Extended Data Fig. 8: Development of the ex vivo drug-efficacy prediction model by using combinations of multiple PPI metrics.
a, Comparison of BCL2 family PPI profiles from the healthy donor PBMC sample and HL60 cells with SMPC (n = 10 independent images). The data were normalized to the HL60 cell. b, Correlation between ex vivo AUC for ABT-199 and combination of BCL2-related metrics (BCL2 total level, BCL2-BIMBH3 PBA, BCL2-BAD PBA, and BCL2-BAX CPX) (One-sided F-test, p-value = 1.3e-10). c, Linear correlations between two different BCL2 PBA metrics. d, Correlations between ex vivo AUC and IC50 of primary AML samples for ABT-199 (n = 32). e, Schematic of Lasso regression analysis for the selection of PPI metrics highly correlated with drug response. The training and test groups were randomly selected from the primary AML sample cohort. The Lasso regression models were generated using the training group and evaluated based on Pearson’s R as well as prediction outcomes for the test group. 67 models were selected from 10,000 different initial models. f, Identification of outliers in the ABT-199 drug efficacy prediction models. The residuals of each outlier in the model (ex vivo AUC – estimated score) were indicated. g, Correlations between ex vivo AUC and IC50 of primary AML samples for AZD-5991 (n = 27). h, Correlation between ex vivo AUC for AZD-5991 and combination of MCL1-related metrics (MCL1 total level, MCL1-BIMBH3 PBA, MCL1-NOXA PBA, and MCL1-BAK CPX) (One-sided F-test, p-value = 2.5e-5).
