Computational tools for biomarker identification in inborn errors of metabolism: A rapid systematic review

Inborn errors of metabolism (IEMs) are a heterogeneous group of rare genetic diseases characterized by enzymatic, transporter, or cofactor deficiencies that disrupt metabolic homeostasis. Biomarker identification is central to their diagnosis, prognosis, and therapeutic monitoring, but phenotypic heterogeneity and limited patient cohorts pose persistent challenges. Computational tools, including genome-scale metabolic modeling, systems biology, and quantitative systems pharmacology, are emerging and aiming to address these limitations. This work aimed to identify, characterize, and synthesize the available evidence on computational tools for predicting or identifying metabolic biomarkers in individuals diagnosed with IEMs. A rapid systematic review was conducted following PRISMA 2020 guidelines, searching PubMed and EMBASE for studies published within the last ten years in English or Spanish. Eligible studies applied computational methodologies, such as in silico modeling, constraint-based modeling, machine learning, or metabolic network analysis, for biomarker discovery, prediction, or validation in IEMs. Three reviewers performed selection independently, resolving discrepancies by consensus. Of 955 records identified, 13 studies met the eligibility criteria. They were classified into four categories: machine learning and artificial intelligence approaches for predictive modeling and risk stratification; genome-scale and constraint-based modeling to simulate enzymatic deficiencies and identify candidate biomarkers; systems biology and network-based modeling for pathway-level signatures and diagnostic visualization; and quantitative systems pharmacology frameworks to model therapeutic response and biomarker trajectories. When combined with conventional methods, computational tools may improve the mechanistic interpretability, scalability, and multi-omics integration during biomarker identification for EIMs. However, most approaches remain exploratory, and prospective clinical validation is limited. Future efforts should prioritize hybrid modeling, standardized reporting, and clinical integration.