NutriGx Advisor — Personalised Nutrition from Genetic Data

Reading "How You Should Eat" from Your DNA – Experience NutriGx Advisor

A few days ago, I dug out my 23andMe genetic test report from a couple years back and suddenly thought: Could those SNP data directly tell me what to eat more or less of? After all, there are countless "genetic diet methods" floating around online, but not many are reliable. Then I found NutriGx Advisor – a small tool that runs locally, specifically designed to extract nutrition-related SNPs from your genetic data and generate personalized dietary recommendations.

To be honest, I was initially skeptical. Nutrigenomics is a pretty specialized field – what could an open-source project achieve? But after looking at its SNP panel, I was somewhat convinced: it carefully selects over 30 loci validated by GWAS and ClinVar, including commonly known ones like MTHFR (folate metabolism), APOE (saturated fat response), FTO (obesity risk), VDR (vitamin D absorption), and others. Plus, all calculations are done locally without uploading any data, so privacy is well taken care of.

Installation Process: Easier Than Expected

The project is on GitHub, under the repository sulfierry/mcp, with the skill located in the skills/nutrigx_advisor directory. I simply cloned it with git, then installed a few Python packages: pandas, numpy, matplotlib, seaborn, and reportlab. reportlab is optional for generating PDFs; I didn't install it and it still worked fine.

git clone <repository-url>
cd mcp/skills/nutrigx_advisor
pip install pandas numpy matplotlib seaborn

After installation, I tried it with my own 23andMe raw data. The command was super simple:

python nutrigx_advisor.py genome.txt

Of course, if you don't have real data, the project also includes a demo script examples/generate_patient.py that can generate random patient genetic data and run the analysis:

python examples/generate_patient.py --run

This demo script creates a simulated 23andMe-format CSV, then automatically calls the main program, outputting reports and charts to the examples/output/ directory. I ran this first, and watching the console output fly by was quite satisfying.

What Does the Generated Report Look Like?

The report is in Markdown format, and when I opened it, the content was surprisingly rich. It's divided into several sections:

• Overall Risk Profile: A radar chart displaying genetic risk scores for various nutrients, such as folate metabolism, vitamin D, Omega-3 synthesis, etc. Each dimension is rated from 0 to 10, with higher scores indicating greater genetic risk (and thus requiring more attention).
• Detailed Explanation of Each SNP: For every analyzed locus, it lists your genotype, risk allele, explanation, and specific dietary recommendations. For example, my MTHFR is C/T heterozygous, with the risk allele T; it recommends increasing folate intake, especially in the form of methylfolate supplements.
• Gene × Nutrient Heatmap: A color matrix showing how different genotypes affect various nutrient states – for instance, red indicates adverse effects, green indicates beneficial ones – very intuitive.
• Reproducibility Package: If you want to share the results with a doctor or nutritionist, it generates a package containing all commands and environment configuration, including SHA-256 checksums, to ensure results are verifiable.

To be honest, seeing this report got me a bit excited – because before, using commercial genetic interpretation services, it often cost hundreds of yuan, and the reports were quite vague. NutriGx Advisor's report is much clearer, and every recommendation comes with literature support (evidence strength is marked in the SNP panel, such as Strong, Moderate), making it feel more professional.

Details That Impressed Me

This tool really put effort into SNP selection. For instance, it covers the FADS1 and FADS2 loci for fatty acid metabolism, as well as APOE ε typing (rs429358 and rs7412), allowing it to determine if you are ε2/ε3/ε4. If you are an ε4 carrier, a high saturated fat diet will cause a spike in LDL cholesterol, and the report will clearly remind you to limit red meat and full-fat dairy products.

Additionally, it considers the CYP1A2 locus (rs762551) for caffeine metabolism. If you are a slow metabolizer (C/C genotype), the report recommends no more than one cup of coffee per day, otherwise cardiovascular risk increases. This information is quite practical for many coffee lovers.

Another thing that surprised me was the detection of lactose intolerance. Instead of using the LCT gene itself, it uses the rs4988235 locus on the MCM6 gene, which is a regulatory region for lactase expression. If your genotype indicates non-persistent lactase expression, the report suggests choosing lactose-free milk or taking lactase supplements.

Several Pitfalls to Watch Out For

Although this tool is powerful, you still need to be cautious when using it:

• Data format must be accurate: 23andMe raw data is in text format (one SNP per line), but different versions have slight format variations. The tool has built-in multiple parsers to automatically recognize 23andMe, AncestryDNA, and VCF formats. However, if you have modified the file yourself (e.g., removed comment lines), parsing may fail. It's best to use the original downloaded file.
• Results are for reference only: Genetic risk is just one factor influencing nutritional needs and cannot replace professional medical advice. ...