Cancer gene co-mutation explorer
Which driver genes are mutated in the same tumours more often than chance allows, and which avoid each other? Every gene pair the Cybergenic scan tested, in each of 50 tumour cohorts, significant or not, read straight from the open dataset.
Try
Loading the dataset (5.1 MB) from Hugging Face…
Read the numbers with care
- Mutations only. A driver mutation here is a nonsense, frameshift, nonstop, splice-site or splice-region mutation, an in-frame insertion or deletion, or a missense mutation at a position recurrently hit in that study. Copy-number changes and fusions are not counted, so an amplified ERBB2 or a deleted CDKN2A is not in these tests.
- Chance holds each tumour's mutation count fixed. The test of record is a burden-conditional exact test (a conditional variant of DISCOVER, Canisius, Martens and Wessels 2016): a heavily mutated tumour is expected to carry more pairs, so the expected count here differs from the naive one, the count expected with no burden effect ((tumours with gene A × tumours with gene B) / tumours), which is shown beside it.
- Established means q at or below 0.05 with a consistent direction and at least a 2-fold odds ratio in that direction, or kept by every gate of the scan. The direction rule of 7 October 2026 is applied to every row here (see the dataset card's correction). Kept means the scan's own rule: q below 0.1, at least 2-fold, a consistent direction, no reversal within strata, and enough tumours.
- q values are corrected within one cohort. Comparing a pair across cohorts, or scanning this whole table for hits, is a new multiple-testing problem of your own.
- MSK-IMPACT 50K and MSK-CHORD share most of their patients (about 78%). A pair seen in both has been seen once; "independent sources" counts TCGA, MSK, OrigiMed and METABRIC.
- Most tests are small and discrete. About three quarters of the p values are exactly 1, and where a cell of the 2×2 is empty the odds ratio rests on a 0.5 correction and can mislead, even in sign, so it is shown as not estimable.
- A row describes how mutations are distributed across tumour samples. It is not evidence of a mechanism and not clinical guidance.