Multi-Omics and Artificial Intelligence in Adult Sepsis: From Risk Stratification to Organ-Dysfunction Prediction and Individualized Therapy

Sepsis encompasses biologically distinct host-response states, organ-injury pathways, and treatment susceptibilities despite similar clinical presentations, limiting population-level diagnostic, prognostic, and therapeutic strategies. This review examines how multi-omics and artificial intelligence can support four connected tasks in adult sepsis: molecular phenotyping, risk stratification, early prediction of organ dysfunction, and individualized treatment. Transcriptomic and single-cell studies provide the most mature frameworks for defining immune states, while spatial transcriptomics, proteomics, metabolomics, and integrated multi-omics add tissue, functional, and organ-specific information. Artificial intelligence can integrate these signals with electronic health records and monitoring data to assign phenotypes and update dynamic risk. However, prediction does not establish treatment benefit, and current evidence remains dominated by retrospective analyses, small or single-center cohorts, inconsistent sampling, dataset shift, limited external validation, and sparse randomized evidence for biomarker- or AI-guided interventions. Progress will require standardized longitudinal sampling, multicenter validation, calibrated and explainable models, and prospective trials testing whether model-guided decisions improve patient-centered outcomes.