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

## TL;DR

This paper introduces GutMIND, a large database and machine learning tools to study the gut-brain connection in neuropsychiatric disorders using microbiome data from 3,492 individuals across 12 countries.

## Contribution

The novel contribution is the creation of the GutMIND database and the MetaClassifier framework for diagnosing neuropsychiatric disorders using gut microbiome data.

## Key findings

- Microbial community heterogeneity was significantly higher in patients with neuropsychiatric disorders compared to healthy controls.
- MetaClassifier achieved a mean AUROC of 0.69 in diagnosing 8 disorders using gut microbiome data.
- Nine core protective microbiota species were identified, linked to glutamate synthesis and acetate production.

## Abstract

Emerging evidence underscores bidirectional communication along the microbiota-gut-brain axis in neuropsychiatric disorders. However, the field lacks dedicated metagenomic resources with standardized phenotyping for these conditions. Existing single-cohort studies face inherent limitations due to restricted sample sizes, confounding heterogeneity, and methodological fragmentation, compromising reproducibility and mechanistic insights. To overcome these challenges, we constructed the Gut Microbiome in Multinational Integrated Neuropsychiatric Disorders (GutMIND) database, a comprehensive resource integrating shotgun metagenomic data with harmonized metadata. Adhering to a standardized preprocessing protocol and rigorous quality control workflow, this dataset represents the largest gut-brain microbiome repository to date, encompassing 31 studies across 12 countries (n = 3,492) spanning 14 neuropsychiatric conditions. Utilizing this dataset, we characterized microbial community heterogeneity, which was significantly elevated in patients compared to healthy controls. Subsequently, we developed a computational framework, MetaClassifier, enabling the diagnosis of neuropsychiatric disorders and the identification of microbial biomarkers. Employing a comprehensive two-stage validation strategy, we first assessed the model utilizing taxonomic abundance profiles via nested cross-validation in the high-quality discovery cohort (n = 2,734), achieving a mean AUROC of 0.69 (range: 0.55–0.78) across 8 disorders. Its robustness was further confirmed in an independent platform-extended validation cohort (n = 400), yielding a mean AUROC of 0.71 (range: 0.60–0.76). We also developed the Microbial Gut-Brain Axis Health Index (MGBA-HI), which effectively distinguished neuropsychiatric status in both the high-quality cohort and the platform-extended cohort. Furthermore, integrative analysis of health-abundant species, index-derived biomarkers, and ecological prevalence, we identified 9 core neuropsychiatric-protective microbiota. These species predominantly exhibited metabolic capacities linked to glutamate synthesis and acetate production. Building upon this, the GutMIND framework ensures robust cross-cohort comparability while minimizing technical heterogeneity, thereby enhancing inferential rigor in gut microbiome-neuropsychiatry research. Notably, the MetaClassifier, MGBA-HI, and core microbiota hold translational potential for developing microbiome-based prognostic tools and personalized therapeutic strategies in neuropsychiatric disorders. The source code and usage instructions for MetaClassifier are accessible at https://github.com/juyanmei/MetaClassifier.