Scoping review: microbiome-driven precision psychiatry along the gut-brain axis
Across 28 predominantly observational studies, gut microbiome biomarkers for depression and anxiety show early biological signal but insufficient evidence for routine clinical use.
| Population | Adults with depression or anxiety (primary human studies) |
|---|---|
| Intervention | Gut microbiome profiles, metabolic biomarkers, multi-omics signatures, AI models, and microbiome-targeted interventions |
| Comparator | Healthy controls or varied comparison groups (heterogeneous across studies) |
| Outcome | Biological stratification via microbiome biomarkers; Taxonomic consistency across populations; Treatment response prediction via multi-omics signatures; Efficacy of microbiome-targeted interventions; Artificial intelligence model performance |
Summary of findings
| Outcome | Effect | 95% CI | Certainty | Clinical relevance | Notes |
|---|---|---|---|---|---|
| Biological stratification via microbiome biomarkers | no pooled effect size reported; narrative only | — | Very low | — | 28 studies |
| Taxonomic consistency across populations | heterogeneous findings across populations; in the quantified measure | — | Very low | — | 28 studies |
| Treatment response prediction via multi-omics signatures | exploratory signal only; in the pooled OR/RR or accuracy metrics reported | — | Very low | — | |
| Efficacy of microbiome-targeted interventions | no quantified effect; interventions described as exploratory | — | Very low | — | |
| Artificial intelligence model performance | no validated accuracy metrics reported; described as exploratory | — | Very low | — |
Context
The gut-brain axis hypothesis has gained traction as a target for precision psychiatry. No prior review systematically mapped the integration of multi-omics, AI, and personalized interventions in this field. The methodological gap is clinically relevant before any adoption.
What the study showed
Of 58 identified records, 28 were included after screening. Evidence is predominantly observational and taxonomically inconsistent across populations. Studies of microbial functions, metabolites, and multi-omics signatures suggest potential for biological stratification, but no pooled effect sizes with 95% CI were reported. Microbiome-targeted interventions and AI models remain exploratory without independent validation.
How it was done
Scoping review following JBI guidance and PRISMA-ScR. Search of PubMed/MEDLINE and PubMed Central from inception to July 2026, supplemented by Google Scholar and backward citation searching. Screening and charting of primary human studies; narrative synthesis without meta-analysis.
Effect magnitude
No pooled effect size with 95% CI was calculated — scoping design precludes meta-analysis. Individual study magnitudes were not consolidated.
Risk of bias
Scoping reviews do not assess risk of bias of included studies (AMSTAR-2 not applicable; RoB 2 or ROBINS-I not applied). Narrative synthesis is subject to interpretation bias. Methodological heterogeneity across 28 studies precludes causal or quantitative conclusions.
What this study does NOT prove
This study does not prove causality between microbiome composition and psychiatric disorders. It does not demonstrate efficacy of any specific intervention nor generalizability across diverse populations.
In clinical practice
No basis exists for prescribing microbiome-targeted interventions for depression or anxiety outside research settings. Clinicians should not use microbiome profiles as diagnostic or predictive biomarkers in current practice. Prospective studies with independent validation and standardized methods are required first.
Limitations
Scoping reviews do not assess risk of bias of included studies (AMSTAR-2 not applicable; RoB 2 or ROBINS-I not applied). Narrative synthesis is subject to interpretation bias. Methodological heterogeneity across 28 studies precludes causal or quantitative conclusions.
What is still missing
Prospective RCTs with standardized sequencing methods and validated clinical outcomes are needed. AI models require external validation before any clinical application.
Technical appendix
Version history
- 1.0 · 2026-10-31 — Auto-generated under Evidence Standard v1.0