Sup. Fig. 1

Sup. Fig. 1 Simulation analyses show accurate dissection of bulk tumor RNA-Seq data by DisHet. (a) Scatterplot shows high correlation between predicted component proportions (X axis) and simulated component proportions (Y axis). N=35. (b) Scatterplot shows high correlation between predicted (X axis) and simulated (Y axis) average expression of each gene. (c) Distribution of the correlations of the predicted and simulated individual expression levels of each gene across all patients.

Sup. Fig. 2

Sup. Fig. 2 Validating immune/stroma component expression dissected by DisHet. (a) Intra-signature correlations of genes from each of the 16 Winslow signatures. Intra-signature correlations are shown in red with pair-wise correlations of all genes as control in purple for the tumor- and immune/stroma- specific components, respectively. (b) Principal Component Analysis plot based on the immune/stroma expression of 9 tumor samples from 4 unique patients, dissected by DisHet, shows strong similarity of immune/stroma expression in different sampling regions of the same patient.

Sup. Fig. 3

Sup. Fig. 3 Comparison of dissection performance of DeMix and DisHet. (a) Correlation of

ssGSEA scores of the ESTIMATE signature in the bulk tumor and predicted tumor stromal

proportions by DeMix. (b) Correlation of predicted tumor proportions by DeMix and

proportions assessed by pathological reviews. (c) Run time comparison for DisHet and

DeMix for simulation data of different sizes. The first panel shows the runtime comparison

on 5 testing datasets containing 1000 randomly selected genes from 35 patients. The

second panel shows run time comparison on datasets containing 1500 genes from 35

patients. The unit of the y-axis is CPU seconds.

Sup. Fig. 4

Sup. Fig. 4 Principal Component Analysis (PCA) shows the

clustering of TCGA pan-RCC patients tumors by the

tumor-specific genes and eTME-specific genes, respectively.

Tumor-specific genes are defined as genes whose

expression in the tumor is at least 3 times higher than both

the normal tissue and immune/stroma components.

Red: ccRCC, blue: pRCC, and yellow: chRCC.

CD4 CD8

Six2-Cre;VhlF/F;Pbrm1F/F

Six2-Cre;VhlF/F;Bap1F/+

100 μm 100 μm

100 μm 100 μm

Sup. Fig. 5

Sup. Fig. 5 CD4 and CD8 immunohistochemistry staining in the Six2-Cre;VhlF/F;Pbrm1F/F and

Six2-Cre;VhlF/F;Bap1F/+ Genetically Engineered Mouse Models (GEMMs). N=3 in each cohort.