A 3D pocket-aware and affinity-guided diffusion model for lead optimization

2025-04-29

AI Summary

In drug discovery, existing 3D generative models struggle to effectively model protein–ligand binding affinity during molecular optimization. This paper introduces Diffleop, a pocket-aware, affinity-guided SE(3)-equivariant diffusion model. Its key innovations are: (i) the first explicit incorporation of gradient signals from an affinity predictor into the denoising process, enabling affinity-driven 3D structural optimization; and (ii) a pocket-conditioned encoder that imposes geometric constraints to ensure spatial complementarity between generated ligands and the target binding pocket. Evaluated on multiple benchmarks, Diffleop achieves a 1.8× improvement in mean predicted binding affinity over state-of-the-art methods, while 92% of its generated molecules satisfy both drug-likeness and pocket-complementarity constraints—marking significant advances in structure-based generative drug design.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Learning to Search
• Machine Learning: Deep Generative Models & Autoencoders

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Molecular optimization, aimed at improving binding affinity or other molecular properties, is a crucial task in drug discovery that often relies on the expertise of medicinal chemists. Recently, deep learning-based 3D generative models showed promise in enhancing the efficiency of molecular optimization. However, these models often struggle to adequately consider binding affinities with protein targets during lead optimization. Herein, we propose a 3D pocket-aware and affinity-guided diffusion model, named Diffleop, to optimize molecules with enhanced binding affinity. The model explicitly incorporates the knowledge of protein-ligand binding affinity to guide the denoising sampling for molecule generation with high affinity. The comprehensive evaluations indicated that Diffleop outperforms baseline models across multiple metrics, especially in terms of binding affinity.

Problem

• Optimizing molecules for improved binding affinity in drug discovery
• Addressing inadequate consideration of protein-ligand affinity in deep learning models
• Enhancing molecule generation with 3D pocket-aware diffusion guidance

Innovation

• 3D pocket-aware diffusion model for molecules
• Affinity-guided denoising sampling technique
• Enhanced binding affinity in lead optimization