Fig. 4. Expression-based predictions of drug response indicate concordance of independent ex vivo datasets and may be improved by incorporating general response across drugs.

Performance (Pearson correlation between observed and model-predicted drug response; y axis) of ridge regression models trained on OHSU data and tested on FIMM data using genes as predictors (Gene Expression), genes and GRD predicted by applying ridge regression to gene expression (Gene Expression + Predicted GRD), or genes and GRD calculated from drug response data (Gene Expression + Observed GRD). Each point corresponds to a drug (n = 87). Drug d is excluded from observed and predicted GRD in modeling d’s response. ****One-sided paired Wilcoxon signed rank p < 0.0001; ***p < 0.001.