Accelerating Reaction Generalization through Domain-Specific Transfer Learning

Le résumé fourni par la source

Generality in synthetic chemistry is often equated with a single set of reaction conditions that performs uniformly across diverse substrates. In practice, however, widely adopted transformations achieve generality not through a universal protocol, but through a wellcharacterized reaction space that enables rapid substrate-specific optimization. Here we report a machine-learning-guided strategy that accelerates this generalization process through active transfer learning. Bayesian optimization was used to generate a compact dataset of 120 experiments that maps the reaction landscape for the decarboxylative olefination of malonic acid derivatives. This domain-specific dataset provides transferable prior knowledge that enables the rapid identification of optimal conditions for new coupling partners, including previously unseen substrates. The method extends across thioesters, esters, and amides, providing a mild and atomeconomical alternative to classical phosphonate-based olefinations. More broadly, this work shows how small, information-rich datasets coupled with accessible machine learning tools can enable practical reaction generality and accelerate the translation of new reactions into routine synthetic use.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.