Search:
Match:
1 results

Analysis

This paper introduces EnFlow, a novel framework that combines flow matching with an energy model to efficiently generate low-energy conformer ensembles and identify ground-state conformations of molecules. The key innovation lies in the energy-guided sampling scheme, which leverages the learned energy function to steer the generation process towards lower-energy regions. This approach addresses the limitations of existing methods by improving conformational fidelity and enabling accurate ground-state identification, particularly in a few-step regime. The results on benchmark datasets demonstrate significant improvements over state-of-the-art methods.
Reference

EnFlow simultaneously improves generation metrics with 1--2 ODE-steps and reduces ground-state prediction errors compared with state-of-the-art methods.