Machine learning-assisted protein engineering for improving stereoselectivity

Biocatalysis is a promising approach to asymmetric synthesis; however, the natural substrate specificity of enzymes often limits their stereoselectivity, and thus, protein engineering is essential to improving enzyme performance. This perspective summarizes machine learning-assisted protein engineering for stereoselectivity, focusing on supervised learning models trained on experimental data to uncover correlations between enzyme/substrate descriptors and stereoselectivity. This approach can provide relatively accurate predictions at low computational cost, thereby improving or reversing enzyme stereoselectivity. Despite these advances, challenges remain, such as the lack of reliable stereoselectivity data and limited predictive performance and generalization ability of models. The integration of large amounts of high-quality data, more accurate structural and physicochemical descriptors, and innovative algorithms holds the promise of developing more robust and generalizable models that can predict the stereoselectivity of a wide range of enzymes and substrates. This approach could pave the way for more efficient and sustainable biocatalytic processes in asymmetric synthesis.