GPCRact: a hierarchical framework for predicting ligand-induced GPCR activity via allosteric communication modeling

Area of Science:

• Computational chemistry and structural biology
• Pharmacology and drug discovery
• Machine learning in bioinformatics

Background:

• Predicting G-protein-coupled receptor (GPCR) activity upon ligand binding is crucial for drug discovery.
• Modeling allosteric communication, the long-range signaling in GPCRs, remains a significant challenge for current sequence-based models.
• Existing models often fail to capture the complex three-dimensional dynamics of GPCR activation, particularly for allosterically complex targets.

Purpose of the Study:

• To introduce GPCRact, a novel computational framework designed to model the biophysical principles of allosteric modulation in GPCR activation.
• To develop a method that accurately predicts ligand-induced GPCR activity by explicitly considering three-dimensional structural dynamics and allosteric pathways.
• To provide a more interpretable and mechanistically grounded tool for structure-guided drug discovery targeting GPCRs.

Main Methods:

• Construction of a high-resolution, 3D structure-aware graph representing functionally critical residues in GPCRs.
• Implementation of a dual attention architecture, including cross-attention for ligand-protein interactions and self-attention for signal propagation, built upon an E(n)-Equivariant Graph Neural Network (EGNN).
• Development of a tailored loss function and inference logic to mitigate error propagation and explicitly model conformational changes.

Main Results:

• GPCRact achieves state-of-the-art performance in predicting GPCR activity.
• The framework demonstrates superior accuracy on a benchmark of allosterically complex receptors where conventional models underperform.
• Analysis of learned attention weights validates the identification of biologically relevant allosteric pathways, enhancing model interpretability.

Conclusions:

• GPCRact offers a significant advancement in accurately predicting ligand-induced GPCR activity by effectively modeling allosteric communication.
• The framework provides a more interpretable and mechanistically grounded approach compared to previous methods, addressing the 'black box' nature of GPCR modulation.
• GPCRact paves the way for more effective structure-guided drug discovery by providing a robust tool for understanding GPCR activation dynamics.