Bayesian Optimization-Assisted Engineering of Formate\nDehydrogenase Encapsulation in Multivariate Zeolitic Imidazolate Framework

Engineering\nof multivariate zeolitic imidazolate frameworks (mZIFs)\noffers substantial potential for optimizing enzyme encapsulation by\nenhancing encapsulation efficiency (EE), enzyme loading capacity (<i>P</i><sub>loading</sub>), retained enzymatic activity (REA),\nand protection. However, this area remains underexplored. In this\nstudy, we rationally employed three imidazole-based ligands with distinct\nfunctionalitiesHeIM (2-ethylimidazole), HTz (1,2,4-triazole),\nand HIM (1-(2-hydroxyethyl)imidazole)to fine-tune hydrophobicity\nand defect simultaneously within FDH@mZIF (FDH = formate dehydrogenase).\nLeveraging an iterative Bayesian optimization-assisted training-design-synthesis-measurement\nworkflow, we efficiently identified F190 as the best FDH@mZIF, achieving\nEE = 89.3%, REA = 14.9%, and <i>P</i><sub>loading</sub> =\n30.3 wt%. This establishes F190 as the leading FDH-based biocatalyst\nin the literature. The optimal FDH-mZIF interactions in F190 were\nreflected by minimal structural perturbation of encapsulated FDH,\nas evidenced by the ATR-FTIR and fluorescence studies. Additionally,\nF190 can effectively safeguard the encapsulated FDH against thermal\nand proteolytic degradation and catalyze CO<sub>2</sub>-to-formate\nconversion while maintaining activity for at least five cycles without\nsignificant activity loss.