Fig. 4

A The log-likelihood gain (log₁₀ scale) after applying the inferred modulatory processes for each type of cancer. The red line marks the mean. B The comparison of additive residues of three models for CRC, melanoma and liver cancers. Baseline refers to the Poisson-resampled model, additive model refers to the NMF model [4] and extended model—the additive model together with the modulatory process. The residues are normalized and colored by total sample-wise mutation burdens. C The impact, i.e. the mean of the modulatory process effect in the three selected types of cancer. D The total log-likelihood gain, normalized by the total mutation burdens. E Hierarchically clustered modulatory signatures of each cancer type. The mean modulatory signature profile across all cancer types is displayed separately at the bottom