(a) Corrected TMB was replotted with tumor purity for TCGA samples to visualize whether our approach generated overcorrected TMB values on the lower end of tumor purity (N=3,788 tumors, BLCA N=405 tumors, BRCA N=778 tumors, COAD N=287 tumors, HNSCC N=504 tumors, KIRC N=367 tumors, LUAD N=508 tumors, LUSC N=477 tumors, SKCM N=462 tumors). We did not observe such phenomenon for all tumor types evaluated, suggesting that our correction factors did not generate high TMB-low tumor purity outliers. Additionally, we found that after using our TMB correction approach, there was now no longer a positive correlation between cTMB and tumor purity (BLCA Pearson R 0.023 p=0.64, BRCA R=-0.013 p=0.71, COAD R=0.019 p=0.0.77, HNSCC R=0.031 p=0.48, KIRC R=0.015 p=0.78, LUAD R=0.071 p=0.11, LUSC R=0.069 p=0.13, SKCM R=0.048 p=0.31). Blue line indicates a linear fit. The Pearson correlation coefficient (R) was used to assess correlations between continuous variables and p values are based on two-sided testing. (b) Effect of TMB correction for tumor purity in cohort 1. Distribution of observed (black circles) and corrected TMB for patients in cohort 1 (N=89 patients) are shown for each tumor purity tier. Corrected TMB values are denoted by purple circles for tumor purity 0.1-0.25 and green circles for tumor purity >0.25, error bars represent 95% confidence intervals. cTMB values are capped at 1000. The second tertile of cTMB was used as a cut-off point to classify tumors as high vs low TMB categories in order to avoid bias related to multiple cutoff point selection. After correcting for tumor purity in cohort 1, ten patients were reclassified in TMB categories. As would be expected with lower tumor purity, there were four patients with tumor purity <20% that switched class from the low observed TMB to the high corrected TMB category but there was also one patient with tumor purity >20% that switched from low observed TMB to high corrected TMB category and also five patients with higher tumor purity (range 46–82%) that switched class from the high observed TMB to the low corrected TMB category. Switching between both TMB categories would be expected as our approach generates corrected TMB estimates that better capture the true TMB distribution and that in turn would affect the percentile threshold that defines high TMB more accurately. NA; radiographic response non evaluable.