2025/11/14 by Pavel A. Dub, Thomas P. Hughes, Thomas J. L. Mustard · 1 voice
Chemical Engineering · Chemistry · Materials Science · #Asymmetric Hydrogenation and Catalysis #Catalysis and Oxidation Reactions #Machine Learning in Materials Science
paper · doi:10.26434/chemrxiv-2025-mm69d
openalex created_date 2025/11/14 · openalex publication_date 2025/11/14 · openalex updated_date 2026/07/14
Identifying molecular catalysts that simultaneously exhibit high selectivity, high turnover frequency, and robust stability remains a major challenge in homogeneous catalysis. Traditionally, catalyst discovery has relied on time-consuming and expensive experimental trial-and-error approaches. Here, we describe a software framework for homogeneous catalyst design based on high-throughput virtual screening of reaction networks. Our Reaction Network Enumeration Profiler (RxnEnumProfiler) workflow automatically enumerates a reference catalytic reaction network based on a user-defined library for enumeration and performs automated calculations and visualizations of the corresponding Free Energy Profiles (FEPs). Depending on the required accuracy and cost, calculations can be performed using quantum mechanical, semiempirical tight-binding, and/or machine-learning potentials. Key metrics relevant to catalyst design are extracted directly from the computed FEPs, enabling automated mechanism-enabled catalyst optimization. We demonstrate this approach across three real-world homogeneous catalytic reactions: organocatalyzed asymmetric hydrogenation, Pd-catalyzed C–O cross-coupling, and ansa-metallocene-catalyzed isotactic propylene polymerization. Beyond homogeneous catalysis, RxnEnumProfiler can be used for non-catalytic transformations to optimize reactivity