GPCRVS - AI-driven decision support system for GPCR virtual screening

Abstract

SummaryG protein-coupled receptors (GPCRs) constitute the largest and most frequently used family of molecular drug targets. The simplicity of GPCR drug design results from their common seven-transmembrane helix topology and well-understood signaling pathways. GPCRs are extremely sensitive to slight changes in the chemical structure of compounds, which allows for the reliable design of highly selective and specific drugs. Only recently has the number of GPCR structures, both in their active and inactive conformations together with their active ligands, become sufficient to comprehensively apply machine learning in decision support systems to predict compound activity for drug design. Here, we describe GPCRVS, an efficient machine learning web service for the online assessment of compound activity against several GPCR targets, including peptide and protein-binding GPCRs, the most difficult for virtual screening tasks. As a decision support system, GPCRVS evaluates compounds in terms of their activity range, the pharmacological effect they exert on the receptor, and the binding modes they could possibly demonstrate for certain types of GPCR receptors. GPCRVS can be applied to compounds ranging from small molecules to short peptides uploaded as common chemical format files. The activity class assignment and the binding affinity prediction are provided in a reference to similar predictions precomputed for already known active ligands of each GPCR target. This multiclass classification in GPCRVS, handling often incomplete and fuzzy biological data, was validated on ChEMBL-retrieved data sets for secretin-like and chemokine GPCR receptors. Availability and ImplementationGPCRVS is freely available at: https://gpcrvs.chem.uw.edu.pl