B
B9
Text
Explainable Machine Learning for the Design of Boron-Based Lewis Acids
Builds interpretable machine-learning models that predict the fluoride ion affinity (a proxy for Lewis acidity) of boron Lewis acids from chemically meaningful descriptors: ab initio computed features and Hammett substituent parameters, evaluated on four constrained molecular scaffolds. The white-box models reach mean absolute error below 6 kJ/mol and outperform black-box deep networks in the low-data regime, and their explainability analysis identifies actionable substituent levers for modulating Lewis acidity, bridging ML with a chemist's reasoning about reactivity.