Machine Learning Reveals In-Cavity Versus Surface Activity for Selective C─H Borylation by Metal-Organic Framework...

Metal-organic frameworks (MOFs) provide an expansive and tunable platform for heterogeneous catalysis, yet distinguishing between catalytic reactions occurring within their pores and those on their external surfaces remains a challenge. This study employs interpretable machine learning to elucidate structure-activity relationships in MOF-supported nickel (Ni) catalysts for selective sp³ and sp² C─H borylation. By analyzing over 470 000 MOF structures, we developed a set of 45 concise and chemically meaningful descriptors that capture key structural variations across MOFs. These descriptors enabled us to identify the critical factors governing sp³ versus sp² selectivity, revealing distinct activation mechanisms: sp³ C─H borylation preferentially occurs within MOF cavities via a radical-mediated hydrogen atom transfer (HAT) mechanism, whereas sp² C─H borylation is associated with surface or defect sites, favoring a concerted metalation-deprotonation (CMD) pathway. Guided by these insights, we designed Ni catalysts that achieve up to 97.8% sp³ selectivity and 88.7% sp² selectivity. This work provides a systematic framework for rational catalyst design and establishes generalizable principles for controlling activity preference in MOF-supported catalysis.