Metabolic modeling: the link between AI and multi-omics data to accelerate the discovery of antibiotic-resistant traits in ESKAPE pathogens

ABSTRACT The development of predictive understanding of bacterial resistance to antibiotics is one of the greatest challenges of modern medicine. Genome-scale metabolic models (GEMs) are mathematical representations of the metabolic network of a given organism assembled from genomic data that predict cellular activity across many environmental and genetic contexts. Because a single model can be solved under thousands of conditions, GEMs generate data at scales that support downstream analysis with artificial intelligence tools, helping to elucidate how pathogens adapt to host environments, evade immune responses, and acquire resistance. Here, we review how GEMs of ESKAPE pathogens are built, what data they require, and what they can and cannot do. We describe the reconstruction workflow from automated draft network through manual curation, gap filling, and experimental validation and the public and experimental data sources each stage depends on. We then describe how accessory genes and transcriptomic measurements are incorporated to yield strain- and condition-specific models and where machine learning contributes both in expanding functional annotation and in consuming model-generated flux distributions. Applications are illustrated with ESKAPE examples throughout, including conserved metabolic vulnerabilities identified across strain collections and flux rewiring observed under antibiotic exposure. Despite this progress, high-quality GEMs exist for only a fraction of ESKAPE species; many are constrained by incomplete annotation or limited experimental validation, and non-metabolic resistance mechanisms are captured only insofar as they impose a metabolic cost. Providing reliable tools to build new models from trustworthy templates, therefore, remains a priority for narrowing the gap between genotype and phenotype.