B E, N. | Search bioRxiv

bioRxiv2026

How Generative Models Approach Molecular Conformational Sampling

Characterising equilibrium conformational ensembles with deep generative models requires assessing not only whether a model reproduces the target distribution, but also the mechanism of how it arrives here. Here we examine two distinct routes to generative conformational sampling-- stochastic relaxation and deterministic transport--through a study of denoising diffusion probabilistic models (DDPM) and rectified-flow (RF) models across molecular systems of increasing complexity. Using systems of increasing complexity, including a multimodal two-dimensional potential, the folded mini-protein Trp-cage, and a high-dimensional dihedral subspace of the intrinsically disordered protein -synuclein, we show that the key distinction between these paradigms lies not only in endpoint fidelity but in how distributional error is resolved during sampling. Diffusion models converge via pronounced late-stage stochastic relaxation and exhibits robust recovery of configurational breadth across neural architectures. Rectified flow approaches the target more gradually through deterministic transport and therefore depends much more strongly on architectural expressivity, particularly in heterogeneous high-dimensional landscapes. Analyses of entropy and moment evolution further show that diffusion more reliably restores both ensemble location and fluctuation structure, whereas RF requires Transformer-level feature mixing to represent the transport geometry accurately. These results establish convergence mechanism as a key design principle for generative sampling.

biophysics↗

bioRxiv2026

Fast Adversarial Generation of Molecular Dynamics Trajectories with Kinetic Fidelity

Molecular dynamics (MD) simulations yield atomic-level insights into molecular motion but struggle to reach the long timescales needed for rare events due to prohibitive computational costs. Generative machine-learning models (e.g., diffusion models and normalizing flows) offer a promising route to accelerate sampling, yet they generate independent equilibrium snapshots without temporal correlation or kinetic information. Autoregressive sequence models can learn time evolution by producing one frame at a time, but this stepwise generation often accumulates errors and drifts from true dynamics. Here, we propose a complementary approach inspired by advances in image and video generation: we treat finite MD trajectory segments as high-dimensional objects and learn their joint distribution using Generative Adversarial Networks (GANs). Using a Wasserstein GAN with gradient penalty, we directly generate entire time-series trajectories in one shot, that remain physically coherent over time without explicitly integrating the equations of motion. We demonstrate the generality of this method on molecular systems of increasing complexity: a 2D triple-well potential energy landscape, a protein-ligand binding process (cytochrome P450), the dynamics of an intrinsically disordered protein (-synuclein) in a latent coordinate space, and even the conditional generation of folding trajectories for the Trp-cage mini-protein. In all cases, the GAN-generated trajectories closely reproduce the true free-energy landscapes and kinetic signatures of the systems, while enabling efficient sampling of rare events that would ordinarily require months of conventional MD simulation.

biophysics↗