On the Choice of Active Site Sequences for Kinase-Ligand Affinity Prediction

Abstract

Recent work showed that active site rather than full-protein-sequence information improves predictive performance in kinase-ligand binding affinity prediction. To refine the notion of an "active site", we here propose and compare multiple definitions. We report significant evidence that our novel definition is superior to previous definitions and better models of ATP-noncompetitive inhibitors. Moreover, we leverage the discontiguity of the active site sequence to motivate novel protein-sequence augmentation strategies and find that combining them further improves performance.

Figure 1

Overview of active site site definitions and representations. A) Visualization of cAMP-dependent protein kinase catalytic subunit alpha (P17612). Residues unique to the active site definitions of refs (8) and (10) are colored in orange and green, respectively. Residues contained in both definitions are shown in red. B) Partial amino acid sequence (residues 48–62) of the same kinase. The upper gray panel displays the four kinase sequence representations examined in this work. The lower gray panel visualizes three kinase augmentation strategies, exemplified on the “combined” active site definition: flipping (i.e., reversing) the entire sequence, flipping contiguous subsequences, and swapping neighboring subsequences. Residues affected by the augmentation are encircled in black.

Figure 2

RMSE in affinity prediction for kinase split on validation and test data. 10-fold cross-validation results on kinase data from BindingDB. Performance of validation (A) and test data (B) is shown. Statistically significant differences between the three different active site configurations are marked with a star.