Machine learning-driven optimization of metabolic balance for β-carotene production

Balancing metabolic pathways is critical for engineering microbial platforms to efficiently and robustly synthesize value-added bioproducts. In the oleaginous yeast Yarrowia lipolytica engineered for β-carotene production, lipid synthesis supports carotenoid storage but also competes with carotenoid synthesis for cellular resources, necessitating precise regulation for optimal resource allocation. In this study, we establish a machine learning framework that captures the complex interactions among three key metabolic modules for β-carotene synthesis: the mevalonate pathway (precursor supply for β-carotene), lipid synthesis (storage capacity), and the β-carotene synthetic cluster (product formation). This computational framework enables the prediction of β-carotene output based on gene combinations and guides iterative gene integration strategies across these interconnected pathways to optimize production. Using this approach, the best-performing strain YLT226 achieved a 7-fold increase in β-carotene titer compared to the initial strain YLT001 through nine rounds of guided gene integration. This work provides a promising strategy for understanding and engineering metabolic flux distributions.