Efficient sampling of short protein trajectories with conditional diffusion models

Sampling continuous molecular dynamics and biomolecular conformational transitions traditionally requires computationally intensive quantum mechanical or hybrid QM/MM simulations. In this work, we propose a conditional diffusion generative modeling framework capable of efficiently generating continuous, physically viable short protein trajectories. By conditioning on end-point structural coordinates and learned molecular representations, our diffusion model reconstructs transitional trajectories with high geometric fidelity, bypassing standard high-cost physics calculations and accelerating conformational sampling.

Journal of Chemical Theory and Computation (ACS), 22(1), 78