Vincenzo Pezzano · Pharmacogenomics & Nutrigenomics EN · IT
Anthropic AI for Science · Application
MetaPlantAI

AI-assisted discovery of metabolic signatures to generate testable, clinician-supervised nutritional hypotheses

A research pipeline that turns raw organic-acid panels into evidence-linked, prioritised nutritional hypotheses, as decision support for clinicians, never a diagnosis or a prescription.

What MetaPlantAI is

Life sciences · metabolomics · clinical nutrition · decision support

Organic-acid (OAT) profiles are widely used to probe mitochondrial, oxidative, gut-microbial and detoxification pathways. Yet their interpretation is still largely manual and unstandardised, and clinicians lack objective tools to turn a raw panel into evidence-linked, testable nutritional hypotheses.

MetaPlantAI is a Python/FastAPI research pipeline that reads a metabolic panel, maps altered metabolites to the biological pathways they imply, and matches those inferred needs to plant secondary metabolites, ranked by their measured concentration in specific plant parts and cross-checked against the scientific literature (PubMed). Every output carries severity grading, safety flags and a mandatory clinician-review disclaimer.

The research question. Across patients (adults and children), do metabolic profiles show recurrent, statistically robust patterns that map to specific pathways and can be prioritised as candidate targets for individualised nutritional support, as decision support, under medical supervision, rather than treatment?

How it works

From raw panel to caveated hypothesis
Step 1

Parse the panel

Reads organic-acid (OAT) panels from a reference laboratory, with multi-laboratory support in progress.

Step 2

Map to pathways

Links altered metabolites to the metabolic pathways and biological activities they imply, with severity grading.

Step 3

Match & cross-check

Matches inferred needs to plant metabolites, concentration-ranked and cross-referenced against PubMed evidence.

Outputs explicitly separate well-supported associations from low-confidence ones, flag weak or extrapolated links, and are drafted as clinician-facing interpretive summaries, decision support, not prescription.

What makes it different

The microbiome as a bidirectional metabolome bridge

A substantial share of OAT markers are microbial in origin, and several are the gut-bacterial transformation of dietary plant secondary metabolites (for example, hippuric acid from polyphenols). Polyphenols, in turn, reshape the microbiota. MetaPlantAI exploits this axis in both directions:

Diagnostic reading

The metabolic signature reveals the state of the microbiota, a snapshot of an otherwise hidden ecosystem.

Interventional lever

Concentration-ranked plant metabolites become candidate levers to modulate that state, with a measurable readout on the same panel.

To our knowledge, no other tool bridges the human and plant secondary metabolomes quantitatively, via the microbiome, in this way. Single microbial markers are always used within patterns, never in isolation.

Planned extension: genotype × metabolome (G×E)

A core research aim is to integrate the individual's genotype (nutrigenetic and pharmacogenetic polymorphisms, from a companion genetic-report pipeline) with the metabolomic phenotype, so that nutritional hypotheses account for personal variants. Examples include MTHFR/MTRR for folate and B12 markers, COMT for catecholamine markers, DAO/HNMT for histamine markers, and detoxification genes (GST, SOD2, NQO1) for oxidative markers.

This genotype-informed calibration is evidence-graded (documented / high-plausibility / hypothesized / speculative) and clinician-validated, never auto-prescriptive.

A preliminary, illustrative signal

Hypothesis-generating only

In an initial set of seven anonymised OAT panels (adults and children), one marker was elevated in every single panel: indican (7/7), a marker of gut-microbiome protein putrefaction. Other recurrences appeared in tryptophan and serotonin turnover, B-vitamin cofactors and mitochondrial energy metabolism.

The sample is small and uncontrolled, so these recurrences are strictly hypothesis-generating, but the fully penetrant indican signal is notable, and it points to a single testable question linking the project's microbial and immune axes. This is a hypothesis to be tested, not a conclusion to be assumed.

Safety, ethics & responsible use

Decision support for licensed clinicians

The current phase is methodological and computational: pattern discovery on anonymised, existing panels plus evidence synthesis. It involves no intervention on patients, and none on minors.

All outputs are decision support for licensed clinicians. The tool does not diagnose, prescribe, or replace medical care, and states this in every report. Recommendations are confined to nutrition and micronutrients within established safety margins, and remain subordinate to conventional care. Any future prospective phase would run under confirmed clinical supervision. All patient data are anonymised.

The team

An interdisciplinary research group within Associazione Virdimura (Palermo)

Biostatistical support is being sought as the cohort scales. All patient data referenced are anonymised (Panel 1…N); no identifying information is used.

Interested in the research?

Collaboration, clinical supervision, data partnership

MetaPlantAI is at an early, self-funded stage, submitted to Anthropic's AI for Science program (currently under evaluation). If you are a clinician, laboratory or researcher interested in the method, we would be glad to talk.

Selected scientific references
  1. Ticinesi et al. 2023, Nutrients. DOI 10.3390/nu15051138
  2. Cheng et al. 2023, Phytomedicine. DOI 10.1016/j.phymed.2023.154979
  3. Meroño et al. 2022, Food Research International. DOI 10.1016/j.foodres.2022.111632
  4. Daneberga et al. 2021, Nordic Journal of Psychiatry. DOI 10.1080/08039488.2021.2014954
  5. Agus et al. 2018, Cell Host & Microbe. DOI 10.1016/j.chom.2018.05.003
  6. West et al. 2020, Gut. DOI 10.1136/gutjnl-2019-319620
  7. Salminen 2022, Ageing Research Reviews. DOI 10.1016/j.arr.2022.101573
  8. Hughes et al. 2012, Brain, Behavior, and Immunity. DOI 10.1016/j.bbi.2012.05.010
  9. Pichersky & Lewinsohn 2011, Annual Review of Plant Biology. DOI 10.1146/annurev-arplant-042110-103814