AiGPro: a multi-tasks model for profiling of GPCRs for agonist and antagonist

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• G protein-coupled receptors (GPCRs) are crucial drug targets, but existing computational models are limited in scope and applicability for high-throughput screening.
• Deep learning offers powerful tools for drug discovery, yet its application to comprehensive GPCR profiling remains underdeveloped.

Purpose of the Study:

• To develop a novel multitask deep learning model, AiGPro, for predicting small molecule agonists and antagonists across all 231 human GPCRs.
• To create a scalable and accessible platform for large-scale GPCR profiling and accelerate GPCR-targeted drug discovery.

Main Methods:

• Developed AiGPro, a multitask deep learning model incorporating multi-scale context aggregation and bidirectional multi-head cross-attention (BMCA).
• Employed a dual-label prediction strategy for simultaneous classification of agonists, antagonists, or both, with confidence scores.
• Validated the model using stratified tenfold cross-validation on 231 human GPCR targets.

Main Results:

• AiGPro achieved a robust performance with a Pearson's correlation coefficient of 0.91, demonstrating broad generalizability.
• The model accurately predicts both agonist and antagonist activities, outperforming previous studies.
• The Bi-Directional Multi-Head Cross-Attention (BMCA) module effectively integrates protein and ligand features for precise molecular interaction prediction.

Conclusions:

• AiGPro represents a first-in-class solution for comprehensive GPCR profiling, enhancing prediction accuracy and accelerating virtual screening.
• The developed web-based platform provides easy access for rapid screening of small-molecule libraries, facilitating GPCR-targeted drug discovery.
• This multitask approach offers a more complete understanding of ligand bioactivity across the diverse GPCR superfamily.