Fig 5. Predicted vs. observed log(IC₅₀) for leave-drug-out cross-validation.

Randomly selected drugs were omitted from the GDSC data set, and random forest regression models were trained using the data set containing the remaining 224 drugs and 990 cell lines. This process was performed for each of 9 drugs. The RMSE, Pearson correlation (R_P), Spearman rank correlation (ρ), and corresponding regression line are shown for each omitted drug. Structures of the omitted drugs are depicted in the insets.