Summary
Biomolecular diffusion models cannot reproduce the equilibrium dynamics of the simulations they are trained on (1), and do not always recover the ground states (2). This is due to undersampling at low- to avoid exploding errors resulting from low variance. A workaround is to use Fokker-Planck regularization during training.
Figures
Ref (1)
1.
Plainer M, Wu H, Klein L, Günnemann S, Noé F. Consistent Sampling and Simulation: Molecular Dynamics with Energy-Based Diffusion Models. 2025; Available from: https://arxiv.org/abs/2506.17139
2.
Sarma S, Truscott H, Xu D, Reid K, Chu L-S, Chen J, et al. Can We Extract Physics-like Energies from Generative Protein Diffusion Models?. openRxiv; 2025. Available from: https://doi.org/10.1101/2025.11.28.690021