Fig. 3: Differences in immune features between male and female patients from TCGA.

a Overview of the propensity score algorithm used to balance confounding effects, including age, race, tumor purity, tumor stage, subtype and smoking history, and to evaluate the sex-associated immune features, including TMB, neoantigen load, TCR/BCR, checkpoints, immune cell population and aneuploidy, across cancer types. b Differences of molecular biomarkers, including TMB, neoantigen load, GEP, CYT and PD-L1 protein expression, reported in immunotherapy data sets between male patients and female patients. Bar plots indicate the number of significant female-biased features minus the number of significant male-biased features. c Differences of relative abundance of six immune cell populations, including active CD4/CD8 T cells, effector memory CD4/CD8 T cells, myeloid-derived suppressor cell and regulatory T cells. d Differences of mRNA expression level of 34 immune checkpoints, including LAG3, CTLA-4, PDCD1 and CD274. X axis denotes immune features. Y axis of b–d denotes 22 cancer types analyzed by the propensity score algorithm and ordered by the number of significant female-biased features minus the number of significant male-biased features in b. Statistical analysis was performed using a propensity score algorithm to identify immune-associated features (see Supplementary methods). p-value was calculated by linear regression model and adjusted by Benjamini and Hochberg correction. FDR is labeled as red dots (female-bias) and blue dots (male-bias) in b–d. Sample size for each data set was listed in Supplementary Table 3.