Ancestral Sequence\nReconstruction Meets Machine Learning:\nEne Reductase Thermostabilization Yields Enzymes with Improved Reactivity\nProfiles

Ene reductases (EREDs) are enzymes that catalyze the\nasymmetric\nreduction of CC bonds. EREDs are potentially useful in the\nlarge-scale synthesis of pharmaceutical compounds, but their application\nas biocatalysts is limited because they are often unstable under process\nconditions. Previous work addressed this limitation by identifying\nstabilized EREDs with ancestral sequence reconstruction (ASR), a bioinformatic\nmethod that predicts evolutionary ancestors based on a set of homologous\nsequences. In this work, we sought to apply ASR to design enzyme libraries\nand leverage machine learning to predict the most stable library variants.\nWe generated an ERED library that targeted residues based on uncertainty\nin the ASR prediction. Screening data from a portion of the library\nwere used to build a machine learning model that could accurately\npredict variants with improved thermostability. The most stabilized\nenzyme outperformed the wild-type and ancestral parent enzymes under\nprocess-like conditions with a panel of substrates. We envision that\nthe combination of ASR and machine learning could be generally applied\nto other classes of enzymes, facilitating the development of high-quality\nindustrial biocatalysts.