Nutrigenomics — Personalised Nutrition from Genetic Data

Skill ID: nutrigenomics

Version: 0.3.2

Status: Beta

Author: David de Lorenzo

Requires: Python 3.11+, pandas, numpy, matplotlib, seaborn, reportlab (optional)

What This Skill Does

The Nutrigenomics generates a personalised nutrition report from consumer

genetic data (23andMe, AncestryDNA raw files or VCF). It interrogates a curated

set of nutritionally-relevant SNPs drawn from GWAS Catalog, ClinVar, and

peer-reviewed nutrigenomics literature, then translates genotype calls into

actionable dietary and supplementation guidance — all computed locally.

Key outputs

• Markdown nutrition report with risk scores and per-SNP genotype calls
• Radar chart of nutrient risk profile
• Gene × nutrient heatmap
• Reproducibility bundle (README_reproducibility.txt, environment.yml, checksums.txt, provenance.json)

Trigger Phrases

The Bio Orchestrator should route to this skill when the user says anything like:

• "personalised nutrition", "nutrigenomics", "diet genetics"
• "what should I eat based on my DNA"
• "nutrient metabolism", "vitamin absorption genetics"
• "MTHFR", "APOE", "FTO", "BCMO1", "VDR", "FADS1/2"
• "folate", "omega-3", "vitamin D", "caffeine metabolism", "lactose", "gluten"
• Input files: .txt or .csv (23andMe), .csv (AncestryDNA), .vcf

Curated SNP Panel

Macronutrient Metabolism

Micronutrient Metabolism

Omega-3 / Fatty Acid Metabolism

Caffeine & Alcohol

Food Sensitivities

Antioxidant & Detoxification

Algorithm

1. Input Parsing (parse_input.py)

Accepts:

• 23andMe .txt or .csv (tab-separated: rsid, chromosome, position, genotype)
• AncestryDNA .csv
• Standard VCF (extracts GT field)

Auto-detects format from header lines. Normalises alleles to forward strand using

a hard-coded reference table (avoids requiring external databases).

2. Genotype Extraction (extract_genotypes.py)

For each SNP in the panel:

1. Look up rsid in parsed data
2. Return genotype string (e.g. "AT", "TT", "AA")
3. Flag as "NOT_TESTED" if absent (common for chip-to-chip variation)

3. Risk Scoring (score_variants.py)

Each SNP is scored on a 0 / 0.5 / 1.0 scale:

• 0.0 — homozygous reference (lowest risk)
• 0.5 — heterozygous
• 1.0 — homozygous risk allele

Composite Nutrient Risk Scores (0–10) are computed per nutrient domain by

summing weighted SNP scores. Weights are derived from reported effect sizes

(beta coefficients or OR) in the primary literature.

Risk categories:

• 0–3: Low risk — standard dietary advice applies
• 3–6: Moderate risk — dietary optimisation recommended
• 6–10: Elevated risk — consider testing and targeted supplementation

> Important caveat: These are polygenic risk indicators based on common

> variants. They are not diagnostic. Rare pathogenic variants (e.g. MTHFR

> compound heterozygosity with high homocysteine) require clinical confirmation.

4. Report Generation (generate_report.py)

Outputs a structured Markdown report with:

• Executive summary (top 3 personalised findings)
• Per-nutrient sections: genotype table → interpretation → recommendation
• Radar chart (matplotlib) of nutrient risk scores
• Gene × nutrient heatmap (seaborn)
• Supplement interactions table
• Disclaimer section
• Reproducibility block

5. Reproducibility Bundle (repro_bundle.py)

Exports to the output directory (not committed to the repo):

• README_reproducibility.txt — step-by-step instructions to reproduce the analysis manually
• environment.yml — pinned conda environment
• checksums.txt — SHA-256 checksums of the SNP panel and output report (input file intentionally excluded to avoid persisting a fingerprint of genetic data)
• provenance.json — timestamp, version, and format arguments (input filename intentionally omitted)

Note: No executable scripts are generated. The reproducibility bundle contains

only text files for documentation and integrity verification.

Execution

To run the analysis on a user-provided genetic file, execute this command directly:

python {baseDir}/openclaw_adapter.py --input <path_to_genetic_file> --format auto

To run a demo without real genetic data (synthetic patient file included with the skill):

python {baseDir}/openclaw_adapter.py --input {baseDir}/tests/synthetic_patient.csv --format 23andme

{baseDir} is replaced by OpenClaw at runtime with the absolute path to this skill's folder. Do not substitute it manually. Output is written to a timestamped directory (nutrigenomics_output_YYYYMMDD_HHMMSS/) in the current working directory and persists until manually deleted.

Supported --format values: auto (default), 23andme, ancestry, vcf.

Usage

# From 23andMe raw data
openclaw "Generate my personalised nutrition report from genome.csv"

# From VCF
openclaw "Run Nutrigenomics analysis on variants.vcf and flag any folate pathway risks"

# Targeted query
openclaw "What does my APOE status mean for my saturated fat intake?"

# Run the demo report (no real genetic data needed)
openclaw "Run a demo nutrigenomics report using the synthetic patient file"

File Structure

skills/nutrigenomics/
├── SKILL.md                      ← this file (agent instructions)
├── nutrigenomics.py            ← main entry point
├── parse_input.py                ← multi-format parser
├── extract_genotypes.py          ← SNP lookup engine
├── score_variants.py             ← risk scoring algorithm
├── generate_report.py            ← Markdown + figures
├── repro_bundle.py               ← reproducibility export
├── data/
│   └── snp_panel.json            ← curated SNP definitions
├── tests/
│   ├── synthetic_patient.csv     ← fixed 23andMe-format test data (for pytest)
│   └── test_nutrigenomics.py           ← pytest suite
└── examples/
    ├── generate_patient.py       ← random patient generator (demo use)
    ├── data/                     ← generated patient files land here (gitignored)
    └── output/
        ├── nutrigenomics_report.md     ← pre-rendered demo report
        ├── nutrigenomics_radar.png     ← demo radar chart (nutrient risk profile)
        └── nutrigenomics_heatmap.png   ← demo gene × nutrient heatmap

> Note: Runtime output directories and randomly generated patient files are

> excluded from version control. Only the pre-rendered demo

> report in examples/output/ is committed.

Privacy

All computation runs locally — no genetic data is ever transmitted to external

servers or third-party services.

What the report contains: The Markdown report includes per-SNP genotype calls

(e.g. AT, TT) for each of the 58 panel SNPs analysed. This is intentional:

knowing your specific genotype at each nutrition-related locus is what makes the

report actionable. Full raw genome data from the input file is not reproduced in

the report; only the 58 panel SNPs are included.

File persistence: Output files (report, figures, reproducibility bundle) are

written to a timestamped nutrigenomics_output_YYYYMMDD_HHMMSS/ directory under

the working directory and persist on disk until manually deleted. The input

file is read-only and is never copied into the output directory.

If you are running this skill on behalf of others or in a shared environment,

delete the output directory once the user has downloaded their results.

Limitations & Disclaimer

1. Not a medical device. This skill provides educational, research-oriented

nutrigenomics analysis. It does not constitute medical advice.

1. Common variants only. The panel covers SNPs with MAF > 1% in at least one

major population. Rare pathogenic variants are out of scope.

1. Population context. Effect sizes are predominantly derived from European

GWAS cohorts. Risk estimates may not generalise equally across all ancestries.

1. Gene–environment interaction. Genetic risk scores interact with baseline

diet, lifestyle, microbiome, and epigenetic state. A "high risk" score does not

mean a nutrient deficiency is present — it means the individual may benefit from

monitoring.

1. Simpson's Paradox note. Population-level associations used to derive weights

may not reflect individual trajectories (see Corpas 2025, *Nutrigenomics and

the Ecological Fallacy*).

Roadmap

• [ ] v0.2: Microbiome × genotype interaction module (16S rRNA input)
• [ ] v0.3: Longitudinal tracking — compare reports across time
• [ ] v0.4: HLA typing for immune-mediated food reactions (coeliac, gluten sensitivity)
• [ ] v1.0: Multi-omics integration (metabolomics + genomics + dietary recall)

References

This skill's SNP panel and methodology are informed by peer-reviewed nutrigenomics research. For verification and additional details, consult:

• PubMed MEDLINE: https://pubmed.ncbi.nlm.nih.gov/
• GWAS Catalog: https://www.ebi.ac.uk/gwas/ (published genome-wide association studies)
• ClinVar: https://www.ncbi.nlm.nih.gov/clinvar/ (variant interpretations)

Users are encouraged to verify specific claims through these authoritative sources and with qualified healthcare providers.

Contributing

The SNP panel (data/snp_panel.json) is maintained by the skill author.

To suggest additions or corrections, contact David de Lorenzo directly via

GitHub (@drdaviddelorenzo) or open

an issue on GitHub.