AI-DRIVEN MICROBIOME-BASED BIOMARKERS FOR EARLY DIAGNOSIS OF CANCER AND AUTOIMMUNE DISEASES: CURRENT ADVANCES AND FUTURE PERSPECTIVES

Abstract

The human microbiome encompassing the gut, oral, and mucosal microbial communities has emerged as a rich, minimally invasive source of biomarkers for diseases that remain difficult to diagnose at an early, treatable stage, including gastrointestinal and non-gastrointestinal cancers and a broad spectrum of autoimmune conditions. Advances in high-throughput sequencing (16S rRNA amplicon and shotgun metagenomics) have generated large, high-dimensional, compositional datasets that exceed the analytic capacity of conventional statistics. Artificial intelligence (AI) and machine learning (ML) spanning random forests, gradient boosting, support vector machines, deep neural networks, graph neural networks, and, most recently, transformer-based genomic foundation models have become central to extracting reproducible diagnostic signals from this complexity. This review synthesizes recent (2021–2026) evidence on AI-driven microbiome biomarker discovery across colorectal, gastric, pancreatic, hepatocellular, oral/head-and-neck, and breast cancers, and across autoimmune diseases including systemic lupus erythematosus, rheumatoid and juvenile idiopathic arthritis, inflammatory bowel disease, multiple sclerosis, and myasthenia gravis. We summarize commonly used algorithms and representative diagnostic performance, describe the growing role of explainable AI and multi-omics integration in improving biological interpretability, and critically appraise persistent barriers to clinical translation including batch effects, cohort heterogeneity, low-biomass contamination, data leakage, limited external validation, and regulatory uncertainty. We conclude by outlining future directions, including federated learning, foundation models for metagenomics, and prospective multicentre validation, that are needed to move AI-microbiome diagnostics from research cohorts into routine clinical screening.