PRECOGx: exploring GPCR signaling mechanisms with deep protein representations

Area of Science:

• Biochemistry and Bioinformatics
• Computational Biology
• Genomics

Background:

• G protein-coupled receptors (GPCRs) are crucial cell surface receptors involved in numerous physiological processes.
• Understanding GPCR signaling and functional repertoire is essential for drug discovery and disease research.
• Existing methods for predicting GPCR-transducer interactions have limitations in scope and accuracy.

Purpose of the Study:

• To develop a machine learning predictor, PRECOGx, leveraging protein language models for GPCR sequence analysis.
• To accurately predict GPCR interactions with G protein and β-arrestin.
• To provide a user-friendly webserver for exploring GPCR functions and analyzing variants.

Main Methods:

• Utilized ESM1b protein embeddings as features for machine learning models.
• Integrated publicly available GPCR binding information.
• Developed PRECOGx, a machine learning predictor, and a webserver for accessibility.
• Employed attention map analysis and predicted intramolecular contacts for interpretability.

Main Results:

• PRECOGx demonstrated superior performance in predicting GPCR-transducer couplings compared to previous methods.
• The tool successfully predicted the impact of disease variants (ClinVar) and alternative splice forms (GTEX) on human GPCRs.
• PRECOGx identified sequence and structural determinants of GPCR coupling and analyzed GPCRome features.

Conclusions:

• Protein language models can effectively encode GPCR structural and functional information for predictive tasks.
• PRECOGx offers a powerful platform for dissecting GPCR signaling mechanisms in both health and disease.
• The webserver facilitates the study of GPCR variants and their functional consequences.