Figure 4.

Binding affinity prediction is specific to each interaction type. For each type of molecular interaction (protein‐small molecule, protein‐DNA/RNA, and protein−protein), we fit a statistical model using cross‐validation (see Methods, Fig. 2) and evaluated each model's accuracy on set‐aside testing data of either the same interaction type or a different interaction type. A: We plot the mean and standard error in Pearson's correlation (r ²) between predicted and experimental pKd, showing how the statistical models trained using each type of training data predicted pKds on testing data of either the same or different type. B: We plot the mean and standard error in r ² between predicted and experimental pKd, comparing a general statistical model trained on all data sets (gray) to specific models (white) trained on each data set, respectively. C: We show the mean and standard error in root mean square deviation (RMSD) between predicted and experimental pKd, comparing a general statistical model (gray) trained on all data sets to specific models (white) trained on each data set.