Multikingdom microbiome-based machine learning enables multiple sclerosis diagnosis.

Emerging evidence suggests a role for the gut bacteria in the pathogenesis of multiple sclerosis (MS); however, the role of other microorganisms and their diagnostic potential for MS remain poorly explored. Here, we analyzed large-scale metagenomic data derived from fecal samples (discovery cohort n = 1152; total n = 1306 across 3 geographically diverse cohorts). Subsequently, we utilized multikingdom gut microbiome data to develop machine learning models to distinguish MS patients from healthy controls. Our analysis identified distinct microbiome alterations, revealing 90 bacterial, 3 fungal, 2 viral species, 119 KEGG orthology genes, and 17 metabolic pathways significantly associated with MS. Machine learning models integrating multikingdom taxonomic and functional features achieved the area under the receiver operating characteristic curves (AUCs) of 0.977 for males and 0.978 for females. On external validation datasets, the ensemble models yielded AUCs of 0.813 in males and 0.745 in females, while the 30-marker models reached AUCs of 0.849 and 0.763, respectively. Notably, the accuracy of the model was associated with Faecalibacterium spp. and L-methionine biosynthesis pathways, which were less abundant in MS patients. Collectively, our findings highlight the potential application of multikingdom and functional gut microbiome markers as non-invasive biomarkers for MS.