Breaking timescales with generative sampling of conformational transitions

Abstract

Molecular transitions, including protein folding, allostery and membrane transport, are central to biological functions, yet remain notoriously difficult to simulate. Their intrinsic rarity places them beyond the reach of standard molecular dynamics, whereas enhanced-sampling strategies are computationally demanding and often depend on arbitrarily chosen parameters and variables that bias outcomes1, 2–3. Here we introduce Gen-COMPAS, a generative committor-guided path-sampling framework that reconstructs transition pathways without predefined collective variables and at acceptable computational cost. Gen-COMPAS couples a denoising diffusion probabilistic model, which produces structurally plausible intermediate targets, with committor-based filtering to identify transition states4,5. Short unbiased simulations from these intermediates yield transition-region ensembles at nanosecond-to-submicrosecond aggregate sampling scales for which conventional approaches require orders of magnitude more sampling. Applied to systems ranging from a miniprotein to a pentameric, ligand-gated ion channel, Gen-COMPAS recovers committors, transition states and free-energy landscapes from known end-point structures alone, without predefined reaction coordinates or prior mechanistic knowledge, thereby providing a computationally tractable route to mechanistic insight in biomolecular systems that have so far resisted conventional simulation approaches. A generative committor-guided path-sampling framework reconstructs rare biomolecular transition pathways and reveals the underlying thermodynamics and kinetics without using predefined collective variables or brute-force sampling, at an acceptable computational cost.