Accurate and generalizable protein-ligand binding affinity prediction with geometric deep learning

2025-04-22

AI Summary

Existing protein–ligand binding affinity (PLA) prediction methods exhibit severely limited generalization to novel, unseen target proteins. To address this, we propose IPBind—the first geometric deep learning framework for PLA prediction that incorporates machine-learned interatomic potentials. IPBind employs an equivariant graph neural network to jointly encode multiple conformations of both bound and unbound states, explicitly modeling physical interatomic potentials. This design enables robust zero-shot prediction for unseen targets. Our approach achieves strong generalization and atomic-level interpretability: it attains Spearman correlations exceeding 0.82 on PDBbind and BindingDB benchmarks; improves cross-protein generalization by over 15% relative to state-of-the-art methods; and supports end-to-end visualization of atomic-level contribution scores.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Protein-ligand binding complexes are ubiquitous and essential to life. Protein-ligand binding affinity prediction (PLA) quantifies the binding strength between ligands and proteins, providing crucial insights for discovering and designing potential candidate ligands. While recent advances have been made in predicting protein-ligand complex structures, existing algorithms for interaction and affinity prediction suffer from a sharp decline in performance when handling ligands bound with novel unseen proteins. We propose IPBind, a geometric deep learning-based computational method, enabling robust predictions by leveraging interatomic potential between complex's bound and unbound status. Experimental results on widely used binding affinity prediction benchmarks demonstrate the effectiveness and universality of IPBind. Meanwhile, it provides atom-level insights into prediction. This work highlights the advantage of leveraging machine learning interatomic potential for predicting protein-ligand binding affinity.

Problem

• Predict protein-ligand binding affinity accurately
• Improve performance with unseen protein-ligand complexes
• Provide atom-level insights into binding interactions

Innovation

• Geometric deep learning for protein-ligand binding
• Interatomic potential between bound and unbound states
• Atom-level insights in affinity prediction