AI and Multi-Omics Data Are Building the Future of Precision Medicine

Complex diseases arise from dysregulation across multiple biological layers simultaneously, yet current medical practice still relies largely on one-size-fits-all diagnostics and treatments that ignore individual molecular differences. This review by Mardinoglu and colleagues, published in Molecular Systems Biology, argues that the convergence of longitudinal big biological data, AI, and systems biology represents the most viable path toward truly personalized medicine. The authors synthesize evidence across hundreds of studies to outline a framework in which multi-omics data is continuously collected, integrated, and interpreted to reveal disease mechanisms at unprecedented resolution.

At the core of the systems biology component are genome-scale metabolic models (GEMs). The review traces their development from the first human metabolic network reconstructions — Recon1 and EHMN in 2007 — through progressively refined models including HMR1, HMR2, Recon2, Recon3D, and ultimately Human1 in 2020. These models encode all known biochemical reactions catalyzed by enzymes and transporters encoded in the human genome, enabling researchers to build tissue-specific models — notably for liver, gut, and other organs — that simulate altered metabolic states in disease. Complementary microbe-specific and microbial community GEMs have extended this framework to the gut, oral, vaginal, and skin microbiomes, enabling study of host-microbiome interactions and identification of microbiome-based therapeutic targets.

The review catalogs a broad array of landmark longitudinal multi-omics studies currently underway worldwide, noting that datasets now routinely integrate genomics, epigenomics, transcriptomics, proteomics, metabolomics, lipidomics, and metagenomics alongside clinical records, wearable device outputs, and dietary data. The authors highlight that wearable devices now continuously capture physiological parameters — calories burned, blood pressure, heart rate, activity levels, and sleep — while electronic health records (EHRs) contribute longitudinal clinical notes, physical measurements, and imaging. This data richness positions AI as indispensable for pattern recognition and biomarker discovery at scales impossible for human analysts.

AI's role is presented in two complementary modes. First, classical machine learning and deep learning algorithms identify disease-relevant patterns within high-dimensional multi-omics datasets, with digital pathology cited as a mature example where large structured datasets enabled clinical deployment to assist pathologists. Second, large language models (LLMs) now enable rapid biomedical text mining, accelerating interpretation of biological data and identification of connections among genes, metabolites, microbiome composition, diet, and environmental toxins. Together, these AI tools are enabling discovery of novel biomarkers and drug targets and compressing drug development timelines.

The review's most forward-looking contribution is its treatment of digital twins — personalized computational models of individual patients constructed from their longitudinal big biological data. Whole-body metabolic models, such as those built on the Harvey/Harvetta framework, serve as the substrate for these twins, enabling real-time simulation of individual biological function and prediction of responses to interventions. The authors argue this infrastructure forms the foundation for AI-driven clinical decision support systems, which could fundamentally transform hospital care from reactive to predictive and preventive.

Key caveats acknowledged by the authors include the current fragmentation of data standards across institutions, privacy and regulatory challenges surrounding longitudinal biological data collection, and the computational costs of running whole-body models at scale. The review is primarily synthetic rather than empirical, drawing on existing studies rather than presenting new experimental data, which limits the ability to quantify specific effect sizes or clinical outcomes directly attributable to these integrated approaches.