Data-Driven Interrogation of Reactivity in Acid-Catalyzed Carbonyl–Olefin Metathesis with Machine Learning and Large Language Models
- 15 May 2026
- journal article
- research article
- Published by American Chemical Society (ACS) in Journal of the American Chemical Society
- Vol. 148 (21) , 21545-21557
Abstract
Carbonyl–olefin metathesis (COM) has emerged as a powerful yet mechanistically complex transformation for forging carbon–carbon bonds. Although diverse Brønsted and Lewis acid catalysts enable COM reactivity, predicting which catalyst will be effective for a given substrate remains challenging. Likewise, machine learning (ML) and large language models (LLMs) are emerging tools for planning experiments in organic synthesis, but their relative strengths in mechanistically rich catalytic systems like COM remain unclear. Here we address both challenges through a practical ML framework that combines a rigorously curated, blinded data set of 147 COM reactions with tiered predictive models to guide catalyst and substrate selection. A Morgan-fingerprint baseline enables rapid structure-based screening, while quantum-derived descriptor models achieve higher accuracy (R2 ≈ 0.92) and retain performance on external substrates; feature-importance and SHAP analyses identify catalyst HOMO energy, dimerization propensity, and substrate carbonyl polarization as dominant drivers of reactivity. Reactivity-cliff analysis distinguishes smoothly interpolating catalysts, such as FeCl3 and ZnCl2, from cliff-rich ion-pair systems, identifying regions where predictions are intrinsically unreliable and targeted experiments are most informative. Variance-guided Bayesian optimization recovers most full-model performance using only ∼40% of the reaction matrix, while a parallel GPT-4o-guided, code-free acquisition protocol offers a lower-barrier alternative that outperforms random sampling and, in this setting, exceeds expert selection. Together, these results establish a mechanistically interpretable, data-efficient framework for predicting, explaining, and prioritizing COM experiments in a controlled reaction matrix, and show how uncertainty-aware ML and LLM-guided selection can augment decision-making in mechanistically related COM systems.Keywords
All Related Versions
Funding Information
- Canadian Institutes of Health Research
- Canada Foundation for Innovation
- Research Corporation for Science Advancement (SA-AUT-2024-022a)
- China Scholarship Council
- Natural Sciences and Engineering Research Council of Canada
- Canada Excellence Research Chairs, Government of Canada
- University of British Columbia
- Acceleration Consortium, University of Toronto
This publication has 48 references indexed in Scilit:
- Exposing the Limitations of Molecular Machine Learning with Activity CliffsJournal of Chemical Information and Modeling, 2022
- Brønsted Acid-Catalyzed Carbonyl–Olefin Metathesis: Synthesis of Phenanthrenes via Phosphomolybdic Acid as a CatalystThe Journal of Organic Chemistry, 2021
- Data Science Meets Physical Organic ChemistryAccounts of Chemical Research, 2021
- Carbonyl–Olefin MetathesisChemical Reviews, 2021
- Brønsted-Acid-Catalyzed Intramolecular Carbonyl–Olefin Reactions: Interrupted Metathesis vs Carbonyl-Ene ReactionThe Journal of Organic Chemistry, 2021
- Models for Understanding Divergent Reactivity in Lewis Acid-Catalyzed Transformations of Carbonyls and OlefinsACS Catalysis, 2020
- Synthesis of Medium-Sized Carbocycles by Gallium-Catalyzed Tandem Carbonyl–Olefin Metathesis/Transfer HydrogenationOrganic Letters, 2019
- FeCl3‐Catalyzed Ring‐Closing Carbonyl–Olefin MetathesisAngewandte Chemie International Edition, 2016
- NBO 6.0: Natural bond orbital analysis programJournal of Computational Chemistry, 2013
- Energy-adjustedab initio pseudopotentials for the second and third row transition elementsTheoretical Chemistry Accounts, 1990