Computational framework for the mechanistic studies and physics-informed prediction and design of photoenzymes - a robust enzyme design strategy will profoundly impact human health, such as by…

Computational framework for the mechanistic studies and physics-informed prediction and design of photoenzymes - a robust enzyme design strategy will profoundly impact human health, such as by achieving asymmetric synthesis of pharmaceuticals not easy to be achieved by small-molecule catalysts. However, photoexcitation has rarely been considered in enzyme design. In recent years, IT has been found that certain enzymes can be repurposed by photoexcitation for non-natural chemical reactions that cannot be easily achieved by small- molecule catalysts or traditionally engineered enzymes. These enzymes, termed photoenzymes, are believed to utilize photoactivatable cofactors, combined with a natural or mutated enzyme scaffold, to reach new reaction spaces. Photoenzymes have emerged as a promising new class of catalysts for non-natural reactions important in pharmaceutical synthesis, such as asymmetric radical reactions important in late-stage functionalization of drug-like molecules. However, the mechanisms of photoenzymes have not been studied well and there has not been a clear rational discovery and design strategy for photoenzymes, not only because these are emergent systems, but also because existing computational methods are not adequate. In this research program, we will develop an integrated computational framework to predict the combined effect of light, cofactor, substrate(s), and protein sequences on photoenzyme reactivity and the mechanisms that lead to this effect, and will develop a physics-informed design strategy that makes use of descriptors derived from both ground and excited electronic states to control the activity and selectivity of photoenzymatic reactions. This will fill the GAP in computational enzyme design where the excited electronic states are not normally considered. In specific, we will 1) develop machine learning-enhanced simulation methods to efficiently simulate both the ground and excited electronic states of photoenzymes to assist mechanistic studies and to inform the prediction and design of photoenzymes, and 2) develop a photoenzyme design strategy centered on descriptors derived from both ground and excited electronic states computed by molecular simulations. We will use flavin-dependent "ENE"- reductases (ereds) as the prototype system for the computational tool development and testing since there have already been a collection of computational and experimental data for ereds, where the computational data are from the PI's group. This research program will not only deepen our understanding of photoenzyme mechanisms, but will also greatly facilitate the design and prediction of photoenzymes for non-natural reactions important in pharmaceutical synthesis. In the long term, IT will also facilitate the identification of natural enzymes that may have previously unknown photo-driven reactivity, which may become new protein scaffolds for developing novel photoenzymatic reactions or become new drug targets.