Nutrigenomics and AI:
towards a diet based on the genetic code

1. The end of the universal diet: why we are all different

Nutrition has long been governed by the logic of the common denominator: universal guidelines for entire populations, identical food pyramids for everyone, macronutrient recommendations based on statistical averages. This approach had historical merits — it helped eliminate severe nutritional deficiencies and reduce certain diseases — but it has a fundamental limitation: it ignores the fact that every human organism is a unique biological system, with a specific genome, an irreplaceable microbiome, and a metabolism that responds in profoundly different ways to the same food.

Research over the past two decades has accumulated overwhelming evidence of this variability. The PREDICT project — one of the largest personalised nutrition studies ever conducted — showed that the postprandial glycaemic response to the same food varies up to tenfold between different individuals. The lipid response to the same dietary fats follows patterns that are nearly unrepeatable from person to person. Even the response to caffeine depends critically on variants of the CYP1A2 gene.

Nutrigenomics — the integration of genetics, genomics, and nutritional science — is the discipline that systematically studies this biological variability in order to translate it into genuinely personalised dietary and supplement recommendations. Not as a futuristic promise, but as a scientific reality that in 2026 is finally finding practical tools for clinical application.

"There is no best diet. There is the best diet for you — and the difference depends on your DNA, your microbiome, and your metabolome."

2. Nutrigenomics: how genes dictate the response to food

Nutrigenomics studies the bidirectional interaction between the human genome and nutrients: on one side, how individual genetic variants influence the absorption, metabolism, and biological efficacy of nutrients (nutrigenetics); on the other, how nutrients influence gene expression through epigenetic mechanisms (nutrigenomics in the strict sense). This two-way direction is crucial: genes are not merely a fixed "blueprint" — they are a dynamic system that nutrients continuously modulate.

The fundamental mechanism through which genetic variants influence nutritional response is the single nucleotide polymorphism (SNP, pronounced "snip"): variations in a single DNA base that alter the function of metabolic enzymes, nutrient receptors, transport channels, or regulatory proteins. A single SNP in the MTHFR gene, for example, can reduce by 40–70% the ability to convert folic acid into its biologically active form (5-MTHF), rendering standard folate supplementation practically ineffective in carriers of this variant.

Understanding these variants has radically transformed nutritional prescribing in precision medicine: no longer "take 400 mcg of folic acid per day" for everyone, but "in carriers of MTHFR C677T, prefer already-methylated 5-MTHF supplements" — with radically different efficacy for the same objective.

3. The key genetic variants for personalised nutrition

Not all SNPs have nutritional relevance. Research has identified a set of variants with high clinical actionability — those for which documented dietary or supplementary interventions exist that are capable of modifying the risk associated with the variant itself. Knowing these variants makes it possible not only to optimise the current diet, but to precisely prevent the chronic diseases to which one's genetic profile predisposes.

4. The microbiota as a mediating layer: multi-omics integration

The genome is the starting point of personalised nutrition — but it is not sufficient on its own. The intestinal microbiome is a second, fundamental biological mediation layer: different bacteria metabolise the same food in radically different ways, producing metabolites with opposing biological effects. The metabolism of TMAO — trimethylamine N-oxide, a pro-atherogenic metabolite derived from the carnitine in red meat — depends critically on microbiome composition, not only on dietary intake or the host's genotype.

The review published in Genes & Nutrition in 2025 synthesises the most recent evidence on multi-omics integration in precision nutrition: genomics, epigenomics, transcriptomics, proteomics, metabolomics, and microbiomics are jointly analysed by machine learning models — in particular transformers and graph neural networks — to build predictions of individual metabolic response with accuracy exceeding 90%.

This multi-layer integration has important practical consequences: one and the same individual carrying APOE ε4 can face very different metabolic risks depending on the composition of their microbiome. The most advanced genomic nutritional recommendations therefore do not rely on genotype alone, but on the complete biological profile — including the microbiota.

"The genome is the architectural blueprint. The microbiome is the building site. Metabolomics is the real-time photograph of how the structure is actually taking shape."

5. Metabolomics: the metabolome as a real-time mirror of diet

While the genome is relatively static — it changes little over time except for somatic mutations — the metabolome is intrinsically dynamic: it represents the totality of all metabolites present in a biological system at any given moment, and is continuously shaped by diet, physical exercise, stress, sleep, and microbiome activity. It is precisely this dynamism that makes it the most valuable tool for monitoring in real time the efficacy of a diet or a supplement protocol.

Metabolomics — the science that measures and analyses the metabolome through advanced spectrometric techniques (NMR, LC-MS) — can detect thousands of metabolites simultaneously from simple biological samples such as blood, urine, or saliva. The individual metabolomic profile — the "metabotype" — allows patients to be stratified into subgroups with homogeneous nutritional responses, predicting how each individual will respond to specific dietary interventions before those interventions have even been tested.

A paradigmatic example is TMAO metabolism: measuring circulating TMAO levels in a patient who consumes red meat makes it possible to determine whether their microbiome is of the "producing" or "non-producing" type — and therefore whether reducing carnitine intake is or is not a metabolic priority for that specific individual.

6. Artificial intelligence in precision nutrition

The volume of biological data needed to build a genuinely personalised nutritional profile — genomic, microbiome, metabolomic, phenotypic, behavioural, and environmental — far exceeds the capacity of conventional human analysis. This is where artificial intelligence becomes not an optional tool but a necessary condition for precision nutrition.

The most advanced machine learning models applied to nutrigenomics — in particular transformers (the same underlying technology as large language models) and graph neural networks — are able to identify interaction patterns between genetic variants, microbiome composition, metabolomic profile, and food response that would be invisible even to the most sophisticated multivariate statistical analysis. The Genes & Nutrition review of 2025 reports accuracy exceeding 90% in predicting individual metabolic response to specific dietary interventions from these integrated models.

In practice, a genomic AI nutrition system functions like a biological consultant: it receives the individual's multi-omics profile as input, identifies risk and response patterns, and generates specific nutritional and supplementary recommendations — with the ability to update those recommendations over time as new metabolomic data are acquired.

7. From data to plate: how it works in practice

Genomic nutrition is not yet available as a mass-market service — but the gap between laboratory and clinic is closing rapidly. In Europe and the United States, dozens of platforms already offer nutrigenetic analyses based on panels of 50–500 nutritionally relevant SNPs, with reports that include specific food, elimination, and supplement recommendations. The quality of reports varies enormously, and the clinical validation of many commercial platforms remains partial.

The genomic nutrition pathway in 5 phases

• 01
  Genotyping: the baseline genetic profile
  Analysis of a panel of nutritionally relevant SNPs from a saliva sample. The most advanced platforms analyse 500+ variants, with reports covering macronutrient metabolism, vitamins, food sensitivities, exercise response, and metabolic disease risks.
• 02
  Microbiome analysis: the modifiable variable
  Intestinal microbiome sequencing (shotgun metagenomics or 16S rRNA) to identify bacterial composition, SCFA production patterns, and the presence of pro- or anti-inflammatory species. Combined with the genotype, it provides the complete biological map of individual food response.
• 03
  Baseline metabolomic profile: the current photograph
  Metabolomic analysis of plasma or urine to identify the individual metabotype and measure key biomarkers (TMAO, homocysteine, 25-OH-D, SCFAs, hsCRP). This is the map of where metabolism stands today — the starting point for measuring progress.
• 04
  AI processing and nutritional plan
  The AI integrates data from the three preceding phases with phenotypic information (age, BMI, physical activity, conditions) and generates the personalised nutritional and supplementary plan, with a ranked list of intervention priorities by impact on longevity and prevention of the diseases identified as most critical in the profile.
• 05
  Metabolomic monitoring and adaptation
  Re-analysis of metabolomics every 3–6 months to verify protocol efficacy: are the biomarkers improving? Is the microbiome diversifying in the expected direction? The AI updates its recommendations based on longitudinal data, creating a cycle of continuous personalisation.

8. The role of supplements in genomic nutrition

In genomic nutrition, a supplement is not a generic addition to a dietary regime: it is a specific response to an identified biological gap. The difference from conventional supplement use is the same as that between a personalised medical prescription and an over-the-counter multivitamin: not necessarily in efficacy, but in precision and intentionality.

Genetic variants identify not only the nutrients for which an increased requirement exists, but also the form in which they must be taken to be effectively usable. A carrier of the MTHFR variant does not respond well to folic acid — they need 5-MTHF. A carrier of low GC gene activity (the vitamin D carrier protein) may require doses far above standard recommendations to achieve functional levels. A carrier of FADS1/FADS2 variants converts ALA fatty acids into EPA and DHA less efficiently — and therefore has an increased requirement for preformed omega-3s.

9. Current limitations and open questions

Nutrigenomics is a rapidly maturing science, but not without important limitations that it is right to communicate transparently. The primary one is the polygenic nature of most nutritional conditions: almost no nutritional trait is determined by a single gene. Obesity, insulin sensitivity, fat response — these are all traits influenced by hundreds or thousands of genetic variants, each with small effects, that interact with each other and with the environment in ways that are still difficult to model with precision.

A second limitation is accessibility and standardisation: commercial nutrigenetic tests vary enormously in quality, and many platforms cite genetic associations with evidence levels insufficient for robust clinical recommendations. The distance between academic research and commercial offerings remains significant.

The third limitation is economic: a complete multi-omics profile — genotyping, microbiomics, metabolomics — still costs several hundred euros and requires specialised interpretive expertise. The democratisation of this technology is underway, but has not yet arrived.