Salivary Microbiome Meta-Analysis for Multi-Disease Prediction, 7,750 Samples

A meta-analysis of 7,750 saliva samples from 22 public cohorts (2016–2024) reveals gender differences and core microbiota. Qixiang Yuan used QIIME2 and Wekemo to analyze 16S rRNA data, identifying nine core microbes and constructing a multi-disease prediction model. The study validated the feasibility of using saliva microbiota and machine learning for non-invasive disease diagnosis.

Use Cases

• Training multi-class random forest models for disease classification based on salivary microbial profiles.
• Identifying core microbial taxa associated with healthy baselines from negative control group data.
• Investigating gender differences in oral microbiota composition and structure.
• Analyzing associations between microbiota structure and specific physiological or pathological states like colorectal cancer or HIV.

Strengths

• Includes 7,750 samples aggregated from 22 cohorts, providing a substantial sample size for meta-analysis.
• Reports robust model performance with AUC scores ranging from 0.898 to 1.0 for multi-disease prediction.
• Identifies nine specific core microbiota genera, such as g:Streptococcus, in the healthy control group.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the underlying feature table is unknown, which may limit suitability assessment.
• Data may reflect temporal and source bias inherent to the PubMed-sourced cohorts from 2016–2024.

Provenance

Source

figshare, author Qixiang Yuan.

Collection Method

Meta-analysis of public 16S rRNA saliva data from PubMed (2016–2024), processed with QIIME2 and Wekemo.

Time Range

Cohorts published between 2016 and 2024.

Freshness

Last updated 2026-04-15 05:46:05; freshness should be verified.