Overcoming Data Starvation: Automated Virtual Reaction Exploration and Machine Learning Discovery of <i>p</i> ‐Block Metal Catalysts for Borrowing Hydrogen

Abstract

ABSTRACT (De)Hydrogenation processes, traditionally dominated by d ‐block transition metals, offer a sustainable route for molecular synthesis using alcohols as feedstocks. However, reliance on noble metals, mechanistic complexity, and limited substrate scope drive the search for alternatives. p ‐Block metals represent an attractive but long‐standing challenge due to their strong oxophilicity and Lewis acidity. In particular, the lack of d ‐orbitals for electronic buffering impedes catalytic dehydrogenation/hydrogenation cycling and confines p ‐block hydrides to stoichiometric use. To overcome these intrinsic limitations and data starvation for AI‐driven design, we present an intelligent framework that bypasses high‐throughput experimentation (HTE) by integrating automated reaction pathway exploration with machine learning (ML). This approach enables de novo discovery of p ‐block catalysts in data‐scarce regimes. We demonstrate its power by developing a homogeneous indium‐based catalyst for borrowing hydrogen (BH)‐mediated N ‐alkylation, featuring broad substrate scope, operational simplicity, and synthetic accessibility. The catalytically active indium‐hydride (In–H) species was confirmed by in situ 1 H NMR. This work not only establishes the first efficient p ‐block BH catalyst but also introduces a mechanism‐informed, artificial intelligence (AI)‐guided paradigm for main‐group catalysis, expanding the frontiers of catalysis and sustainable synthesis.