PFM: perturbed flow matching for structure-based drug design.

Perturbed Flow Matching (PFM) enables protein pocket-conditioned 3D molecular generation on CrossDocked2020, achieving the best average Vina score among unguided methods at -7.12.

• Why it matters: Structure-based 3D molecular generation needs flexible forward noising and conditional process design, but diffusion methods often rely on marginal distributions with closed-form solutions.
• What they did: PFM uses a perturbed conditional probability path that combines pocket binding-site information with coupled atom type-coordinate modeling, then evaluates unguided and PFM-G variants.
• The result: On CrossDocked2020, PFM-G with tailored classifier-free guidance improves the average Vina score to -8.70, and results on additional datasets support effectiveness and generalizability.

AI summary of the abstract · may contain errors · not medical advice · read the paper.