FlexiFlow: decomposable flow matching for generation of flexible molecular ensemble

2025-11-21

AI Summary

In drug discovery, molecular 3D conformational diversity critically influences physicochemical properties and target binding affinity; however, existing 3D de novo design models—such as flow matching and diffusion-based approaches—generate only a single conformation per molecule. To address this limitation, we propose the first decomposable, SE(3)-equivariant flow matching framework enabling joint molecule–conformer ensemble sampling. Our method is the first to extend flow matching to conformational ensemble generation while strictly preserving SE(3) equivariance and atomic permutation invariance. By decoupling molecular scaffold generation from conformational sampling, it ensures both chemical validity and conformational diversity. The model supports both unconditional and protein-conditioned ligand generation. On QM9 and GEOM-Drugs benchmarks, it achieves state-of-the-art performance: >98% molecular validity, >95% novelty, conformational coverage on par with physics-based methods (RMSD < 0.5 Å), and >10× faster inference speed.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: 3D Computer Vision

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

Sampling useful three-dimensional molecular structures along with their most favorable conformations is a key challenge in drug discovery. Current state-of-the-art 3D de-novo design flow matching or diffusion-based models are limited to generating a single conformation. However, the conformational landscape of a molecule determines its observable properties and how tightly it is able to bind to a given protein target. By generating a representative set of low-energy conformers, we can more directly assess these properties and potentially improve the ability to generate molecules with desired thermodynamic observables. Towards this aim, we propose FlexiFlow, a novel architecture that extends flow-matching models, allowing for the joint sampling of molecules along with multiple conformations while preserving both equivariance and permutation invariance. We demonstrate the effectiveness of our approach on the QM9 and GEOM Drugs datasets, achieving state-of-the-art results in molecular generation tasks. Our results show that FlexiFlow can generate valid, unstrained, unique, and novel molecules with high fidelity to the training data distribution, while also capturing the conformational diversity of molecules. Moreover, we show that our model can generate conformational ensembles that provide similar coverage to state-of-the-art physics-based methods at a fraction of the inference time. Finally, FlexiFlow can be successfully transferred to the protein-conditioned ligand generation task, even when the dataset contains only static pockets without accompanying conformations.

Problem

• Generating multiple molecular conformations beyond single structures
• Capturing conformational diversity to assess molecular properties accurately
• Overcoming limitations of current models in flexible ensemble generation

Innovation

• Extends flow-matching models for joint molecular-conformation sampling
• Preserves equivariance and permutation invariance in generation
• Generates conformational ensembles faster than physics-based methods