ML-Guided Engineering to Enhance Cell-Organelle Capacity for Multi-Enzyme Pathway Compartmentalization
Abstract
Repurposing organelles for specialized metabolic functions offers a powerful approach for optimizing metabolic engineering in fermentable organisms such as Saccharomyces cerevisiae. Peroxisomes, in particular, provide an attractive target for re-engineering, as they are dispensable for yeast viability under glucose growth conditions and offer the ability to compartmentalize heterologous enzymes, thus separating native cytosolic metabolism from engineered peroxisomal pathways. Despite this potential, peroxisomes are repressed when not required, limiting their functional capacity for heterologous protein. To address this limitation, we identified 25 peroxisome-related genes with the potential to enhance capacity. Testing all gene combinations in either overexpression or normal states creates a vast combinatorial space, posing a significant challenge due to limited experimental resources. To overcome this, we implemented a machine learning-based strategy that integrates experimental testing with in silico predictions, allowing us to iteratively refine and identify the optimal gene combination to maximize peroxisome capacity. The optimal combination ends up achieving a 137% enhancement in peroxisome functional capacity relative to the wild-type strain. This improvement facilitated the effective compartmentalization of the eight-enzyme pathway to biosynthesize the monoterpene geraniol, enabling an 80% increase in geraniol titers, representing the highest titer yet reported.