AlphaGEM enables precise genome-scale metabolic modelling by integrating protein structure alignment with
Weishang Han1, Luchi Xiao1, Haocheng Sun1
1State Key Laboratory of Microbial Metabolism, School of Life Science and Biotechnology, Shanghai Jiao Tong University, Shanghai, China.
Nature Communications
|July 16, 2026
Summary
AlphaGEM is a new toolbox that builds high-quality genome-scale metabolic models (GEMs) for any organism. It uses advanced AI to quickly discover metabolic functions and create accurate models for diverse species.
Area of Science:
- Systems Biology
- Metabolic Engineering
Background:
- Constructing genome-scale metabolic models (GEMs) for non-model organisms is complex and time-consuming.
- Existing methods often struggle with identifying all metabolic functions, especially those from nonhomologous proteins.
Purpose of the Study:
- To develop a versatile and efficient toolbox, AlphaGEM, for automated GEM reconstruction.
- To improve the accuracy and scope of GEMs by leveraging advanced computational techniques.
Main Methods:
- Utilized proteome-scale structural alignment and protein language models (PLMSearch) for enhanced homologous relationship identification.
- Employed an ensemble deep learning approach to mine "dark" metabolic functions from nonhomologous proteins.
- Validated the toolbox across diverse organisms including prokaryotes, eukaryotes, and mammals.
Main Results:
- AlphaGEM demonstrated superior performance in identifying homologous relationships compared to traditional methods.
- The ensemble deep learning procedure effectively expanded species-specific metabolic networks by uncovering novel functions.
- Generated high-fidelity GEMs comparable to manually curated models and outperforming existing automated tools.
- Successfully reconstructed GEMs for 332 yeast species, showcasing scalability.
Conclusions:
- AlphaGEM enables precise and rapid GEM construction across diverse biological domains.
- Provides a robust foundation for universal functional analysis in organisms with available genome sequences.
- Significantly advances the field of metabolic modeling for non-model organisms.
