A Hybrid Structure-Based Machine Learning Approach for Predicting Kinase Inhibition by Small Molecules

Kinases have been the focus of drug discovery programsfor threedecades leading to over 70 therapeutic kinase inhibitors and biophysicalaffinity measurements for over 130,000 kinase-compound pairs. Nonetheless,the precise target spectrum for many kinases remains only partly understood.In this study, we describe a computational approach to unlocking qualitativeand quantitative kinome-wide binding measurements for structure-basedmachine learning. Our study has three components: (i) a Kinase InhibitorComplex (KinCo) data set comprising in silico predictedkinase structures paired with experimental binding constants, (ii)a machine learning loss function that integrates qualitative and quantitativedata for model training, and (iii) a structure-based machine learningmodel trained on KinCo. We show that our approach outperforms methodstrained on crystal structures alone in predicting binary and quantitativekinase-compound interaction affinities; relative to structure-freemethods, our approach also captures known kinase biochemistry andmore successfully generalizes to distant kinase sequences and compoundscaffolds.

Automated summary

This study presents a computational approach for qualitative and quantitative kinome-wide binding measurements using structure-based machine learning. The approach outperforms methods trained on crystal structures alone and structure-free methods in predicting kinase-compound interaction affinities. It also successfully captures known kinase biochemistry and generalizes well to distant kinase sequences and compound scaffolds.