Fig. 4From: Mutation-Attention (MuAt): deep representation learning of somatic mutations for tumour typing and subtypingUMAP projection of MuAt tumour-level features in PCAWG data. MuAt recognized tumour subtypes in a prostate cancers with a subgroup defined by SPOP driver mutations and increased somatic structural variant burden, b medulloblastomas, c microsatellite-stable colorectal cancers, microsatellite-unstable cancers (MSI) and polymerase \(\epsilon\) deficient cancers (POLE), d skin melanomas subtypes with CCND1 amplifications, e chronic lymphocytic leukemias associated with somatic hypermutability and f pancreatic neuroendocrine tumours with germline MUTYH mutations. Specific example tumours are indicated by numbers, with SNV, indel and SV types, and SNV/indel/SV proportions shown. Tumours coloured by a specific single-base or doublet-base signature exposure indicated with SBS or DBS, respectivelyBack to article page