Yeast-MetaTwin for Systematically Exploring Yeast Metabolism through Retrobiosynthesis and Deep Learning

Abstract Underground metabolism plays a crucial role in understanding enzyme promiscuity, cellular metabolism, and biological evolution, yet experimental exploration of underground metabolism is often sparse. Even though yeast genome-scale metabolic models have been reconstructed and curated for over 20 years, more than 90% of the yeast metabolome is still not covered by these models. To address this gap, we have developed a workflow based on retrobiosynthesis and deep learning methods to comprehensively explore yeast underground metabolism. We integrated the predicted underground network into the yeast consensus genome-scale model, Yeast8, to reconstruct the yeast metabolic twin model, Yeast-MetaTwin, covering 16,244 metabolites (92% of the total yeast metabolome), 2,057 metabolic genes and 59,914 reactions. We revealed that K m parameters differ between the known and underground network, identified hub molecules connecting the underground network and pinpointed the underground percentages for yeast metabolic pathways. Moreover, the Yeast-MetaTwin can predict the by-products of chemicals produced in yeast, offering valuable insights to guide metabolic engineering designs.