Programming 2,3-butanediol biosynthesis in Saccharomyces cerevisiae with AI-optimized dual-optogenetic switches

Microbial cell factories promise sustainable biomanufacturing, yet balancing cellular growth and product synthesis remains a fundamental productivity bottleneck. While traditional static metabolic engineering fails to precisely orchestrate metabolic flux across temporal and spatial dimensions, optogenetic approaches offer unprecedented spatiotemporal control and programmability. This study developed an integrated framework combining dual-mode optogenetic switches with a machine learning-guided optimization strategy to achieve precise temporal orchestration of metabolic pathways. Using Saccharomyces cerevisiae (strain BY4742, which exhibits minimal endogenous 2,3-butanediol (2,3-BDO) production) as the model host, the 2,3-BDO biosynthetic pathway was constructed. The deployment of dual light-responsive modules for protein degradation and stabilization, together with systematic optimization of illumination conditions, enabled dynamic coordination of the growth and production phases, increasing the 2,3-BDO titre to 3.52 times that of the static control in shake flasks. To extend beyond empirical optimization, a convolutional neural network (CNN) was integrated with a differential evolution (DE) algorithm to identify the optimal dynamic illumination schedule, resulting in a further 15.60% increase in the predicted and experimentally validated 2,3-BDO titre. The optimized strategy was subsequently validated in a 5-L photobioreactor, where dynamic optogenetic regulation combined with optimized fermentation conditions yielded 23.95 g L^-1 2,3-BDO under fed-batch conditions. This study establishes a machine learning-assisted optogenetic framework for dynamic metabolic regulation and provides an effective strategy for improving microbial biomanufacturing.