Fig. 1

A Observed and predicted mutation counts (first and second panels), additive residual \(X - {\widetilde{X}}\) (third panel) and multiplicative residual \(X/{\widetilde{X}}\) (fourth panel), where X is observed and \({\widetilde{X}}\) is NMF model fitted mutation counts of one example sample (colorectal adenocarcinoma (ColoRect-AdenoCA) PCAWG sample SP16934). Bars are colored according to their base SBS type. B Multiplicative and additive residual correlation clustering. The upper triangle represents the multiplicative residual correlations, while the lower triangle represents the additive residual (the lower triangle correlation values are ordered according to the ordering of the upper triangle). Correlation scale is limited to the range of [0, 1]. The color-bar on the right show the cancer type (refer to Additional file 1: Fig. S1C for legend). C Mean residual profiles of the zoomed in block consisting mostly of Liver-HCC samples (see text). D A scatter plot of the mutation counts vs. NMF model predicted counts for this Liver-HCC cluster shows a systematic overestimation of C>G and underestimation of T>A mutations. These systematic biases are also evident from the mean multiplicative residual profile of panel C