Wild-type by absence: cancer cell line labels tested against curated records, repeat sequencing and functional readouts
Year: 2026
Type: preprint
Abstract: Cancer cell lines can be chosen as wild-type for a gene because a mutation file reports no variant in it. We asked how often other data disagree with that inference and which kinds of disagreement go with low functional readouts. For 2,232 cancer lines in DepMap Public 26Q3, we compared the absence of a protein-changing variant in the default sequencing entry with copy number, other sequencing entries, curated Cellosaurus records, Sanger Cell Model Passports, CCLE 2019 and expression. Of 842 lines with no reported TP53 variant, 85 (10.1%) had contrary evidence. Four of 12 CRISPR readouts from the literature met a separation bar written before this analysis read the screens, and not registered; other comparisons were exploratory. Lines with contrary evidence fell below the TP53 cut in 61% of cases (36/59), against 13% of lines without it (47/362; P = 1.7 × 10⁻¹⁴), or 17% under cross-validation. The difference held within lineages and in p53 target expression (57/81 versus 113/654) and Project Score screens (12/12 versus 7/56). RB1 was weaker (55% versus 16% centred, 38% versus 10% raw); in Project Score and protein, lines with low copy number or expression fell below the cut (4/4, 4/6), while lines with sequence evidence alone were too few to show or exclude an effect (3/10, 3/19). STAG2 rested on few lines, and the APC readout was not specific to APC. Pooled, 65% of lines with low copy number or expression and 43% with truncating changes or deletions fell below the cut. Missense and in-frame calls seen only elsewhere mostly did not in TP53 (1/15), with RB1 higher (7/21, five of those with one recurrent in-frame deletion). We also describe three ways this study's own checks misread these data, on chromosome X, flagged synonymous TP53 variants and lineage centring. These are associations in one snapshot of public data: we sequenced no line and knocked out no gene, and a low score does not establish a line's genotype.Claude (Anthropic), deployed as the AI Operator of O-mega Enterprise, Inc., designed the study, wrote and ran the analysis code, made the figures and wrote the text. Yuma Heymans, chief executive of O-mega Enterprise, Inc., commissioned the work, set its brief, decided to publish it and stands behind it as its named author, and did not design, run, write or review it. It has not been peer reviewed. Competing interests and funding are stated in the paper.Analysis code and summary statistics: https://doi.org/10.57967/hf/11474. The same code, source file checksums and the DepMap-derived line-level tables: https://cybergenic.im/data/cell-line-labels/2026-11-06/README.txt. A shorter account: https://cybergenic.im/blog/cell-line-wild-type-labels-depmap.The paper is shared under CC BY 4.0. The line-level tables are not part of this record and stay under DepMap's terms of use.
Source: Figshare
Author: Yuma Heymans
Language: English
Subfield: Cancer Research
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7225728145