Machine Learning-Guided Cofactor Engineering for Tunable Azaphilone Component Ratios in Monascus spp.

Abstract Monascus azaphilones (MAs) are bioactive compounds that have been utilized in food for over a millennium. These molecules are classified into orange (OMAs), yellow (YMAs), and red MAs (RMAs), with compositional ratios determining the product quality. Cofactors critically influence component ratios, but their individual roles in controlling MA ratios remain unclear. This study established a metabolic library by chemically perturbing the cofactor levels in M. purpureus HJ11. Machine learning analysis of cofactor concentrations and MA components identified NADH as the regulator driving YMAs biosynthesis while suppressing OMAs and RMAs. To validate this, we engineered a cofactor supply system by replacing native isocitrate dehydrogenase with Pseudomonas stutzeri phosphite dehydrogenase, enabling NADH control via an isomaltose adjustment. The engineered strain produced azaphilones with purities of 91% (YMA), 93% (OMA), and 87% (RMA) under optimized protocols. These findings reveal the regulatory role of cellular redox balance in azaphilone diversity and establish a programmable framework for metabolite composition control in Monascus species.