Systematically exploring yeast metabolism through retrobiosynthesis and deep learning
Journal article, 2026
A systematic understanding of cellular metabolism is essential for engineering yeast and uncovering the principles of metabolic robustness and evolution, yet much of its metabolic space remains unexplored. Although yeast genome-scale metabolic models have been reconstructed and curated for over two decades, more than 90% of the yeast metabolome remains uncovered. Here, to address this gap, we have developed an integrated workflow that combines retrobiosynthesis, deep learning-based enzyme annotation and enzyme–substrate prediction to systematically explore yeast underground metabolism. Using the framework, we reconstruct a yeast metabolic twin model, Yeast-MetaTwin, comprising 16,244 metabolites, 1,976 metabolic genes and 59,865 reactions. The model reveals systematic differences in Km distributions between the known and underground networks and identifies key hub metabolites linking the underground network. Moreover, Yeast-MetaTwin predicts by-product formation in yeast cell factories, and we experimentally validate two genes converting geraniol to geranial during geraniol biosynthesis. (Figure presented.)
Author
Ke Wu
Tsinghua University
Haohao Liu
Tsinghua University
Yao Zhou
Sun Yat-Sen University
Manda Sun
Tsinghua University
Runze Mao
Tsinghua University
Yindi Jiang
Shenzhen Institute of Advanced Technology
Eduard Kerkhoven
Chalmers, Life Sciences, Systems and Synthetic Biology
Novo Nordisk Foundation
Yu Chen
Shenzhen Institute of Advanced Technology
Jens B Nielsen
Chalmers, Life Sciences, Systems and Synthetic Biology
BioInnovation Institute
H. T. Tang
Sun Yat-Sen University
Feiran Li
Tsinghua University
Nature Catalysis
25201158 (eISSN)
Vol. In PressSubject Categories (SSIF 2025)
Molecular Biology
Bioinformatics and Computational Biology
Microbiology
DOI
10.1038/s41929-026-01523-w