Global machine learning model predicting activity level of any GPCRs based on protein structure.

Abstract

GPCRs are the target for one-third of all FDA-approved drugs. However, the development of new drug molecules targeting GPCRs is limited by the lack of mechanistic understanding of the GPCR structure-activity-function relationship. Understanding the connection between structure and activity level of GPCRs through explainable methods will help elucidate the activation mechanism. In this study, we develop ML models to predict the conformation state of GPCR proteins with high prediction accuracy (97.37%). Additionally, we predict the activity level of GPCRs based on their structures using ML models trained with conserved features in GPCRs (MAE is 8.55%). Furthermore, we build ML models to explore key features in the transition between GPCRs activity states. We applied these models to molecular dynamics trajectories of ß2AR protein and predicted the activity of every single frame based on the structural information. We validated the accuracy of our model by correlating the global activation features with our ML activity regressor. Moreover, this model helps us predict potential correlation and causality among protein features along the activation path. These models address the challenges with the receptors where our knowledge about their three-dimensional structures in different states is limited due to challenges in X-ray crystallography. The proposed models can be applied to two problems; the first is to build configurations of proteins with a specific activation level where their three-dimensional structures are unknown, second is to characterize signaling pathways in GPCRs which is critical in designing efficient drugs targeting GPCRs.