Machine learning-guided engineering of chalcone synthase enables high-selectivity phloretin biosynthesis in yeast

Abstract

Flavonoids are a diverse class of plant secondary metabolites with broad applications in pharmaceuticals, nutraceuticals, and cosmetics. However, microbial production of flavonoids with high selectivity remains challenging due to substrate promiscuity of chalcone synthase (CHS), a key biosynthetic enzyme. Despite advances in metabolic engineering, enzyme design for substantial improvements in CHS activity and selectivity remains limited. Here, we construct a library of 243 CHS variants and apply machine learning to guide CHS engineering. According to the model prediction, we identified a triple mutant EbCHS that markedly improved product selectivity by 10-fold. When producing the phloretin in Saccharomyces cerevisiae using p-coumaric acid as the substrate, the EbCHS mutant increases the titer by 2.14-fold. Combining the multiple enzyme engineering approach, the titer of phloretin reach 132.85 mg L⁻¹. Structural analysis revealed that the mutations reshaped the active site and improved substrate binding, collectively enhancing both catalytic efficiency and product selectivity. This work demonstrates the potential of machine learning-guided enzyme engineering and provides a generalizable framework for optimizing biosynthetic enzymes toward the selective microbial production of high-value natural products.