FLOWR -- Flow Matching for Structure- and Interaction-Aware De Novo Ligand Generation

Julian Cremer, Ross Irwin, Alessandro Tibo, Jon Paul Janet, Simon Olsson, Djork-Arné Clevert

Track: Machine learning: computational method and/or computational results

Nature Biotechnology: No

Keywords: Flow Matching, Structure-Based Drug Discovery, Equivariance, Generative Chemistry, Pocket-ligand-Interactions, Fragment-based design

Abstract: We present our progress on overcoming key challenges in applying generative models to 3D ligand design, including generating high-quality binders and reducing inference times. We introduce FLOWR, a flow matching framework for 3D ligand generation conditioned on a protein pocket and a set of desired interaction between the protein and the ligand. To thoroughly evaluate our model we also introduce SPIRE, a refined dataset of high-quality protein-ligand complexes derived from crystallographic data. Evaluations on this dataset show that FLOWR outperforms an existing state-of-the-art diffusion model, while achieving up to a 50-fold speed-up in inference time. We also propose an interaction-aware training and inference strategy that enables the generation of novel ligands tailored to predefined interaction profiles. Our findings suggest that FLOWR is an important step forward for efficient, AI-driven de novo ligand generation.

Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.

Presenter: ~Ross_Irwin1

Format: Yes, the presenting author will definitely attend in person because they attending ICLR for other complementary reasons.

Funding: No, the presenting author of this submission does *not* fall under ICLR’s funding aims, or has sufficient alternate funding.

Submission Number: 108