Data-Driven, MechanisticallyGuided Prediction ofYield and Chemoselectivity in SuFEx Reactions

Sulfur(VI) fluoride exchange (SuFEx) has emerged as a powerful click reaction for constructing diverse S(VI)-based linkages across chemical biology, materials science, and drug discovery. However, the diversity of SuFEx hubs and native nucleophiles generates a high-dimensional reaction landscape in which yields and chemoselectivities are not always straightforward to anticipate, limiting its predictability in synthetic planning. Here, we demonstrate that both SuFEx yield and chemoselectivity can be rendered predictable by integrating machine learning with curated reaction data and mechanistic insight. Using large-language-model-enabled agents, we constructed a comprehensive literature-derived data set encompassing successful, low-yielding, and failed SuFEx reactions. Models trained on this data set provide quantitative estimates of reaction yields and enable qualitative forecasting of chemoselectivities across diverse SuFEx hubs, nucleophiles, and additives. DFT calculations reveal how SuFEx-hub electrophilicity, nucleophile acidity, and additive basicity cooperatively govern chemoselectivity, leading to a mechanistically transparent framework for chemoselectivity control. Deployment as a publicly accessible platform, SuFExPredictor, enables model-guided reaction optimization, recovery of unproductive transformations, and late-stage functionalization of drug-relevant molecules.