The significance of association of eight mutation-based and two copy number-based features with therapeutic response was compared with that of TMB (log 10 of feature to TMB P value ratios visualized). For each cohort, the feature with the most significant association with therapeutic response is marked with an ‘x’ mark. For all ICB cohorts, pTMB outperformed TMB in predicting ICB response. Notably, the best-performing feature in NSCLC was clonal pTMB (MW P = 1.03 x 10 -4, P = 1.60 x 10 -3 for NSCLC-Anagnostou and NSCLC-Shim, respectively), while in melanoma the number of multi-copy persistent mutations (MW P = 5.42 x 10 -7 ) and in mesothelioma the number of only-copy persistent mutations (MW P = 3.15 x 10 -2 ) better distinguished between responding and nonresponding tumors. pTMB outperformed loss-prone mutation load in distinguishing responding from nonresponding tumors (HNSCC: MW P = 0.16 versus P = 0.05; melanoma: P = 1.92 x 10 3 versus P = 2.25 x 10 -6 ; mesothelioma: P = 0.09 versus P = 0.03; NSCLC-Anagnostou: P = 1.03 x 10 -2 versus P = 1.26 x 10 -4 ; NSCLC-Shim: P = 3.20 x 10 -2 versus P = 1.87 x 10 -3 ). Tumor aneuploidy or WGD alone failed to predict ICB response. P values are two-sided. CN, copy number; MW, Mann–Whitney; NR, nonresponding; R, responding.
