pdCSM-GPCR: predicting potent GPCR ligands with graph-based signatures

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• G protein-coupled receptors (GPCRs) are crucial drug targets involved in numerous cellular processes.
• Developing specific and safe GPCR ligands is challenging due to receptor diversity and complexity.
• Existing computational methods for GPCR drug development have limitations in performance and generalization.

Purpose of the Study:

• To develop a novel computational method, pdCSM-GPCR, for rapid and accurate screening of GPCR ligands.
• To create a comprehensive computational resource for GPCR bioactivity prediction.
• To identify key features of potent GPCR ligands.

Main Methods:

• Utilized graph-based signatures to develop predictive models for GPCR ligands.
• Curated bioactivity data (IC50, EC50, Ki, Kd) for 36 major GPCR targets across 4 classes.
• Employed stratified 10-fold cross-validation and blind tests to evaluate model performance.

Main Results:

• Achieved high prediction accuracy with Pearson's correlations up to 0.89, outperforming previous methods.
• Identified common features of potent GPCR ligands, such as bicyclic rings and high aromaticity.
• Developed the most comprehensive computational resource for GPCR bioactivity prediction to date.

Conclusions:

• pdCSM-GPCR offers a significant advancement in computational screening of GPCR ligands.
• The method can assist in enriching compound libraries and ranking potential drug candidates.
• The predictive models and datasets are publicly available via a web server.