A Machine-Learned Symbolic Committor for a Chemical Reaction: Isomerization in Retinal and p -Coumaric Acid
Abstract The thermal cis-trans isomerization of C═C double bonds in chromophores is a prototypical high-barrier reaction whose mechanism hinges on subtle out-of-plane bending motions. Combining machine-learning-guided path sampling (AIMMD) with symbolic regression, we learn an interpretable committor for two chemically distinct chromophores, N-retinylidene-lysine and p-coumaric acid anion, in vacuo from unbiased molecular dynamics trajectories generated by two-way shooting. Parametrizing the committor with a neural network and sampling it uniformly across the full range allows the model to resolve the reaction coordinate across the full transition region, not only at the isocommittor surface pB(x) = 0.5. A feature-attribution analysis identifies four proper dihedrals around the reactive bond as the informative coordinates, while the improper dihedrals at the sp2-hybridized carbon atoms prove unsuitable because reactant, transition, and product states share the same values. Symbolic regression then distills the committor into compact analytical expressions─to our knowledge the first committor both learned and distilled into closed form for a chemical reaction─and shows that a coupling of all four dihedrals is required to reproduce the S-shaped, stepwise pathway seen in the transition-path ensemble. This S-shape is absent from the minimum free-energy path: it arises from the dynamics of the short transition events combined with the mass asymmetry between heavy-atom and hydrogen-bearing dihedrals, and the larger asymmetry of p-coumaric acid yields the more pronounced S-shape this picture predicts. An interpretable, machine-learned committor, though a function of configuration alone, thus exposes dynamical features of the mechanism to which the free-energy surface is blind. The workflow requires no prior assumptions about the reaction coordinate and extends naturally to other high-barrier isomerizations and, combined with a reactive potential, opens a path toward bond-breaking chemical reactions.
Authors
- Gianluca Lattanzi (ORCID: https://orcid.org/0000-0002-0808-6457)
- Gianmarco Lazzeri (ORCID: https://orcid.org/0000-0002-0593-5572)
- Kai Töpfer (ORCID: https://orcid.org/0000-0002-4650-9641)
- Roberto Covino (ORCID: https://orcid.org/0000-0003-0884-0402)
- Bettina G. Keller (ORCID: https://orcid.org/0000-0002-7051-0888)
- Florian Renner
- Christopher J.I. Bläß
- Vittoria Ossanna (ORCID: https://orcid.org/0009-0008-6546-374X)
Institutions
- Goethe University Frankfurt (DE)
- University of Trento (IT)
- Frankfurt Institute for Advanced Studies (DE)
- Freie Universität Berlin (DE)
Publication Details
- Journal
- Journal of Chemical Theory and Computation
- Published
- 2026-10-30
- DOI
- https://doi.org/10.1021/acs.jctc.6c01496
- Primary Topic
- Machine Learning in Materials Science
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Deutsche Forschungsgemeinschaft
- Volkswagen Foundation
- Goethe-Universität Frankfurt am Main
- Freie Universität Berlin
- Johanna Quandt Stiftung
- Frankfurt Institute for Advanced Studies