Machine learning-based analysis of gut microbial shifts along the normal–obese–MetS axis in metabolic syndrome using 16S rRNA amplicon data

Metabolic diseases are increasingly associated with gut microbiome dysbiosis; however, integrated approaches combining taxonomic profiling, ecological interpretation, statistical testing, and explainable artificial intelligence remain limited. This study applied a comprehensive amplicon-based microbiome workflow to investigate gut microbial signatures across Normal, Obese, and Disease (MetS) groups. Raw 16S rRNA amplicon reads were quality assessed, trimmed, and taxonomically classified using kraken followed by abundance re-estimation by bracken and diversity analysis. Taxonomic profiling of 144 samples after metadata and quality filtering, identified a diverse gut microbial community across multiple ranks, including 31 phyla, 52 classes, 114 orders, 203 families, 388 genera, and 535 species. Disease samples showed marked taxonomic shifts, reduced alpha diversity, and clear Bray–Curtis separation from Normal and Obese groups. PERMANOVA confirmed significant community-level differences across all taxonomic ranks using Bray–Curtis distances. Differential abundance analysis revealed extensive disease-associated taxa, particularly at genus and species levels, with stronger separation in Normal vs. Disease and Obese vs. Disease comparisons than in Normal vs. Obese. Random Forest models showed strong classification performance for disease-related comparisons, with high ROC-AUC values, while Normal vs. Obese showed weaker discrimination. SHAP-based explainable AI highlighted candidate species-level biomarkers, including Anaerostipes hadrus , Anaerobutyricum hallii , Dorea longicatena , Faecalibacterium prausnitzii , Phocaeicola vulgatus , and Morganella morganii . Overall, this study demonstrates that integrating 16S rRNA amplicon sequencing, statistical testing, machine learning, explainable AI, and network ecology to identify exploratory gut microbial signatures and candidate microbial markers associated with MetS that require independent validation.