Integrating Interfacial Ligand Tailoring and Machine Learning for Heterogeneous H2-Driven Chemoenzymatic Synthesis

Enzyme cofactor-mediated biochemical cascades in living cells showcase nature’s capacity for efficient, selective multistep transformations, inspiring artificial chemoenzymatic systems that merge enzymatic specificity with the versatility of synthetic catalysts via the cofactor. Practical implementation of such systems is, however, constrained by poor selectivity, enzyme incompatibility, and limited turnover in nonenzymatic in situ cofactor regeneration. Here, molecular ligand-directed interfacial tailoring of supported Pt catalysts is shown to overcome these challenges, achieving >99% regioselective NADH regeneration from NAD+ with H2 as a clean reductant, while enhancing the activity and stability. Machine learning-guided analysis of a structurally diverse ligand library, validated experimentally, identifies the surface Pt2+ fraction as a robust, predictive electronic descriptor for selectivity classification across catalyst variants. Integration of the ligand-optimized catalyst with an NADH-dependent dehydrogenase delivers asymmetric hydrogenations with >99% enantiomeric excess and benchmark in situ NADH turnover numbers. This study demonstrates an H2-driven chemoenzymatic platform that bridges heterogeneous metal catalysis and enzymatic catalysis, offering immediate relevance to a wide range of applications, including fine chemical synthesis and artificial photosynthesis.