Microbial communities play important roles in diverse environmental, health-related, and industrial contexts due to their metabolic diversity and adaptability. Thus, accurate metabolic modeling of microbial communities is essential for ecological understanding, microbiome engineering, and biotechnological innovations. Genome-scale metabolic models (GEMs), which systematically represent metabolic networks via gene-protein-reaction rules and stoichiometric matrix constraints, provide powerful computational frameworks to explore microbial capabilities and predict responses to different environmental conditions. Despite the availability of numerous automated GEM reconstruction tools, the models generated by individual tools often exhibit incomplete metabolic coverage. To address these limitations, consensus GEMs, generated by integrating multiple independently reconstructed models, were explored in this thesis. In Chapter 3, I systematically demonstrated that consensus community GEMs substantially improved metabolic network completeness, reduced dead-end metabolites, and enhanced genomic support compared to single-tool reconstructions, thus significantly improving the prediction of interspecies metabolic interactions. To further refine the community GEM predictions, Chapter 4 introduced IMIC (Integration of Metatranscriptomes Into Community GEMs), a novel method developed to incorporate metatranscriptomic data directly into microbial community-scale metabolic models. IMIC improved predictions of individual microbial growth rates and community metabolite interactions beyond traditional abundance-based approaches by explicitly reflecting functional activity within communities. Finally, Chapter 5 presents GEM-ORACLE, an enzyme-constrained GEM framework designed to leverage mutant library phenotypes to predict functions of uncharacterized proteins in the model green alga Chlamydomonas reinhardtii. GEM-ORACLE accurately predicted known gene-reaction associations and uncovered novel annotations for orphan reactions, with validation provided by complementary computational analyses. In summary, these results show that integrating multi-omics data and phenotype-driven constraints into consensus and enzyme-constrained GEMs significantly improves predictive accuracy, biological relevance, and robustness of microbial metabolic modeling, providing robust analytical tools to advance microbiology research, microbial ecology, and biotechnological applications.
Yun-Li Hsieh (Thu,) studied this question.