Machine Learning\nOptimizing Enzyme/ZIF Biocomposites\nfor Enhanced Encapsulation Efficiency and Bioactivity

In this study, we present the first example of using\na machine\nlearning (ML)-assisted design strategy to optimize the synthesis formulation\nof enzyme/ZIFs (zeolitic imidazolate framework) for enhanced performance.\nGlucose oxidase (GOx) and horseradish peroxidase (HRP) were chosen\nas model enzymes, while Zn(eIM)<sub>2</sub> (eIM = 2-ethylimidazolate)\nwas selected as the model ZIF to test our ML-assisted workflow paradigm.\nThrough an iterative ML-driven training-design-synthesis-measurement\nworkflow, we efficiently discovered GOx/ZIF (G151) and HRP/ZIF (H150)\nwith their overall performance index (OPI) values (OPI represents\nthe product of encapsulation efficiency (<i>E</i> in %),\nretained enzymatic activity (<i>A</i> in %), and thermal\nstability (<i>T</i> in %)) at least 1.3 times higher than\nthose in systematic seed data studies. Furthermore, advanced statistical\nmethods derived from the trained random forest model qualitatively\nand quantitatively reveal the relationship among synthesis, structure,\nand performance in the enzyme/ZIF system, offering valuable guidance\nfor future studies on enzyme/ZIFs. Overall, our proposed ML-assisted\ndesign strategy holds promise for accelerating the development of\nenzyme/ZIFs and other enzyme immobilization systems for biocatalysis\napplications and beyond, including drug delivery and sensing, among\nothers.