Machine-Learning-Aided\nEngineering Hemoglobin as Carbene\nTransferase for Catalyzing Enantioselective Olefin Cyclopropanation

In this study, we developed a machine-learning-aided\nprotein design\nstrategy for engineering <i>Vitreoscilla</i> hemoglobin\n(VHb) as carbene transferase. A Natural Language Processing (NLP)\nmodel was used for the first time to construct an algorithm (EESP,\nenzyme enantioselectivity score predictor) and predict the enantioselectivity\nof VHb. We identified critical amino acid residue sites by molecular\ndocking and established a simplified mutation library by site-saturated\nmutagenesis. Based on the simplified mutant library, the trianed EESP\nscored 160,000 virtual mutants, and 15 predicted high-score mutants\nwere chosen for experimental validation. Among these mutants, VHb-WK\n(Y29W/P54K) demonstrated the highest diastereoselectivity and enantioselectivity\nof carbene transferase for the olefin cyclopropanation in aqueous\nconditions. Subsequently, molecular dynamics simulations were performed\nto explore the interaction between protein and substrates, finding\nthat the high enantioselectivity of VHb-WK stems from the interactions\nof R47, Q53, and K84, which narrows the entrance of the enzyme’s\npocket, favoring the restriction of the formation of reaction intermediates.\nIntegrating the NLP model and enzyme modification offers significant\nadvantages by reducing economic costs and workloads associated with\nthe protein engineering process.