Reconstruction of human metabolic models with large language models

Area of Science:

• Metabolic Engineering
• Computational Biology
• Systems Biology

Background:

• Genome-scale metabolic models (GEMs) are crucial for studying human metabolism.
• Existing models require enhanced precision and collaborative curation methods.

Purpose of the Study:

• Introduce Human2, a consensus human GEM with improved accuracy and biological relevance.
• Enable the reconstruction of tissue- and organ-specific models for various human groups.
• Develop a dynamic whole-body framework for simulating human metabolism.

Main Methods:

• Leveraged large language models (LLMs) and GitHub Actions for automated, efficient, and collaborative GEM curation.
• Integrated transcriptomic, proteomic, and kinetic data for model refinement.
• Constructed an enzyme-constrained dynamic model for simulating interorgan metabolite exchange.

Main Results:

• Human2 offers enhanced precision and biological relevance in human GEMs.
• Reconstructed tissue- and organ-specific models tailored to sex- and age-specific human groups.
• Revealed distinct metabolic features, including differences in arachidonic acid and leukotriene metabolism.
• Integrated models into a dynamic whole-body framework simulating nutritional states (feeding to fasting).

Conclusions:

• Large language models (LLMs) significantly transform GEM reconstruction.
• The dynamic whole-body simulation provides a powerful resource for multiscale human metabolism research.
• Human2 facilitates a deeper understanding of human metabolic variability and responses.