TCGA driver mutation and HLA affinity analysis.

(A) Illustration of the 2 matrices underlying the study’s analytical approach. Driver hot spot missense mutations were identified in TCGA data (see Methods) and a patient x mutation matrix was generated with each cell containing a binary value (mutation observed or unobserved). A corresponding HLA affinity matrix was then calculated with values representing PHBR values. (B) Mutation frequencies for the 20 most frequent driver mutations. Grey scale shown in top right. Pan cancer mutation frequencies indicated by bar plot on top. (C) Box plots compare PHBR values between observed (+) and unobserved (-) mutations with or without randomization of patients/mutations as indicated. Box plots indicate median values and lower/upper quartiles with whiskers extending to 1.5 times the interquartile range. P values calculated using Wilcoxon rank-sum test. (D-E) Pan cancer (D) and per cancer (E) logistic regression analysis between log PHBR (observed variable) and mutation status (0/1, response variable). Analysis was performed both using the within-patient (rows of the matrices) and within-mutation (columns) regression model as indicated (see Methods for details). Bar plot (D) and dots (E) show odds ratios with 95% confidence intervals. Asterisks indicate significance at 5% FDR.