Improving protein-ligand complex generation with force field guidance

Helen Lai, Tingyu Wang, Hassan Sirelkhatim, Joe Eaton, Howard Huang, Brad Rees, Ola Engkvist, Jon Paul Janet, Xiaoyun Wang, Alessandro Tibo

Keywords: Structure-based drug design, Protein–ligand generation, Diffusion models, Flow matching, Guidance, Force fields, Cheminformatics

TL;DR: We introduce an energy-guided sampling framework that improves the physical realism and docking performance of generative protein--ligand models, and provably ensures monotonic energy descent without retraining.

Abstract: Generative models based on diffusion and flow matching have recently been applied to structure-based drug design, but their outputs often include unrealistic protein–ligand interactions that fail to obey the laws of physics. We present an energy guidance framework that incorporates a molecular mechanics force field (MMFF94) directly into the sampling process. The method steers molecular generation toward more physically plausible and energetically stable conformations without retraining the underlying model. We evaluate this approach using two state-of-the-art architectures, SemlaFlow, a flow matching model, and EDM, a diffusion model, on the PDBBind dataset. Across both models, energy guidance improves binding affinity predictions, reduces strain energy by up to 75%, and generates over 1,000 ligands with better docking scores than native ligands. These results demonstrate that lightweight, physics-based guidance can significantly enhance generative drug design while preserving chemical validity and diversity.

Release To Public: Yes, please release this paper to the public

Submission Number: 18