Sequential optimization of multivariate metal–organic framework based biocatalysis

Efficient biocascades require integrated optimization of both biocatalysts and reaction conditions. In the present study, we employ a Latin hypercube sampling–coupled Bayesian optimization (LHS-BO) workflow to sequentially optimizes multivariate Zr-based E-MOF (E-MOF = enzyme@metal–organic framework biocomposite) design and the downstream glucose oxidase–horseradish peroxidase (GOx–HRP) biocascade. Enzymatic assays, attenuated total reflectance Fourier transform infrared spectroscopic (ATR-FTIR) and ultraviolet-visible (UV-Vis) spectroscopic analyses confirm that the optimized multivariate E-MOFs, ZG67, and ZH16, stabilize GOx and HRP in bioactive conformations. ZG67 and ZH16 achieve high encapsulation efficiency (92.2% ± 0.7% and 90.2% ± 1.3%), retained activity (87.1% ± 2.2% and 102.9% ± 14.9%), and enhanced stability under thermal (51.3% ± 1.4% and 45.9% ± 8.7%) and solvent treatments (>80% and ≈60% residual activity, respectively, in ethanol or acetonitrile). The optimal GOx–HRP cascade condition identified by LHS-BO (R49) achieves over 95% of the theoretical maximum 2,3-diaminophenazine (DAP) production rate predicted by the microkinetic model. The strong agreement amongst the experimental data, machine learning modeling, and kinetic modeling validates the robustness of this sequential E-MOF optimization framework. This generalized LHS-BO strategy provides a powerful tool for rational E-MOF and multi-enzyme cascade design, advancing biocatalysis and reaction engineering.