Decoding Regioselectivity in Cu-Catalyzed Borylation of Alkynes: Insights from Machine Learning and Artificial Intelligence

CuI-catalyzed hydroboration of alkynes is a cost-effective route to trans-alkenyl boronates (valuable intermediates for C–C cross-coupling) but controlling regioselectivity in these reactions is challenging. Here, we address this challenge by developing a machine learning model, using high-throughput computational-derived descriptors and a combined experimental/literature data set to predict regioselectivity (expressed as ΔΔG‡). The resulting support vector regression (SVR) model achieved high accuracy (cross-validated R2 > 0.8, RMSE ∼0.4–0.6 kcal/mol) and revealed mechanistically relevant trends through feature importance analysis. For example, the model rationalizes why certain N-donor ligands that fail to promote Cu-catalyzed hydroboration can effectively catalyze alkyne hydrosilylation, linking this divergence to differences in geometric ligand descriptors. Moreover, ML-guided screening identified promising ligands that improved hydroboration outcomes (increased yield and regioselectivity), as confirmed in experiments (e.g., enabling reduced catalyst loading without sacrificing selectivity). Overall, this integrated ML strategy offers a powerful tool for understanding and predicting regioselectivity in CuI-mediated reactions and, with appropriate calibration, could be extended to other organocopper systems.