Breaking the Timescale Barrier: Generative Discovery of Conformational Free-Energy Landscapes and Transition Pathways

TL;DR

Gen-COMPAS is introduced, a generative committor-guided path sampling framework that reconstructs transition pathways without predefined variables and at a fraction of the cost, uniting machine learning and molecular dynamics for broad mechanistic and practical insight.

Abstract

Molecular transitions -- such as protein folding, allostery, and membrane transport -- are central to biology yet remain notoriously difficult to simulate. Their intrinsic rarity pushes them beyond reach of standard molecular dynamics, while enhanced-sampling methods are costly and often depend on arbitrary variables that bias outcomes. We introduce Gen-COMPAS, a generative committor-guided path sampling framework that reconstructs transition pathways without predefined variables and at a fraction of the cost. Gen-COMPAS couples a generative diffusion model, which produces physically realistic intermediates, with committor-based filtering to pinpoint transition states. Short unbiased simulations from these intermediates rapidly yield full transition-path ensembles that converge within nanoseconds, where conventional methods require orders of magnitude more sampling. Applied to systems from a miniprotein to a ribose-binding protein to a mitochondrial carrier, Gen-COMPAS retrieves committors, transition states, and free-energy landscapes efficiently, uniting machine learning and molecular dynamics for broad mechanistic and practical insight.

Figures (4)

• Figure 1: Gen-COMPAS in a nutshell(A) Overall framework of Gen-COMPAS. (B) Denoising diffusion model: Training and inferring to find the intermediates. (C) Obtaining physically meaningful structures corresponding to transition states using targeted MD (TMD). (D) Overall sampling strategy between two metastable states of Gen-COMPAS guided by the committor ($q$) in a schematic free-energy landscape.
• Figure 2: Gen-COMPAS of Trp-cage fast-folding protein(A) Representative structures of Trp-cage in the folded, transition and unfolded states. The $\alpha$-helix and the end-to-end distance, d, are highlighted. (B) Learned committor ($q$) projected onto the distance root mean square deviation (RMSD) with respect to all the C$_{\alpha}$ atoms, the RMSD with respect to the C$_{\alpha}$ atoms of the $\alpha$-helix, and the end-to-end distance. The two sampled pathways are also depicted. (C) Free-energy landscape projected onto the same collective variables (CVs) obtained by Gen-COMPAS and by using the DESRES simulations Lindorff-Larsen2011fold(D) Normalized values of the CVs and corresponding committor values along the pathways.
• Figure 3: Gen-COMPAS of RBP binding-upon-folding process(A) Representative structures of RBP-ribose unbound state and bound state, the transition states, and pathways for the RBP binding-upon-folding process. The three collective variables describing the process: d, hinge, and twist ravindranathan_ribose_2005. (B) The committor and committor-consistent pathways revealing the two distinct binding-folding mechanism by Gen-COMPAS. (C) Free-energy landscapes obtained by Gen-COMPAS showing the binding-upon-folding mechanism.
• Figure 4: Gen-COMPAS of the mitochondrial ATP/ADP carrier (AAC)(A) The three meta-stable states, transition states and transition pathway found by Gen-COMPAS from C-state to M-state through O-state of holo-AAC (ADP$^{3-}$-bound). The pivoting motion of ADP$^{3-}$ from O-state to M-state. The conformational transition pathway C $\xrightarrow{}$ O $\xrightarrow{}$ M as well as the basins of the free-energy landscape (FEL) in the three-dimensional CV space. (B) Holo-state of AAC in membrane and the collective variables describing the AAC conformational transition: d1, d2, and d3. (C) Committor and committor-consistant pathway connecting three states and (D) the FEL projected onto two dimensions. (E) The two metastable states of apo-AAC and its FEL projected onto d1 and d2.