Stochastic Control Policies for Robust Molecular Transition Path Sampling

Transition path sampling (TPS) aims to efficiently generate rare molecular transition trajectories between metastable states and is essential for understanding biomolecular mechanisms. A major class of learning-based methods learns control forces during explicit molecular dynamics (MD) rollouts, which preserves the underlying dynamics and tends to produce more physically plausible trajectories than endpoint-conditioned generators, but has been reported to exhibit unstable and strongly seed-dependent performance. We investigate stochasticity placement as a design choice for improving exploration and optimization robustness. We develop two stochastic policies: FS-TPS, which directly parameterizes a state-dependent Gaussian distribution over the control policy output, and LaS-TPS, which samples a compact latent control variable and decodes it into structured, cross-atom-correlated force variation. In multi-seed experiments on alanine dipeptide, chignolin, and the fast-folding protein BBL, stochastic policies consistently improve transition success and path quality over deterministic-policy baselines while substantially reducing sensitivity to random initialization.