计算生物学
基因组
比例(比率)
计算机科学
生物
遗传学
基因
地理
地图学
作者
Weishang Han,Luchi Xiao,Haocheng Sun,G. M. Xiang,Qianxi Jia,Haoyu Wang,Boyang Ji,Cheng Zhang,Eduard Johannes Kerkhoven,Jens Nielsen,Hongzhong Lu
标识
DOI:10.1038/s41467-026-75549-w
摘要
Constructing high-quality genome-scale metabolic models (GEMs) for non-model organisms remains challenging. To address this, we developed AlphaGEM, a versatile toolbox leveraging proteome-scale structural alignment, protein language models (PLMSearch), and deep-learning-based predictions for efficient genomic mining to generate GEMs ready for applications. AlphaGEM enhances homologous relationship identification compared to traditional sequence-based methods. Crucially, it employs an ensemble procedure empowered by multiple deep learning toolboxes to effectively mine dark metabolic functions encoded by nonhomologous proteins, thereby expanding species-specific networks. We validated AlphaGEM across prokaryotes (Klebsiella pneumoniae, Bacillus subtilis), eukaryotes (Rhodosporidium toruloides, Pichia pastoris), and complex mammals (Mus musculus, Cricetulus griseus), achieving predictions comparable to manually curated models while outperforming existing tools. Furthermore, we demonstrated its scalability by automatically reconstructing high-fidelity GEMs for 332 distinct yeast species. In summary, AlphaGEM enables precise, rapid GEM construction across diverse domains, providing a solid foundation for universal functional analysis of organisms having genome sequences available. Building genome-scale metabolic models for less-characterized organisms remains challenging due to limited methods for reliable functional annotation. Here, the authors present AlphaGEM, a toolbox that integrates structural alignment and deep-learning approaches to improve protein relationship detection and expand metabolic networks, producing reconstructions that align with curated models across diverse species.
科研通智能强力驱动
Strongly Powered by AbleSci AI