Heme proteins such as cytochromes P450s are known for their excellent regio-, chemo-, and stereoselectivity in oxene-transfer reactions. Over the past decade, heme proteins have been engineered to catalyze several unnatural carbene and nitrene transfer reactions such as cyclopropanation and C-H amination. In Chapter 1, we engineer cytochrome P450 variants to catalyze stereoselective atom-transfer radical cyclizations. Controlling the stereochemistry of these free-radical processes in traditional synthetic systems is challenging due to the lack of stereoinduction strategies. However, in this biocatalytic platform, the stereoselectivity of the radical addition and halogen rebound steps is controlled by the enzyme scaffold, enabling precise control over the reactive radical.
Directed evolution has proven to be a powerful tool for improving various enzyme properties such as stability, selectivity, and activity, but its efficiency in navigating the fitness landscape is limited. In Chapter 2, we advance machine learning (ML) algorithms to develop synthetically important enzyme functions not found in nature, such as the new-to-nature reactivity discussed in Chapter 1. Our algorithm, MODIFY, significantly accelerates enzyme engineering campaigns by designing high quality starting libraries that optimize both diversity and fitness, reducing the occurrence of non-expressing variants from the screening pool.
Overall, this work expands the scope of the biocatalytic metalloenzyme platform to control radical intermediates for asymmetric catalysis and demonstrates the potential of ML algorithms in providing superior starting points for protein engineering campaigns.