Representation learning of somatic missense mutation landscapes across cancer types

MMRL learns self-supervised embeddings of somatic missense mutation profiles from ~50,000 tumor samples, yielding higher cross-validated prognostic discrimination than TMB across pan-cancer cohorts.

• Why it matters: Existing cancer mutation representations often reduce missense variation to burden, signatures, or drivers, obscuring mutation-specific context and higher-order dependencies linked to outcomes.
• What they did: The MMRL framework integrates gene identity, trinucleotide context, AlphaMissense pathogenicity, amino acid substitution, and protein positional information, then trains self-supervised learning on pan-cancer missense profiles.
• The result: In three pan-cancer cohorts, MMRL embeddings remain significantly associated with overall survival after adjustment for cancer type and TMB (and age where available), including in immune checkpoint inhibitor–treated patients.

AI summary of the abstract · may contain errors · not medical advice · read the paper.