Additional file 1 of A reference profile-free deconvolution method to infer cancer cell-intrinsic subtypes and tumor-type-specific stromal profiles

Additional file 1 : Figure S1. A schematic illustration of DeClust algorithm. Figure S2 Flowchart of the study. Figure S3. Simulation results when gene expression was simulated under negative binomial distribution. Details can be found in legend of Fig. 1. Figure S4. Simulation results when gene expression was simulated under negative binomial distribution. Details can be found in legend of Fig. 1. Figure S5. Comparing cell fraction estimations by different gene expression deconvolution methods with the ones based on MethyCIBERSORT (treated as “ground truth”) for immune compartment (AB) and stromal compartment (CD) using Spearman’s correlation coefficients (AC) or Median Absolute Deviation (BD). The p-values above are the difference between DeClust and other methods according the two-sided paired t-test. Figure S6. Kaplan-Meier curves of patients with high/low stromal compartment fractions as defined by DeClust (left) and ESTIMATE (right) in TCGA dataset (top) and non-TCGA dataset (bottom) for KIRC (A) and BLCA (B). Figure S7. Pathway analysis of stromal proles across 13 TCGA datasets using Canonical pathways (left) or hallmark_cancer pathways (right) from MsigDB. The color indicates the significance of the up/down-regulation of that pathway in that stromal prole as compared to other stromal proles (−log10 (p-value of Wilcoxon rank-sum test), see Method). Red color represents up-regulation and blue color represents down-regulation. Figure S8. Number of genes with subtype-specific methylation before (A) and after (B) methylation correlated with immune or stromal cell frequencies were removed. Figure S9. Overlap between CRIS subtypes and subtypes defined by different methods. Figure S10. Plots of the correlation between tumor purity (Consensus Purity Estimation*) and subtypes defined by different methods, grouped by methods (A) or by cancer types (B). Figure S11. Kaplan-Meier curves of DeClust subtypes (left) and TCGA subtypes (right) for TCGA dataset (top) and non-TCGA dataset (bottom) for KIRP (A), KIRC (B) and LUAD (C). Figure S12. (A) Spearman’s correlation coefficient between immune and stromal proles estimated by DeClust (y-axis) and reference proles used by EPIC (x-axis). (B) Exemplary scatter plot of reference expression prole versus immune and stromal prole estimated by Declust using TCGA BLCA dataset.(C) Correlation between different reference proles(x-axis) and bulk tissue expression proles (gray), immune proles (red) and stromal proles (blue) estimated by DeClust using TCGA BLCA dataset. (D) Cell type-specific markers identified in EPIC reference proles and their expression in immune and stromal proles estimated by DeClust. Figure S13. t-SNE plot of scRNAseq data for pRCC samples. Cell-type specific markers used to annotate the cell clusters were shown below. Figure S14. Correlation between mean expression prole of each epithelial cell cluster and subtype-specific cancer proles estimated by DeClust or by TCGA for ccRCC scRNAseq data (A) and pRCC scRNAseq data(B). Error bars indicated 95% confidence intervals. Figure S15. Comparison of inferred cancer proles and scRNAseq data. Figure S16. Violin plot of estimated immune (left) and stromal (fraction) fraction in each subtype of BLCA. Figure S17. Proportions of different immune subtypes within each subtypes defined by DeClust. Figure S18. Comparison of immune subtypes and subtypes defined by DeClust in association with overall survival. Plots are grouped by methods (A) and by cancer types(B). Figure S19. (A) The association of overall survival and immune subtypes within each cancer cell-intrinsic subtype defined by Declust.(B) Kaplan-Meier curves of immune subtypes within Atypical_1 subtype of HNSC. Figure S20. (A) The association of overall survival and immune/stromal cell fraction within each subtype defined by Declust. The color in the heatmap represents the –log10(p-value) of the association (log rank test). Red means the higher cell fraction corresponds to worse survival, and blue means the opposite. Only cancer subtypes and cell fractions with at least one significant association (p