Purity correction

To assess the purity of the samples, we could rely on the pathologists’ estimate, where samples with a purity below 40% were discarded. To compensate for potential bias in the pathologist’s estimate, we also applied two different tools to estimate tumor purity. TPES [21] estimates tumor content based on the set of variants called by at least two variant callers in each sample and was complemented with the FACETS tool [22], which looks at common human SNP sites and measures purity based on an Expectation Maximization of the ASCAT formulas [23]. For TPES, the variants obtained through our variant calling strategy were used as inputs (VCFs), together with the seg files obtained from the GATK CNV pipeline. Because those filtered VCFs contain less, but high-confident variants, we had to lower the minSNVs parameter to 3 [23]. FACETS was run using default parameters [22]. For patient UZ001, matching RNA material was available such that we could assess Tumor Infiltrating Leukocytes (specifically we tested for infiltration of B cells, CD4 T cells, CD8 T cells, Monocytes, Neutrophils and NK cells) through transcriptome deconvolution, done using EPIC [24].