Population-scale Characterization of the Oral Microbiome and Associations with Metabolic Health

Code to reproduce the analyses in “Population-scale Characterization of the Oral Microbiome and Associations with Metabolic Health”.

Our Contributions:

1. Population-scale, high-resolution metagenomics with deep metabolic phenotyping: We profile standardized bilateral buccal-swab whole-metagenome data in 9,431 HPP adults, paired with 44 metabolic measures spanning liver ultrasound, CGM, and DXA.
2. A unified, rigorous multi-layer MWAS framework: We systematically test associations across strain, gene-family, and pathway layers using covariate-adjusted regression and layer-wise multiple-testing control, enabling direct comparison of signals across metabolic systems.
3. Actionable outputs with translational and external support: We deliver a multi-system oral–metabolic association atlas with prioritized cross-phenotype markers, demonstrate proof-of-concept metabolic disease classification using phenotype-selected oral features, and provide independent directional replication at genus resolution.

Data Access

HPP (Human Phenotype Project)

• Controlled Access: Due to ethical and IRB requirements, HPP data is available through a controlled-access portal.
• Access Portal: https://humanphenotypeproject.org/data-access
• Process: Researchers must submit a statement of purpose and sign a data use agreement. Upon approval, data can be accessed in a secure environment.
• TRE Tutorial: After obtaining access, please refer to User-guide-for-TRE.pdf for a detailed guide on how to use the Trusted Research Environment (TRE).
• Ethics Approval: Weizmann Institute IRB #1719-1.

Environment Setup

conda create -n oral_hpp python==3.11
pip install -r requirements.txt

Running Pipeline

The analysis follows a sequential workflow where inputs and outputs are chained. While the high-level steps are outlined below, please refer to: * Subdirectory READMEs: Each folder contains a local README.md with detailed execution instructions and script-level documentation.

1. Preprocess (preprocess/)
   Clean phenotypes; standardize strain/pathway/gene-family abundance (zero-replacement → normalization → PPM → log₁₀).

2. Association analysis (association_analyse/)
   OLS (age, sex, smoking); Then Bonferroni correction (correct_P_value_*).

3. Key oral features (Identification_key_oral_features/)
   Rank by association breadth and take top features per system.

4. Oral feature grouping (oral_features_classfication/)
   Classify significant strain/pathway into Favourable / Adverse / Mixed from association directions across liver, CGM, body.

5. Metabolic disease classfication (metabolic_diseases/)
   Select pathways linked to disease-related phenotypes (5-fold CV); train classifiers (LightGBM) on strain/pathway abundance; evaluate with cross-validation.

6. Replication (replication_study/)
   In an independent cohort (NHANES): preprocess genus and phenotype, run association (BMI, waist circumference), then compare direction.

---