A 7-gene signature accurately predicts response to bortezomib-based therapy in PADIMAC and the independent CoMMpass data sets. (A) MCCs and F measures of bortezomib-good assignments by 4 to 11 gene signatures derived from synthetic annealing after cross-validation within the PADIMAC data set. (B-C) MCCs and F measures of bortezomib-good assignments by the 7-gene signature after multiple (n = 100) cross-validations within the PADIMAC data set (signature) compared with the MCCs and F measures of permuted assignments (null). The P values are those of the Wilcoxon-Mann-Whitney test, under the null hypothesis that the distributions of observed and null performances are the same. (D) Kaplan-Meier plot showing the progression-free survival (PFS) of patients who received bortezomib-based therapy within CoMMpass (n = 147) and who were predicted to benefit (n = 39; broken line) or not (n = 108; solid line) from bortezomib-based therapy by the 7-gene signature after training in PADIMAC. The P value and hazard ratios (HRs) are those obtained from Cox regression analysis. (E) HRs for disease progression of bortezomib-good vs bortezomib-standard patients who received bortezomib-based therapy in CoMMpass. Predictions were made by the 7-gene signature, trained in PADIMAC, and followed repeated (n = 100) training/validation splits (signature). The HRs are compared with a null data set of HRs obtained after permutations of the assignments (null). The P value is that of the Wilcoxon-Mann-Whitney test, under the null hypothesis that the distributions of observed and null performances are the same.