FLOWR: Flow Matching for Structure-Aware De Novo, Interaction- and Fragment-Based Ligand Generation

2025-04-14

AI Summary

To address key challenges in 3D ligand generation—including weak structural awareness, coarse-grained protein–ligand interaction modeling, insufficient fragment utilization, and low-quality training data—this paper introduces FLOWR: the first protein-pocket-conditioned 3D ligand generation paradigm integrating continuous/discrete flow matching with equivariant optimal transport. We curate SPINDR, a high-fidelity ligand–pocket structural dataset; support three generation modes—interaction-aware sampling, fragment-guided generation, and de novo design; and achieve, for the first time, multi-objective targeted sampling without retraining (FLOWR.multi). On the PoseBusters benchmark, FLOWR significantly outperforms existing diffusion- and flow-based models in binding pose accuracy, interaction recovery rate, and molecular validity, while accelerating inference by up to 70×.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation

Application Category

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

Abstract

We introduce FLOWR, a novel structure-based framework for the generation and optimization of three-dimensional ligands. FLOWR integrates continuous and categorical flow matching with equivariant optimal transport, enhanced by an efficient protein pocket conditioning. Alongside FLOWR, we present SPINDR, a thoroughly curated dataset comprising ligand-pocket co-crystal complexes specifically designed to address existing data quality issues. Empirical evaluations demonstrate that FLOWR surpasses current state-of-the-art diffusion- and flow-based methods in terms of PoseBusters-validity, pose accuracy, and interaction recovery, while offering a significant inference speedup, achieving up to 70-fold faster performance. In addition, we introduce FLOWR.multi, a highly accurate multi-purpose model allowing for the targeted sampling of novel ligands that adhere to predefined interaction profiles and chemical substructures for fragment-based design without the need of re-training or any re-sampling strategies

Problem

• Generating 3D ligands with structure-aware design
• Improving ligand-pocket interaction accuracy and validity
• Enabling fragment-based ligand sampling without retraining

Innovation

• Continuous and categorical flow matching integration
• Efficient protein pocket conditioning enhancement
• Multi-purpose model for targeted ligand sampling