate checkbox was duplicative.

v.73.1 (2026-04-23)

• Two new buttons in the Cell Line Browser toolbar: Inspect Gene Effects and Inspect Correlations, operating on the currently-selected cell lines.
  • Inspect Gene Effects — fast (< 1 s for typical selections). Two panels: most depleted in selection (lowest mean GE) and largest Δ vs the rest (most differentially essential). Click a row to open Gene Effect inspect for that gene.
  • Inspect Correlations — unbiased all-vs-all Pearson correlation across the selection. Shows a size estimate + confirmation first; during compute a progress bar with ETA and a Cancel button. Two panels: strongest pair correlations in the selection and largest Δr (selection r − rest r). Click a row to open the correlation inspect for that gene pair.
  Helpers _showProgress / _hideProgress wrap a generic cancellable progress overlay that’s ready for future long-running analyses.

v.73.0 (2026-04-23)

• Cell Line Browser gains a Curated collections dropdown next to Tissue / Subtype. Multi-gene / phenotype groups that aren’t well served by a single-gene filter:
  • DNA repair: MMR-deficient (MSI-high), HR-deficient (BRCAness)
  • Mutation burden: Hypermutated (top decile)
  • Chromatin: SWI/SNF-deficient
  • Oxidative-stress: NRF2-activated
  • Expression signature: Neuroendocrine phenotype, EMT-high
  • Breast subtype: Triple-negative (TNBC), HR+ (luminal), HER2+ — calls are relative to breast-only thresholds
  Memberships computed at load (mutation-based ones) and on expression-data load (signature / breast-subtype ones). Dropdown shows live count per collection and is grouped by category. (Inspect correlations / gene effects buttons for selected cell lines coming next.)

v.72.5 (2026-04-23)

• Export dialog gains three size presets — Small (5 cm), Medium (10 cm), Large (15 cm) — that set the width and auto-scale the height from the on-screen plot aspect. New Lock aspect checkbox (default on): when enabled, using +/− on Width or Height rescales the other dimension to preserve the plot aspect, so the figure doesn’t stretch accidentally.

v.72.4 (2026-04-22)

• Default export DPI raised from 300 to 600; max raised to 2400. Step is now 100 (with the −/+ buttons). The hint reads “300 = screen / draft; 600 = print; 1200+ = high-quality print”. Existing exports were noticeably soft because the network is rasterised from a DOM canvas at screen density.
• Network legend frame is now opt-in. The faint gray box around the bottom legend is gone by default; the dialog gains a checkbox Draw a thin frame around the legend for users who want it back. Same toggle works for both PNG and SVG network exports.

v.72.3 (2026-04-22)

• Number boxes throughout the UI now have − / + step buttons on either side (matching the Plot Size / Axis Ranges pattern in the inspect modal). Wired up to: Min Cell Lines on the main page, Highlight / Color font size in the inspect modal, Min n in the Gene Effect modal, the export-options Width / Height / DPI fields, and the UMAP/PCA Width / Height. Backed by a new app.adjustNumber(id, delta) helper that respects each input’s min/max/step and fires change events so re-renders run as expected.
• Default Highlight / Color font size in the inspect modal raised from 3 to 5 for better readability.
• Inspect (correlation) modal no longer closes on backdrop click or on Escape — same exit-safety policy as the Cell Line Browser. The X in the upper-right is the only exit; otherwise typed input or a stray click could lose all filter / gate / highlight state.

v.72.2 (2026-04-22)

• Publication-quality export dialog now appears for every chart export, not just the Plotly-based ones. Added to the inspect scatter (was missing), the network PNG, the network SVG, and the UMAP/PCA export. Each respects the chosen size, DPI, and background; PNG outputs include the pHYs chunk so Word / Office / LaTeX honour the physical size, and SVG outputs use cm units on the outer tag.
• Removed the “Solid white is recommended for PowerPoint / PDF…” line from the export dialog — the app is for general use and shouldn’t recommend specific downstream tools.

v.72.1 (2026-04-22)

• “Adjust graph size with Width/Height sliders…” hint under the Gene Effect chart is now hidden whenever the sliders themselves are hidden (By Tissue overview + By Hotspot scan). It only appears in the detailed tissue view and Hotspot Mutation Inspect where the sliders actually exist.

v.72.0 (2026-04-22)

• Clicking any chart PNG / SVG Export button now opens a publication-quality export dialog. Defaults: width 5 cm, height auto-scaled to preserve the on-screen aspect ratio, 300 DPI for PNG, and solid white background. Width / height / DPI / background are editable per export; the last chosen values are remembered for the session.
• PNG export now honours the chosen DPI: the canvas is sized to widthCm × dpi / 2.54 px, and a pHYs chunk is written into the PNG so Word / PowerPoint / LaTeX place the image at its real physical size instead of the default 96-DPI guess.
• SVG export sets absolute cm units on the outer <svg> tag so downstream tools (Illustrator, Inkscape) open at the intended print size.
• Transparent backgrounds are opt-in — default is solid white. This fixes the faint edge lines that appeared when transparent PNGs were pasted into PowerPoint and then exported to PDF (alpha-edge compositing artifact).

v.71.3 (2026-04-22)

• Caution banners in Hotspot Mutation Inspect, the By Hotspot scan, and the correlation-analysis mutation view now add a concrete recommendation: For meaningful analysis, narrow down using the All tissues and subtype filters above so WT and mutated cell lines come from the same cancer context. (Previously only warned about the confound, without telling the user what to do.)

v.71.2 (2026-04-22)

• “Draw rectangles on plot to define gates” status text now lives inside the Gates group span, so it hides together with the gate buttons in By Tissue and By Hotspot views (where gates aren’t available) and only shows in Hotspot Mutation Inspect.

v.71.1 (2026-04-21)

• Swept all chart exports in V2 through a single helper (_exportPlotly → _expandSvgToContent) so every PNG/SVG download now matches the on-screen plot and never crops content. Previously 6 separate export functions each had their own “Plotly.toImage → blob → canvas” pipeline (upset plot direct-PNG from Plotly, correlation analysis by tissue/hotspot, tissue by-gene chart, expression-correlate scatter, gene-effect chart, mutation-inspect chart) — some missed viewBox expansion and the upset-PNG path skipped SVG entirely. All now: match _fullLayout dimensions, get SVG from Plotly, measure true content bbox, expand viewBox, then either serialize (SVG) or rasterise at 2× (PNG). Removes hundreds of duplicated lines.

v.71.0 (2026-04-21)

• Gene Effect chart PNG / SVG export now matches the on-screen plot exactly. Previous code forced a minimum height of 650 / 550 + 40 “for safety” which stretched the exported plot taller than it appears on screen, and still cropped the stats annotations below the box plot (the Skin p-value line was cut off in the last report). The new path uses _fullLayout.width / height as rendered (which already includes Plotly’s reserved bottom margin for annotations), then runs a post-pass _expandSvgToContent helper that measures the true content bounding box and enlarges the viewBox so nothing outside the nominal figure gets cropped. Applies to both formats.

v.70.9 (2026-04-21)

• Gates are now Mutation-Inspect only. The root problem: when a gate rectangle was drawn on the By Tissue box plot, the filtering code silently ignored the Y axis (_geGatePlotData is only populated for the mutation-inspect scatter) and matched cells across all tissues at the same gene-effect level — so a rectangle around 10 cells in one tissue returned hundreds of cells across all of them. The Gates UI now only appears in Mutation Inspect view, where the 2D rectangle is meaningful (gene effect × 0/1/2 mutation level). Switching to By Tissue or By Hotspot hides the group and clears any active gates.

v.70.8 (2026-04-21)

• GE gate buttons: the Compare button sometimes stayed hidden after both gates were drawn. Cause: the show-handler set style.display = '' which clears the inline style but lets the stylesheet default take over, and in some render paths that default was resolving to none before the inline-flex parent recomputed. Set display: inline-block explicitly for both the Compare and Clear buttons on gate creation so they always show.

v.70.7 (2026-04-21)

• By Hotspot scan: clicking a hotspot-gene row now hands off directly to Hotspot Mutation Inspect with that gene pre-selected (0/1/2 plot in the full mutation-analysis inspect layout), instead of rendering the old separate detailed graph. Same data, more useful chrome (compare buttons, custom exports, filter context).
• Hotspot Mutation Inspect now shows the same caution banner that the By Hotspot scan has: hotspot mutations are often enriched in specific cancer types, so GE differences can reflect lineage rather than the mutation itself.
• Δ Tissue / Δ Hotspot / Δ Fusion compare buttons moved from row 2 (Filters) to row 3 (View + Gates), appearing after the Gates group. They’re compare actions that only show in Mutation Inspect mode, so grouping them with the view / gate actions fits better than living in the filter row.

v.70.6 (2026-04-21)

• In Gene Effect → By Tissue, clicking the pinned All row at the top of the stats table now returns to the all-tissues overview when a specific tissue is currently focused (previously the “All” row was inert). The hover title reads “Click to return to the all-tissues overview.” Any active filters stay applied.

v.70.5 (2026-04-21)

• Gene-Effect modal toolbar reorganised. Row 2 is now a dedicated “Filters:” row (mirroring the “Gene Effect:” legend on row 1), ending at the compare buttons. Row 3 is a dedicated “View:” row with the three mode buttons + “Gates:” group — so view buttons and gate buttons always sit together regardless of window width (no more wrapping apart).
• Third view button renamed from “Mutation Inspect” to Hotspot Mutation Inspect to make clear it’s specifically the hotspot-mutation 0/1/2 stratification (not any mutation type).
• Hotspot-gene selector (the one next to Analyze, which sets the Y-axis stratification gene) now shows the mutation count per gene under the currently-active filters — e.g. TP53 (88 mut). Options are also sorted descending by count so the most-informative genes surface first. Label changed to Hotspot mutation (Y):.

v.70.4 (2026-04-21)

• Mutation Inspect refined. Only the mutation-analysis inspect’s existing hotspot-gene selector (the one next to Analyze) is shown now — the extra dropdown next to the View toggle was redundant. Clicking Mutation Inspect defaults the hotspot gene to TP53 (or the first available one) and hands off to the full mutational-analysis layout; change the stratifying gene in the selector next to the plot. The View toggle (By Tissue / By Hotspot / Mutation Inspect) now stays visible in all three states so the user can switch back to the standalone views from Mutation Inspect without closing the modal.

v.70.3 (2026-04-20)

• Third Gene-Effect view renamed to Mutation Inspect and reworked to reuse the existing mutational-analysis inspect layout instead of a separate simplified plot. Clicking Mutation Inspect surfaces a hotspot-gene picker; picking a gene bootstraps a minimal mutation-analysis context on the fly and hands off to showGeneEffectDistribution — the resulting view is identical to the Inspect opened from a mutation-analysis results table (full-width chart, hotspot / fusion filters, Δ Tissue / Δ Hotspot / Δ Fusion compare buttons, custom export). Clicking By Tissue or By Hotspot from that state cleanly restores the standalone GE layout. By Hotspot (scan) label reverted to By Hotspot now that the three views are clearly distinct. Removed the v.70.2 stand-alone “By Mutation” box plot (duplicated what mutation-analysis inspect already does better).

v.70.2 (2026-04-20)

• Gene Effect inspect gains a third view: By Mutation. Pick a hotspot-mutated gene from the dropdown and the plot stratifies the current gene’s effect by 0 / 1 / 2 mutation levels of that gene — the same view you get from “Inspect” in a mutation-analysis result table, now reachable from the standalone GE inspect. Rolled back v.70.1’s attempt to share the toggle with the mutation-analysis inspect (“By Hotspot” means subtly different things there, which was more confusing than helpful). “By Hotspot” renamed to “By Hotspot (scan)” in the standalone inspect to make its scan-table semantics explicit. Mutation-analysis inspect is back to its original chromeless look.

v.70.1 (2026-04-20)

• Mutational-analysis inspect and gene-effect inspect are now the same modal with a stratify toggle. Previously they were two separate entry points (“Inspect” in the mutation-analysis results vs. clicking a gene name elsewhere) that looked similar but showed subtly different views — a recurring source of confusion. Now the mutation-analysis inspect keeps the existing By Hotspot look and feel (0/1/2 level plot, compare buttons, exports) but exposes the existing By Tissue | By Hotspot toggle, so the user can flip the stratification axis without leaving the mutation-analysis context. Switching to By Tissue re-plots the same gene grouped by lineage while keeping the mutation-analysis filters (lineage filter, excluded tissues, subLineage filter) in effect. Switching back to By Hotspot returns to the mutation-stratified plot exactly as before.

v.70.0 (2026-04-20)

• Cell Line Browser is now exit-safe. Previously a stray click on the dark backdrop or an Esc keystroke (e.g. while dismissing a gene tooltip) closed the whole browser, taking all filters, selections, and the current inspect state with it. The X in the upper-right corner is now the only exit path. Other dialogs that are cheap to reopen (Aa text settings, help popovers, etc.) still dismiss on outside click / Esc as before. Same change applied to V1 and CoExpress.

v.69.9 (2026-04-20)

• Escape on a pinned gene tooltip was bubbling up and closing the Cell Line Browser (and any other modal that listens for Esc). The tooltip now intercepts Escape in the capture phase, stops propagation, and only dismisses itself — the underlying modal stays open. Ported to V1 and CoExpress as well.

v.69.8 (2026-04-20)

• Gene hover tooltip now carries links and supports Shift-to-pin. Every tooltip has a small footer with a direct NCBI Gene ↗ link (and HGNC when available) so the source of the summary is always one click away. Shift-hover (or press Shift while a quick tooltip is showing) pins the tooltip: it widens, shows the full untruncated summary, gets a close button and a green border, and stays open until you press Esc or click outside. Non-pinned tooltips still fade on mouseleave as before.

v.69.7 (2026-04-20)

• UMAP/PCA hover tooltip: · HTML entities were leaking through as literal text because Plotly’s hover renderer doesn’t parse named entities. Replaced with the actual Unicode middle-dot (·). The “click to inspect” hint was also a washed-out grey that didn’t read against the tooltip background — now shown in italic at the default tooltip text color so it’s clearly visible.

v.69.6 (2026-04-20)

• PCA loadings sidecar now top-aligns with the plot. The “Loadings (top 8)” header sits at the same vertical position as the Plotly plot title, so the two axes boxes share a common top border.
• PNG and SVG export now include the loadings graph when it’s visible. The plot SVG and the loadings SVG are composed into a single outer SVG (plot on the left, title + loadings on the right), then downloaded as SVG or rasterised to PNG at 2x. Previously the loadings were on screen only.
• New button under the plot: Select in list above. After lassoing a cluster, press it to tick those cell lines in the browser list — they then flow into Export GE and Full Data Export. Pairs naturally with the existing Copy selected.
• PNG, SVG and Aa buttons moved from the “Display” row (row 2) to the main Run/Reset row (row 1), with a small divider between them and the coloured action buttons. They’re less buried now.

v.69.5 (2026-04-20)

• UMAP/PCA point hover tooltip rebalanced. Instead of just “Hotspot mutations: N” (which on its own said little about genetic complexity), hover now shows Hotspot N · Damaging N · Fusions N side-by-side, so one glance compares the three mutation axes. A small “click to inspect” hint is also added.
• Clicking a point in the plot now opens the inspect panel for that cell line (same as clicking its row in the list above). The matching list entry is scrolled into view and highlighted, so you can jump straight into the Wiki, gene lists, etc. for any cell line you spot on the plot.

v.69.4 (2026-04-20)

• Short explainer added above the Lasso select / Copy selected buttons under the UMAP/PCA plot: draw a lasso on the plot to pick a cluster of cell lines, then copy their names for downstream use (paste into Custom CLs above, a new session, or a spreadsheet). Both buttons also gained descriptive tooltips.

v.69.3 (2026-04-20)

• Dimensionality-reduction tissue dropdown gains a new top option: “Visible in list above (n=X)”. Pick it to project only the cell lines currently rendered in the Cell Line Browser list — so whatever you’ve filtered upstairs (search, custom CLs, mutation state, subtype, …) drives the UMAP/PCA directly. The count stays in sync with the list above, and selecting it hides the subtype dropdown (it would be meaningless against an arbitrary cell-line set). Default selection on open.

v.69.2 (2026-04-20)

• UMAP/PCA plot defaults to 500 × 500 (square) instead of 580 × 480, so the scaleanchor-constrained axes don’t leave awkward dead space.
• PCA loadings moved from a floating inset to a proper sidecar next to the main plot, labeled “Loadings (top 8)”. Size bumped 30% (140 px → 182 px) with correspondingly bigger gene labels and axis tags. No more accidental cover-up of the PCA scatter.

v.69.1 (2026-04-20)

• Dimensionality-reduction control panel reorganised from six ragged rows into three aligned ones. Row 1 is the compute inputs (method, data, tissue, subtype, color-by, Run, Reset); row 2 is display (axes, dot size, width/height, PNG/SVG/Aa/Top-Genes/Loadings); row 3 is overlays (Color-by-gene, Split, Gate-Compare). All controls share a common 24 px height with a consistent 11 px font, group dividers between related button clusters, and a flex-spacer that pushes Run/Reset to the right so the eye can find them in a fixed place
• Default UMAP/PCA dot size bumped from 6 to 9 so cell lines are visible without reaching for the size slider
• Fixed: splitting by mutation and then pressing Clear left the right-hand WT and Mutated panels on screen as ghosts. _renderUmapPlot now purges the child Plotly instances and wipes the plot container before re-rendering the single-panel view

v.69.0 (2026-04-20)

• Cell-line inspect panel reorganised for readability. Subtype and other long values no longer run flush against their labels on wrap (fixed with proper right-alignment and a label gap). Wiki deep-dive button moved to the top of the panel so it’s visible before scrolling. Hotspot, Damaging and Fusion gene lists now show the first 10 cancer-panel genes (bold) with a Show all toggle for the rest — Jurkat’s 658 damaging mutations and 48 fusions no longer dominate the panel. Gene Effect section separated from the mutation block by a thin divider and relabeled (“Gene Effects” with Genes scored / Mean / Range), which keeps Most Depleted and Most Enriched as clear sub-sections underneath. Removed the redundant “Cell Line Detail Report” CSV button (the Wiki already covers this and now has a Download button that saves a standalone, styled HTML file per cell line).
• Version scheme switched to single-decimal (v.69.0, v.69.1 … v.69.9, v.70.0). Small changes bump the tenths digit, large changes bump the integer.

v.68.02 (2026-04-19)

• Pathway-scan section in the Mutation profile clarified. The intro now states explicitly that the scan is based on mutation calls only (copy-number, fusions and expression are excluded); drops the inaccurate "hotspot = oncogenic / damaging = tumour suppressor" shorthand; and explains the asterisk (gene carries both a hotspot and a damaging mutation). Pathway rows restructured so the name, plain-English description, and gene list each sit on their own line, with a clear "Mutated genes in panel:" label before the tags so the genes no longer run on directly after the description. Cryptic shorthand like "RTK (receptor tyr kinase) — receptor TKI targets — check amplification / fusion" is replaced with full sentences, e.g. a note on receptor tyrosine kinases explaining that the mutation panel misses amplifications and fusions — common driver mechanisms for this pathway — so a clean panel does not rule it out

v.67.92 (2026-04-19)

• New Pathway status (genotype + dependency) sub-section in the Mutation profile. For 5 clinically-important pathways (p53, RB / cell cycle, RAS-MAPK, PI3K/AKT, EGFR + BCR-ABL when relevant) it combines mutation status with CRISPR knockout gene-effect to infer whether the pathway is functionally active. Catches the common mismatch where a TP53-wild-type cell line shows neutral TP53 knockout (pathway dormant despite WT genotype), or where TP53-WT cells with positive TP53 gene-effect are actively restraining growth and become MDM2-inhibitor candidates. Each card shows genotype, the relevant gene-effect read-outs, and a one-line interpretation

v.67.82 (2026-04-19)

• Mutation-profile flags turned into titled cards instead of a running block of colored pills. Each flag now has a short clear title ("Mismatch repair is broken", "Very high mutation burden", "TP53 is mutated", "G1/S cell-cycle brake lost", etc.) with a one-line plain-English explanation of what it means and why it matters. Much easier to scan

v.67.72 (2026-04-19)

• Mutation-section intro corrected: hotspots aren’t exclusively oncogenic-driver positions — they include recurrent tumour-suppressor mutations too (TP53 R175H, R248Q are classic hotspots but inactivate a tumour suppressor). New wording covers both cases

v.67.62 (2026-04-19)

• Subtype-hallmarks phrasing tightened: "Altered hallmark genes (N of M)" / "Hallmark genes without detected alterations: ...". Previous phrasing was ambiguous ("not detected") or over-explained

v.67.52 (2026-04-19)

• First pass at clarifying hallmark phrasing

v.67.42 (2026-04-19)

• Sex section rephrased: "annotation" / "annotated male" instead of "patient records say" (we never access patient records directly — we use the sex annotation supplied with the cell line)

v.67.32 (2026-04-19)

• Drug bullets now show a plain-English viability score ("kills most cells / kills many / modest killing / little effect") alongside the AUC number. Hover on the AUC value shows a tooltip explaining what AUC is
• Oncotree code in Classification is now a clickable link that opens Oncotree's page for the specific code, so users can see where this disease sits in the hierarchy

v.67.31 (2026-04-19)

• Wiki rewrite: plainer English, acronyms spelled out (MMR → mismatch repair, HR → homologous recombination, TSG → tumour suppressor gene, ICB → checkpoint immunotherapy, MSI → microsatellite instability). Generic phrasing throughout (no brand-name references). Each section gets a one-line intro in plain language explaining what it shows
• Authentication section now embeds the actual STR reference profile for the cell line (parsed from Cellosaurus for 1105/1186 cell lines). Table shows expected alleles at each marker — the core ANSI/ATCC 9-marker panel plus extended markers. Users can compare lab-report numbers directly against this table
• Offline pipeline: new scripts/extract_str_profiles.py parses the Cellosaurus flat-file dump and merges STR profiles into cellLineMetadata.json

v.67.30 (2026-04-19)

• Cell Line Wiki: new Drug response section (DepMap PRISM Repurposing Secondary, 25Q2). Curated panel of 89 clinically or mechanistically relevant compounds with editorial target / MOA / indication annotations. For each cell line shows: Standout sensitive compounds (z < −1 vs the panel mean), Standout resistant (z > +1), and Context-aware cross-checks that cross-reference mutations/fusions/lineage already detected in the wiki (BRAF-mut → BRAFi/MEKi, EGFR-mut → EGFR TKI, ALK fusion → ALK inhibitors, BCR-ABL → ABL TKI, HR-deficient → PARP, TP53-WT → MDM2, CDKN2A loss → CDK4/6, Lymphoid → BCR/BCL2 axis). Offline pipeline: new scripts/process_drug_response.py + web_data/drug_response.json (1.1 MB)

v.67.29 (2026-04-19)

• Cell Line Wiki: added a Subtype hallmarks section. Curated knowledge base of ~30 Oncotree subtypes (CLL, Burkitt Lymphoma, DLBCL GCB/ABC, MCL, FL, Hairy-Cell Leukaemia, Plasma Cell Myeloma, B-ALL, T-ALL, AML, Hodgkin Lymphoma, Plasmablastic Lymphoma, Melanoma, Lung Adeno/Squamous, Colon, Breast, HGSOC, PDAC, GBM, Prostate, Bladder, HCC, Ewing, Osteosarcoma, ccRCC) with expected mutations/fusions per subtype. Green box shows hallmarks observed in the cell line, blue box shows what's typical, red box flags when zero canonical hallmarks are seen (possible atypical driver or misidentified cell line)
• Cell Line Wiki: added Pathway scan in the Mutation profile. 14 cancer pathways × ~90 curated genes (p53/apoptosis, cell cycle/RB, RAS-MAPK, PI3K/AKT/mTOR, RTK, WNT, MYC/MAX, SWI-SNF, epigenetic writers, HR/FA, MMR, NOTCH, hematopoietic TFs, telomere). Each pathway shows which genes are hit (hotspot in red, damaging in amber)
• Cell Line Wiki: new flags — HR-deficient (PARP-inhibitor candidate), TP53 mutant, cell-cycle checkpoint loss (RB1 or CDKN2A)
• GE signature turned interpretive: druggable dependencies (top-essential genes that have approved/clinical inhibitors, ~60-gene curated panel, highlighted purple with 💊 marker), pathway dependencies (top-essential genes grouped by cancer pathway), tumor-suppressor-positive-on-KO flag (canonical TSGs scoring positive on knockout indicates they're still functional)
• Expression signature turned interpretive: per-lineage marker panels (e.g. Lymphoid: CD19/MS4A1/CD79A/PAX5/CD3D/CD8A…; Breast: ESR1/PGR/ERBB2/FOXA1/GATA3…), highly-expressed druggable targets flagged

v.67.28 (2026-04-19)

• New Cell line Wiki modal (button in the CLB detail panel). Sections: Classification (Oncotree subtype/code + PatientSubtypeFeatures — exposes CLL/Burkitt/DLBCL/MCL etc. under Lymphoid), Patient/tumor origin (age, race, primary/metastasis, collection site, grade, growth pattern, engineered model), Sex deep-dive with Y-loss / XIST-silencing reasoning, Mutation profile with MMR-deficiency and hypermutation flags, Fusion profile with caveat for counts >30 (hypermutated / chromothripsis artifacts), GE signature (top essential / enriched), Expression signature (top expressed, XIST, Y-mean), STR authentication explainer (ANSI/ATCC 9-marker + Promega PowerPlex 16 panel, Cellosaurus reference profiles), and outbound links to DepMap Portal, Cellosaurus, COSMIC, PubMed, ATCC
• Every section carries a Source citation line. No live scraping — outbound links only
• web_data/cellLineMetadata.json enriched via new scripts/enrich_cellline_metadata.py with 14 new per-cell-line fields from Model 25Q3 (age, ageCategory, patientRace, primaryOrMetastasis, sampleCollectionSite, growthPattern, patientTumorGrade, depmapModelType, oncotreeSubtype, oncotreeCode, patientSubtypeFeatures, rrid, engineeredModel, culturedResistanceDrug)

v.67.27 (2026-04-19)

• GE detailed view: width slider can no longer push the plot past the container edge — capped at container.offsetWidth so the stats table is always visible regardless of slider value
• GE detailed view: clicking a disease subtype row (when a tissue filter is active) now correctly shows the subtype-only box plot. Previously the filter used d.lineage === group even in subtype mode, so the subtype's box was empty and only "All cells" rendered
• The pinned "All" row in the tissue stats table is no longer clickable (it's a reference aggregate, not a selectable tissue)

v.67.26 (2026-04-19)

• GE detailed view is now square-ish by default (both width and height derive from the same base size, ~500 px). Sliders still adjust independently. Fixes V2 plot overflowing its container when the user clicks a tissue
• Statistics table: always pinned "All" aggregate row at the top (counts, mean, SD across all currently-filtered cell lines). Non-sortable — stays pinned regardless of sort order

v.67.25 (2026-04-19)

• GE plot width/height defaults now match V1 — width ratio 1.2 → 1.0, height ratio 1.1 → 1.0. Fixes plot overflowing the chart container and overlapping the table when entering detailed tissue view. Sliders still adjust plot size if you want it larger

v.67.24 (2026-04-18)

• AI Full Export from the GE modal now respects active toolbar filters (tissue, subtype, hotspot, fusion, mutation-level, oncoprint, custom cell-line list). Previously the non-mutation-inspect branch of _getAICellLines('ge') returned the full 1147-cell-line dataset. Now routes through getGETissueFilteredData(). AI mutation-analysis export also now respects the custom cell-line filter

v.67.23 (2026-04-18)

• GE toolbar: Gate A / Gate B / Compare / Clear wrapped in an inline-flex group so they stay together when the toolbar wraps (instead of splitting across rows)

v.67.22 (2026-04-18)

• GE gating now matches scatter gating and UMAP gating: tabbed results panel (Tissue / Mutations / Δ Gene Effect / Expression) in #geGateComparePanel with sortable tables. Removed the older flat-stacked-tables implementation (_geStartGate, _geClearGates, _geCompareGates) and the duplicate event listeners that were wiring the same buttons to both implementations. Callers updated to use startGEGateSelection / clearGEGates / compareGEGates

v.67.21 (2026-04-18)

• GE by-hotspot table now scrolls horizontally when the modal is narrow (prevents the p-value column from being cropped off)

v.67.20 (2026-04-18)

• Removed Gene Sets / Pathway Signatures feature entirely — 18 methods, all UI entry points (geGeneSetBtn, geGeneSetSelect, geShowIndividualGenesBtn, geIndividualControls), event listeners, and data files (hallmark_gene_sets.json, kegg_gene_sets.json, pathway_signatures.json). Closes the last divergence that was hidden-but-still-shipped from V1/CoExpress

v.67.19 (2026-04-18)

• Removed Expression vs Gene Effect panel and its UI entry points (was confusing alongside the Expression Correlates modal)
• Removed Growth Rate analysis — 6 methods, UI entry points, and the CLB list’s purple growth-rate indicator. Defensive growthRateData null-checks left in place; they’re dead code but harmless

v.67.18 (2026-04-18)

• Mutation-inspect plot: title and x-axis label pushed further from the plot area (title y 1.45 → 1.65; x-label y −0.15 → −0.20). Top/bottom margins bumped (160/55 → 180/75 px) so the labels have enough pixel budget regardless of width/height slider values

v.67.17 (2026-04-18)

• Mutation-inspect plot: title and x-axis label nudged further from the plot area (title y 1.35 → 1.45; x-label y −0.08 → −0.15) so they sit clear of the axis ticks

v.67.16 (2026-04-18)

• CLB list: cell line name now gets its full natural width (up to entry width − 50px) and does not shrink. Tissue · subtype fills the remaining space and truncates first when cramped

v.67.15 (2026-04-18)

• Cell Line Browser modal widened (1200 → min(98vw, 1600)). In the list, cell line name now gets up to 75% of the entry width and tissue/subtype shrinks first when cramped, so names are cropped only when excessively long

v.67.14 (2026-04-18)

• CLB help modal: fusion entries clarified as "fusion/translocation"

v.67.13 (2026-04-18)

• Cell Line Browser: new ? info button opens a help modal explaining filters, sort options, sex annotation vs sex (expression), and symbol meanings

v.67.12 (2026-04-18)

• Dev helper: window._geStressTest() — async matrix test over 28 width × height combinations that re-renders the GE mutation-inspect plot and asserts the three layout invariants (placeholder hidden; title fits; y-label disjoint from ticks). Logs PASS/FAIL per combo

v.67.11 (2026-04-18)

• Gene Effect mutation-inspect view: title no longer clips at narrow widths or with long gene names. Canvas measureText checks whether the title fits at 25 px bold; if not, wraps greedily at word boundaries; if still too wide, steps font size down to a minimum of 12 px. Preserves user drag positions via _geUserTitlePos

v.67.10 (2026-04-18)

• Gene Effect mutation-inspect view: y-axis label no longer overlaps tick labels. Left margin and label x-position are now computed from Canvas measureText of the widest tick label (e.g. "0/WT (n=1059)"), so the label always sits just left of the ticks regardless of width-slider value or tick-content length. User drag positions still respected via _geUserYLabelPos

v.67.9 (2026-04-18)

• Gene Effect modal: placeholder ("Type a gene name above and click Analyze…") moved to its own sibling element. Single visibility predicate (_updateGEPlaceholderVisibility) enforces the invariant that at most one of {placeholder, tissue view, hotspot view} is visible at any time. Fixes cases where the placeholder showed below a rendered plot

v.67.8 (2026-04-18)

• Removed redundant "Sex sort" control — the sex filter already serves this purpose

v.67.7 (2026-04-18)

• New Sort: Tissue + Subtype option — sorts alphabetically by tissue, then by subtype within each tissue, then by name

v.67.6 (2026-04-18)

• Cell Line Browser list: ♂ / ♀ / ? symbol next to each cell line (annotation first, falls back to expression in italic, ? when both unknown)
• Detail panel: Sex (expression) expands to "Unknown (likely Y-chromosome loss)" or "Unknown (likely XIST silencing)" based on annotation context
• New Sex sort control — groups cell lines by sex (Male → Female → Unknown); main sort still orders within each group. Independent from the main Sort dropdown

v.67.5 (2026-04-18)

• Sex-by-expression classifier now uses independent Y-marker and XIST thresholds (Y>1.0 → male; XIST>1.0 with Y low → female). Earlier score-based threshold required XIST to be extremely high, causing many XIST-silenced female cancers (esp. breast) to fall into Unknown
• Breast cancer cell lines called female by expression: 26/53 (was 15/53)
• DepMap-female cell lines correctly identified: 54% (was 34%)
• Note: XIST-silenced breast cancers (27/53) still land in Unknown — this is a biological limitation of expression-based sex inference

v.67.4 (2026-04-18)

• Cell Line Browser sex filter: annotation and expression are now independent axes — filter options are Male/Female (by annotation), Male/Female (by expression), or Unknown
• Detail panel shows both Sex (annotation) and Sex (expression) — expression call highlighted when it disagrees with annotation

v.67.3 (2026-04-18)

• Cell Line Browser: new Sex filter
• Detail panel shows Sex and Damaging Mutations (N) count
• Unified Sort by dropdown: Name / Hotspot muts / Damaging muts / Fusions / GE for gene
• Sex imputation from Y-chromosome markers + XIST expression (thresholds tuned to <1% false-positive rate)

v.67.2 (2026-04-18)

• CSV exports now include DepMap release and date in the filename (e.g. correlations_DepMap25Q3_2026-04-18.csv)
• Footer shows DepMap release version (25Q3)
• Centralized DEPMAP_VERSION constant in app.js

v.67.1 (2026-04-18)

• Fix stale folder path in data regeneration scripts after rename to correlate_v2

v.67 (2026-03-28)

• Fix Gene Sets nav button — now works without a gene selected
• Individual Genes: Back button, dedicated hotspot overlay dropdown, clear title with set description
• Dropdown shows full descriptions ("RAS/MAPK transcriptional targets") instead of code names
• Gating in GE analysis: Gate A/B buttons on By Tissue and By Hotspot boxplots
• Draw rectangles to select cell populations, Compare shows tissue enrichment + differential GE
• Default chart width 0.7, filter descriptions in all titles
• Connect lines toggle in Individual Genes view with hotspot coloring
• Fix Expression Correlates showing for gene set data
• Fix labels: titles now show "(Expr)" or "(GE dep.)" throughout
• Fix melanoma subtype filter

v.43 (2026-03-28)

• Rank-based expression scoring — batch-effect resistant, wins 6/10 vs z-score
• Scores = mean percentile rank of pathway genes within each cell line (0-1 scale)
• WNT/CTNNB1 d=1.77, RAS/KRAS d=1.24, P53/TP53 d=0.94 (all improved)

v.42 (2026-03-28)

• 15 curated pathway output signatures (transcriptional targets, not machinery)
• Focus on transcriptional TARGETS not pathway machinery — 8x better delta for KRAS (d=1.12)
• Sets: RAS/MAPK, WNT, P53, E2F, Hypoxia, Proliferation, MYC, EMT, IFN, OXPHOS, TGF-b, Notch, DDR, Apoptosis, PI3K/AKT
• Table sorting now syncs with boxplot order (By Tissue view)
• Fixed "GE" label → shows "Score" for gene sets in hotspot table headers
• Standalone "Growth Rate" and "Gene Set" buttons in GE analysis — no gene input needed
• Click Growth Rate → immediate boxplot of growth rate by tissue
• Click Gene Set → dropdown to pick Hallmark pathway → boxplot of pathway score by tissue
• Both work independently of gene effect, with full filter support

v.40 (2026-03-27)

• "By Gene Set" comparison in scatter — correlate all 50 Hallmark scores with current genes
• "Gene Sets" button in mutation analysis — differential pathway scores between Mut vs WT
• Gene set scoring: 50 MSigDB Hallmark pathway signatures as scatter axis
• New "Set" axis option — select a Hallmark set or enter custom gene list
• Scores computed as mean z-scored expression across genes in set per cell line

v.39 (2026-03-27)

• "Find Top Growth Correlators" scans all genes vs growth rate with current filters

v.38.1 (2026-03-27)

• "By Growth" comparison button in scatter inspect — correlate each gene vs growth rate
• Growth axis blanks gene input and shows "Growth Rate" placeholder
• By-tissue breakdown of growth rate correlations

v.38 (2026-03-27)

• Growth rate included in correlation analysis (toggle on/off, default on)
• Shows as "⚡ Growth Rate" in correlations table, purple diamond in network
• Click scatter from growth rate correlation auto-sets axis to Growth

v.37 (2026-03-27)

• Growth rate data: CRISPR-inferred growth rate per cell line (100% coverage)
• New "Growth" axis option in scatter inspect (X or Y axis)
• Growth rate shown in cell line browser
• Growth rate included in AI exports (gate cell lines)
• Data source: CRISPRInferredModelGrowthRate.csv (DepMap 25Q3)

v.36 (2026-03-27)

• AI-ready JSON export for all 5 data tables (mutations, gates, correlations, clusters, expression correlates)
• Each export includes full context, filters, all results, natural language summary, and AI prompt suggestion

v.35.0 (2026-03-26)

• Harmonized all 3 apps (V2, V1, CoExpress) — consistent fonts, labels, defaults
• Publication-ready scatter: 25px title, 15px subtitle, 20px axis labels, 17px ticks/legend
• GE by-tissue/hotspot: annotation-based draggable titles, larger fonts (12-14pt ticks)
• Box plots default B&W (color schemes via Aa panel)
• Default node size 20 (was 25), slider step=1 for snapping
• "No hotspot mutation" label, 2-line mutation inspect subtitle

v.34.0–v.34.8 (2026-03-26)

• Table font 12px, Δ GE (1+2v0) label, mutation table PNG/SVG export
• SVG clip fix: keep plot area clips, only remove legend/scrollbox
• Network Aa banner font size + position preserved in export
• Aa settings preserve input/correlated node color distinction
• Oncoprint Apply no longer overwrites hotspot selector
• UpSet: gene counts from all CLs, configurable N (3/5/7/10), removed bar click
• Network edge click carries filters to scatter inspect
• General tissue selector with checkboxes, subtypes, Apply (matches mutation analysis)
• Lineage filter scales with oncoprint filters

v.33.0–v.33.9 (2026-03-21–26)

• GE x-axis label includes gene name
• UpSet highlights bars matching oncoprint selection (green/gray)
• Mutation inspect subtitle: all filters with WT/Mut labels
• GE modal: oncoprint filters actually filter data + show tags + count
• CLB→GE: skip hotspot when oncoprint active, carry CLB tissue filters
• Network filter text includes all oncoprint filters
• CLB active filter summary: "Showing N of M: filters..."
• CLB Reset clears all filters including oncoprint
• All exports include oncoprint filters in metadata

v.32.2 (2026-03-21)

• CLB active filter summary bar: "Showing 137 of 1186: Skin · Melanoma, TP53 Mut, KRAS WT"
• CLB filter counts cascade with oncoprint selections
• CLB Reset clears all filters including oncoprint selections
• CLB tissue/subtype filters carry through to GE modal
• Min n per tissue control in GE modal (default 1 = show all)

v.31.0–v.31.9 (2026-03-21)

• Tissue breakdown: expandable sub-tissues with checkboxes for selection
• Oncoprint drag fix, UpSet close doesn't close oncoprint
• Network/CSV/SVG exports include all oncoprint mutation filters
• Mutation inspect chart: x-axis always includes 0, wider left margin
• CLB toolbar: Hotspot filter, Fusion filter labels, Sort by GE for gene
• Expr vs GE button green (was purple)
• Mutation filter labels: "Mutation filter: NRAS WT, HRAS WT"

v.29.0–v.30.9 (2026-03-20–21)

• Oncoprint: two boxes per gene (include mutated / exclude mutated)
• Oncoprint multi-gene filters apply to all analyses and inspect views
• Oncoprint: SVG, PNG, CSV export
• UpSet intersection plot with counts/percent/names toggles and export
• Oncoprint in cell line browser with live filtering
• Oncoprint filter label replaces param hotspot controls when active
• All selectors respect excludedTissues from tissue breakdown
• Compact Min Cell Lines + P-value on same row
• Green border on mutation analysis sub-type selector

v.28.0–v.28.7 (2026-03-20)

• CLB mutation/translocation filters: searchable (input + datalist), sorted by count
• CLB gene click carries all active filters (tissue, hotspot, fusion) to GE modal
• Enrichr filter dropdown: Top 10/20/50/100, WT vs 1+2, WT vs 2, 2 vs 1, Δ GE, Visible
• GE modal shows all tissues when mutation/fusion filter active (min group size 1)
• Sort ALL dropdowns by count descending (tissues, subtypes, mutations)
• Oncoprint button in mutation analysis (first version)

v.27.6 (2026-03-20)

• Cascading filters: all selectors reflect active lineage/sublineage, sorted by count descending
• Hotspot/damaging/translocation selectors hide genes with 0 mutations in filtered set

v.27.5 (2026-03-20)

• Cell line browser gene click applies active parameter filters (lineage, hotspot, fusion)
• Mutation analysis: new Δ GE (2v1) and p(2v1) columns — 2-mutation vs 1-mutation comparison

v.27.4 (2026-03-20)

• Warn when mutation analysis skips genes due to min cell lines threshold

v.27.3 (2026-03-20)

• Lower min cell lines by 50%: default 25 (was 50), mutation mode 5 (was 10)

v.27.2 (2026-03-20)

• Bold and Italic toggle buttons in Aa text settings for title, subtitle, axis labels

v.27.1 (2026-03-20)

• Tissues breakdown button for damaging mutations (matching hotspot/translocation)
• Subtle line separator before Δ GE columns in mutation table

v.26.9 (2026-03-19)

• Synonym mode: test genes loads mouse orthologs/aliases (Trp53, p21, PD-1, etc.)
• Hide Exclude Tissues and Find Synonyms button in synonym mode

v.26.8 (2026-03-19)

• Full visual settings stored in export metadata for network, mutation inspect, gene effect
• Network restore applies fontSize, nodeSize, edgeWidth, color settings

v.26.7 (2026-03-19)

• Fix mutation inspect restore: populate hotspot selector before setting value
• Store lineage filter in mutation inspect metadata

v.26.6 (2026-03-19)

• Remove Outline text feature (opentype.js) — Arial font and tspan flattening solve the problem

v.26.5 (2026-03-18)

• Deep SVG cleanup: strip all Plotly class/data- attributes, clipPaths, empty groups, cursor/white-space CSS
• Exported SVGs are now clean, minimal, and work correctly in both Illustrator and Inkscape
• Removed "Open" from nav bar (remains in sidebar and inspect footer)
• Default graph font: Arial; tspan flattening for editable titles
• All export types restore their specific view when opened

v.25.9 (2026-03-18)

• Fix SVG dimensions for Inkscape — width/height now match viewBox, removed empty style and class from root element

v.25.8 (2026-03-18)

• Reset all filters before restoring from SVG/PNG — prevents filter bleed-through
• Mutation inspect exports now open GE modal in hotspot view (not tissue)
• PCA/UMAP exports show info message directing to Dimensionality Reduction panel

v.25.7 (2026-03-18)

• New "Outline text" checkbox for SVG exports — converts text to paths via opentype.js for Illustrator/Inkscape compatibility
• Supports Open Sans, Arial, Helvetica, Georgia, Times New Roman, Roboto Mono, Courier New

v.25.6 (2026-03-18)

• Open button now restores network exports (gene list, mode, cutoff → re-runs analysis)
• Gene effect exports open GE modal, scatter-like exports restore inspect view

v.25.5 (2026-03-18)

• Open button handles all export types — detects graph type from embedded metadata
• Accepts both SVG and PNG files from any graph type

v.25.4 (2026-03-17)

• Expr vs GE: replaced Expr/GE ratio with dependency Score (Expr × −GE) — handles zero expression, default sort by score descending
• Open button: moved next to Cell Lines, accepts both SVG and PNG exports
• All exports (SVG/PNG) now embed Correlate metadata (app, version, graph type, date, genes)
• PNG exports embed metadata in tEXt chunks for restore support
• Tissue/expression/correlation analysis exports upgraded to SVG-first pipeline with Illustrator sanitization

v.25.3 (2026-03-17)

• Expr vs GE: moved higher in cell line panel, Aa text settings, draggable titles, sortable table columns
• UMAP/PCA: annotation-based draggable title and labels, title no longer overlaps modebar
• All annotation-based plots now center title over graph area

v.25.1 (2026-03-17)

• New default font sizes: Title 25, Subtitle 15, Labels 20, Ticks 17, Legend 17, Markers 10
• Number inputs: removed native up/down arrows, only custom −/+ buttons

v.25.0 (2026-03-17)

• Scatter text settings now persist across GE↔Expression switches
• Subtitle line spacing tighter by default, scales with subtitle font size
• Y-axis label nudged left and left margin wider to prevent cropping
• Dot color picker added next to Zero Lines

v.24.7 (2026-03-17)

• Network filter banner is now draggable — click and drag to reposition

v.24.6 (2026-03-17)

• CLB export: clearer button names (GE Matrix, Cell Line Report, Full Data Export) with tooltips
• Full Data Export: auto-loads expression data if not yet available
• Expression data preloads in background 2s after app startup

v.24.5 (2026-03-17)

• Network: font size now absolute — stays exactly as set regardless of rebuild/toggle
• Network Aa: full text settings (font family, legend size/color, banner size/color/text, node colors)
• Network: black node borders (toggleable), font/color persist across rebuild
• Expression loading: progress overlay in scatter (metadata → download → decompress → map)
• Aa panel: auto-closes on tab switch and inspect modal close

v.24.0 (2026-03-17)

• Scatter: gene input boxes wider, text settings preserved on gene switch
• Subtitle editing: multiline textarea, rows preserved on apply
• Network: absolute font size (no zoom scaling), Aa settings panel, node border toggle
• CLB: comprehensive export (GE/Expr/Hotspot/Damaging/Translocation matrix)
• CLB: Expression vs Gene Effect scatter + table for selected cell line
• SVG export: Illustrator-compatible (inline styles, rgb→hex, XML declaration)
• SVG export: embedded state metadata for re-opening plots (Open SVG button)

dev-23.6 (2026-03-17)

• Mutation inspect: zero line toggle (on by default)
• Mutation table: 2-only columns hidden by default with toggle to show

dev-23.5 (2026-03-17)

• Aa panel: font-family selector (Open Sans, Arial, Helvetica, Georgia, Times, Roboto Mono, Courier)
• Aa buttons now use app font (Open Sans) instead of serif
• Top Genes popout: opens beside plot (not overlapping), draggable via header
• Plot W/H controls: set width and height from input fields
• Export matches screen: PNG/SVG use actual layout dimensions
• Fix missing data: impute with gene mean instead of 0, skip genes with >30% missing
• Top Genes: floating popout, click gene to color plot, data type matches analysis
• PCA loadings: draggable biplot inset box instead of cluttered overlay arrows
• Layout: split controls into two rows to prevent wrapping with long tissue names
• Export: PNG and SVG buttons, Aa text settings for UMAP/PCA plots
• Squarer default graph (580x480), condensed layout

dev-23 (2026-03-16)

• UMAP/PCA overhaul: clean layout, Run Analysis button, complete Reset button
• PCA: compute 5 components, axis selector (PC1 vs PC3, etc.), loadings arrows on plot
• UMAP: 3D computation, dimension selector (UMAP 1 vs UMAP 3, etc.)
• Fix lingering gates/splits/graphs on re-run

dev-22 (2026-03-16)

• Aa: box color scheme selector — Essential (red/gray/green), B&W, Blue gradient, Red-Blue diverging, Viridis, Uniform steelblue

dev-21.1 (2026-03-16)

• Mutation inspect: title, X/Y axis labels now draggable annotations (same as scatter)
• Dragged positions persist across re-renders within the same session

dev-21 (2026-03-16)

• Draggable axis labels: X and Y axis labels are now annotations — drag with mouse or nudge with arrow keys
• Subtitle size control in Aa panel
• Subtitle format: "mean (...) median (...)" with spaces

dev-20 (2026-03-16)

• Aa Text Settings: editable title/subtitle/axis text, toggle visibility, Scale All +/−
• Fix: title/legend font size changes now work (scatter uses annotations, not layout.title)
• Annotations: click to select, arrow keys to nudge (Shift for larger steps), drag directly on plot
• Screenshot PNG: high-res (3x) page capture incl. nav-bar
• Fix: network edge thickness legend + blue/red colors in all states

dev-19.2 (2026-03-16)

• Full Page SVG: exports parameters, input genes, and correlation table as clean vector SVG for publication

dev-19 (2026-03-16)

• UMAP gates: lasso/box selection (not only rectangles), colored overlay dots with A/B labels
• UMAP hover: shows hotspot mutation count per cell line
• UMAP gate Enrichr: Enrichr buttons in mutations, diff GE, and expression tabs
• Aa text settings panel: draggable floating panel to adjust title, axis, tick, legend, marker sizes (scatter + gene effect plots)

dev-18 (2026-03-16)

• PCA as alternative dimension reduction method (with variance explained %, top loadings)
• Gene color gradient overlay: color UMAP/PCA dots by any gene's GE or expression value (blue-white-red)
• Top genes panel: shows genes most correlated with UMAP/PCA coordinates, clickable to apply color
• UMAP from Gate A now supports PCA method and stores gene names for top genes analysis

dev-17.3 (2026-03-16)

• Clickable genes in UMAP gate comparison: mutation bar chart, GE box plot, expression box plot
• Fix CLB modal scrolling: UMAP section and gate comparison now scrollable
• Marker size slider (range 2–14, live update)

dev-17 (2026-03-16)

• UMAP gate comparison: draw Gate A/B rectangles, compare tissue, mutations, Δ GE, expression
• UMAP from Gate A: full UMAP recalculation using only Gate A cell lines (min 15)
• Gate tables with sortable columns, gene search, p-value filter
• Reset button clears all UMAP state including gates

dev-16 (2026-03-14)

• Fix UMAP constructor error (umap-js CDN module export compatibility)
• UMAP: more condensed clusters (higher minDist/spread), bigger filled markers
• UMAP split by mutation: 3 panels (All / WT / Mutated) with shared axes
• UMAP split: searchable input+datalist for hotspot or translocation genes, sorted by count
• UMAP: subtype filter (appears when tissue selected)
• Default drag mode set to lasso for easier selection

dev-15 (2026-03-14)

• UMAP visualization in Cell Line Browser (GE, Expression, or both; top 1000 variable genes)
• UMAP: color by tissue/subtype, split by mutation, lasso select + copy cell lines
• Custom cell line filter (paste IDs/names) in scatter, mutation inspect, and CLB
• GE vs Expression axis selectors in correlation scatter (plot GE on one axis, Expression on other)
• Logo centered with "extended version" subtitle

dev-14 (2026-03-14)

• Export preview modal with adjustable graph settings (dimensions, fonts, marker size, scale, background)
• Color-by-tissue legend no longer cropped (max-height: 40vh)

dev-13

• Axis range: +/- buttons, "Reg. line" and "Zero lines" toggles on same row
• Gate A/B labels shown on scatter plot rectangles
• Gene plot margins standardized for export alignment
• "Diff GE" renamed to "Δ Gene Effect"

dev-12

• Export: separate PNG/SVG buttons, descriptive filenames
• Mutation bar chart: 4 stacked traces (WT/Mut for both gates)
• GE inspect gene plots square (300x300)
• P-value filter dropdown on gate comparison tabs
• Enrichr prompts for p-value threshold

dev-11

• Fix axis range "-" input, correlation line toggle label
• Gene search input in gate comparison tables
• Enrichr A/B buttons for mutations (relaxed filter)
• Hotspot filter counts update with tissue filter
• DepMap source files listed in info box
• GE inspect gates: draggable rectangles, both visible
• Enrichr buttons on GE inspect gate tabs

dev-10

• Draggable rectangle gates (replace select-based)
• Full gate comparison in mutation analysis inspect

dev-9

• Both hotspot AND damaging mutations in gate comparison (type badges)
• Expression comparison tab (on-demand loading)
• Clickable gene names → box/bar plots comparing Gate A vs B
• Enrichr fix (wrong element ID) + buttons on all tabs
• Axis range: remove inputmode=decimal
• Correlation line toggle checkbox

dev-8

• Gate A/B colored overlay markers on scatter (blue/red)
• Expression correlates available in regular scatter inspect

dev-7

• Gate comparison: sortable tables, Enrichr, subtissue, scroll fix
• Fix bone cancer bug (gates match filtered data)
• Gate A/B columns in scatter CSV download

dev-6 and earlier

• Damaging mutations integration (8926 genes)
• Gate comparison feature (tissue, mutations, differential GE)
• Scatter performance fix (limit dropdown options)

×

Curated collections — inclusion criteria

×

Working…

Preparing…

Cancel

Min mutated: Export CSV Close

Cell Line Browser — Help

×

Filters — stack with AND. Click Reset to clear.

• Tissue / Subtype — DepMap OncotreeLineage and primary disease subtype.
• Sex — annotation and expression are two independent axes (see below).
• Hotspot filter — only show cell lines mutated in this gene at a hotspot position.
• Fusion filter — the dropdown lists clinically relevant fusion pairs first (★ BCR-ABL1, ★ EWSR1-FLI1, …) followed by the gene-based long tail. Selecting a pair filters to the curated set; selecting a gene returns any line with any fusion involving that gene.
• Oncoprint — mutation heatmap across filtered cell lines.

Hotspot variants — when DepMap's inferred-subtype pipeline calls a specific codon (KRAS p.G12D, BRAF p.V600E, EGFR p.L858R, JAK2 p.V617F, EGFR exon-19 del), the variant is appended in green next to the gene name in the Hotspot Mutations list. Generic group calls like KRAS p.G12 (any G12 not specifically D or C) are also shown when no specific call is available. Source: OmicsInferredMolecularSubtypes.csv.

Functional loss — integrated tumor-suppressor inactivation calls (RB1, TP53, PTEN, NF1, CDKN2A, VHL, MTAP, APC) shown as red-bordered chips in the detail panel. A gene is "lost" if any of: WGS-based relative CN < 0.3, likely-LoF mutation with AF > 0.5, or expression < 0.1 log-TPM. This catches deletion-driven losses that the damaging-mutation matrix alone misses — e.g. CDKN2A homozygous deletion in 445 lines (~37% of the panel), most of which look mutation-clean. Source: OmicsInferredMolecularSubtypes.csv.

MSI badge — small red MSI chip next to the cell-line ID when MSIsensor2 score ≥ 20 (DepMap's threshold for microsatellite instability).

Genome signatures (small row near the top of the detail panel) — Ploidy (PureCN avg chromosome dosage), WGD (whole-genome doubling, shown in red when positive — ~58% of the panel), Aneuploidy (Ben-David 2021 score, 0–39), CIN (chromosomal instability). These shape interpretation when comparing gene effect across cell lines — WGD-positive lines are systematically different. Source: OmicsGlobalSignatures.csv.

Each new section in the detail panel has an inline ? with a hover tooltip recapping the key explanation, so you don't need to come back to this modal to remember what a tier or column means.

Clinically relevant fusions — shown as green/amber/red chips in the cell line detail panel.

• Curated list of ~50 well-validated driver fusions (BCR-ABL1, EWSR1-FLI1, EML4-ALK, PML-RARA, TMPRSS2-ERG, …) sourced from COSMIC fusion gene census.
• Each call is validated per cell line on three orthogonal signals: lineage match, partner expression z-score vs lineage-matched negatives, and partner dependency (CRISPR gene effect) z-score. Only the signals the curator marked as informative for that fusion are scored — e.g. BCR-ABL1 doesn't elevate ABL1 mRNA, so expression isn't checked and the call rests on the strong ABL1 dependency.
• high — both informative signals support the call (or one strong + the other directionally supportive).
• med — partial orthogonal support, or curated + lineage match without orthogonal validation.
• low — neither signal supports and lineage doesn't match the canonical disease — suspect, often artifact.
• ⚠ atypical — the cell line's tissue isn't the canonical context for this fusion (e.g. EML4-ALK in a Bowel line). Lineage is a modifier, not a gate: when orthogonal evidence is strong the call is kept and flagged, not hidden — could be a misclassified line or a genuine rare event worth investigating.
• Source: Arriba-annotated fusion calls from OmicsFusionFilteredSupplementary.csv, filtered to confidence=high with read-through transcripts dropped.

Sex — two independent axes:

• Sex (annotation) — from DepMap's Model table (patient-reported). Values: Male, Female, Unknown.
• Sex (expression) — computed from Y-chromosome markers (RPS4Y1, DDX3Y, EIF1AY, KDM5D, UTY, USP9Y) and XIST expression. Rules: Y-mean > 1.0 log-TPM → male; XIST > 1.0 with Y low → female; otherwise unknown. Validated on DepMap-labeled cells: ~3% false-positive on males, ~0.14% on females.

When expression is “Unknown”:

• Annotation Male → “Unknown (likely Y-chromosome loss)” — Y loss is common in cancer, esp. older male patients and advanced tumors.
• Annotation Female → “Unknown (likely XIST silencing)” — XIST silencing is documented in breast and other cancers.

Sex symbols in the list

• ♂ blue, upright: Male by annotation.
• ♀ pink, upright: Female by annotation.
• ♂ blue, italic: Male by expression only (annotation unknown).
• ♀ pink, italic: Female by expression only (annotation unknown).
• ? gray: Unknown in both axes.

Sort options

• Name — alphabetical.
• Tissue + Subtype — alphabetical by tissue, then subtype within each tissue.
• Hotspot muts — number of hotspot-mutated genes per cell line (descending by default).
• Damaging muts — number of genes with damaging (deleterious / loss-of-function) somatic mutations per cell line (descending). Source: DepMap OmicsSomaticMutationsMatrixDamaging.
• Fusions — number of distinct fusion/translocation partner genes per cell line (descending).
• GE for gene(s) — enter one or more genes; cells sorted by CRISPR gene effect (Chronos score). For multiple genes the mean across the panel is used — rank cells by combined essentiality.
• Expression of gene(s) — enter one or more genes; cells sorted by log₂(TPM+1) expression (descending by default, so highest-expressing cells surface first). Multiple genes are averaged, so a panel like CDKN2A, RB1 or ASCL1, NEUROD1, CHGA, SYP finds cells with the strongest combined signature.

Gene-sort input: comma, semicolon, newline or whitespace can separate multiple symbols.

Data source: DepMap 25Q3. All filters and sorts use the data currently loaded in the browser; changing tissue / subtype filters narrows the visible set that subsequent sorts operate on.

Collections: ?

Custom CLs: Apply Clear

Export Preview

×

Download PNG Download SVG

Export .json.gz for LLM analysis ×

Question for the LLM (optional)