Exploring Enzyme-Mimicking Metal-Organic Frameworks for CO2 Conversion through Vibrational Spectra-Based Machine Learning.

Abstract

In pursuing the benefits of natural enzyme catalysts while overcoming their limitations, we find metal-organic frameworks (MOFs), renowned for their highly tunable functionalities, stand out in biomimetic applications. We used unsupervised machine learning on density functional theory-computed vibrational infrared and Raman spectral features to screen 300 Zn-MOFs for CO2 conversion, similar to carbonic anhydrase (CA). Our findings confirmed that MOFs with spectroscopic attributes closely resembling those of CA hold the potential for replicating CA's electronic and catalytic properties. Unlike previous studies that relied on heuristic or trial-and-error methods and focused on geometric configurations, our research uses vibrational spectral features to explore structure-property relationships, making them more accessible through spectroscopy. Moreover, we highlight vibrational spectral features as efficient carriers for highly dimensional chemical information, enabling the simultaneous optimization of multiple performance parameters. These findings pave the way for pioneering designs of enzyme-mimetic MOFs and concurrently expand the application scope of spectroscopic tools in biomimetic catalysis.