GutMIND: A multi-cohort machine learning framework for integrative characteristics of the microbiota-gut-brain axis in neuropsychiatric disorders.

The GutMIND database represents the largest gut-brain microbiome repository to date, integrating 31 studies across 12 countries spanning 14 neuropsychiatric conditions.

• Total sample size of n=3,492 participants
• Data sourced from 31 studies across 12 countries
• Covers 14 neuropsychiatric conditions
• Utilized shotgun metagenomic data with harmonized metadata
• Adhered to a standardized preprocessing protocol and rigorous quality control workflow

Microbial community heterogeneity was significantly elevated in neuropsychiatric patients compared to healthy controls.

• Heterogeneity was characterized across the full GutMIND dataset
• This finding applied across multiple neuropsychiatric conditions
• The analysis used taxonomic abundance profiles from shotgun metagenomic data
• The elevated heterogeneity was detected after standardized preprocessing and quality control

The MetaClassifier framework achieved a mean AUROC of 0.69 across 8 neuropsychiatric disorders in the discovery cohort using nested cross-validation.

• Discovery cohort contained n=2,734 high-quality samples
• Mean AUROC of 0.69 with range of 0.55–0.78 across 8 disorders
• Validation used a nested cross-validation strategy on taxonomic abundance profiles
• This represented the first stage of a two-stage validation strategy

MetaClassifier demonstrated robustness in an independent platform-extended validation cohort, yielding a mean AUROC of 0.71.

• Independent platform-extended validation cohort comprised n=400 samples
• Mean AUROC of 0.71 with a range of 0.60–0.76
• This represented the second stage of the two-stage validation strategy
• Results confirmed cross-platform generalizability of the MetaClassifier

The Microbial Gut-Brain Axis Health Index (MGBA-HI) effectively distinguished neuropsychiatric status in both the high-quality cohort and the platform-extended cohort.

• MGBA-HI was developed as a composite index derived from microbial biomarkers
• Validated in the high-quality discovery cohort (n=2,734) and the platform-extended cohort (n=400)
• The index was designed to reflect gut-brain axis health
• Demonstrated cross-cohort applicability for distinguishing neuropsychiatric from healthy status

Nine core neuropsychiatric-protective microbiota were identified through integrative analysis of health-abundant species, index-derived biomarkers, and ecological prevalence.

• Identification relied on integrative analysis combining three evidence streams: health-abundant species, MGBA-HI-derived biomarkers, and ecological prevalence
• These 9 species predominantly exhibited metabolic capacities linked to glutamate synthesis and acetate production
• The species were characterized as 'neuropsychiatric-protective' based on their associations across multiple disorders
• These findings hold translational potential for microbiome-based therapeutic strategies

The GutMIND framework was designed to minimize technical heterogeneity and ensure robust cross-cohort comparability in gut microbiome-neuropsychiatry research.

• Standardized preprocessing protocol applied across all 31 studies
• Rigorous quality control workflow implemented to reduce batch effects and methodological fragmentation
• Framework addresses limitations of prior single-cohort studies including restricted sample sizes and confounding heterogeneity
• Source code and usage instructions for MetaClassifier are publicly accessible at https://github.com/juyanmei/MetaClassifier