Related projects — what this is, and what it is not¶

Scope of this page: the projects biology-as-code sits next to, and a stated position on each — consume, cite, watch, or leave alone. Written to answer issue #10 and to stop this repository reinventing a layer that already exists.

As of 2026-08-29. Every licence and activity claim below was checked against the artefact itself on that date, not against a summary. Where a check failed to find a licence, this page says OPEN and stops — the same rule FDP-1 §4 applies to nutrient values. OPEN means not known; it does not mean unlicensed.

How to read the verdicts¶

Verdict	Meaning
UPSTREAM	We consume it, or should. Its identifiers or terms belong in our vocabulary; we do not rebuild them.
ADJACENT	Real overlap in subject, different altitude. Cite it, link to it, do not vendor it.
PROCESS	Read it for how they work, not for what they ship.
WATCH	No overlap today; a plausible path to overlap. Named tripwire below.
UNRELATED	Surfaced in the same searches, shares no layer. Listed so the question stops being asked.

The table¶

Project	What it actually is	Licence (verified)	Verdict
FoodOn	Food entity ontology, 40 MB release, active	CC BY 4.0 — agrees in repo, OBO Foundry, Bioregistry and `foodon.owl` header	UPSTREAM
CDNO	Food-composition component vocabulary	CC BY 3.0 in `cdno.owl`; repo `LICENSE` file is CC0-1.0 — see Licence traps	UPSTREAM
FOBI	Food ↔ biomarker ontology	CC BY 4.0 in `fobi.owl` and repo; OBO Foundry and Bioregistry both still say 3.0	UPSTREAM (dormant since 2022-05)
CompTox / DSSTox	EPA chemical-exposure dashboard, >1M chemicals, DTXSID ids	US public domain; API key free but request-by-email	UPSTREAM
Bioregistry	Metaregistry of 2,768 prefixes	MIT	UPSTREAM — and we owe it two prefixes
joint-food-ontology-wg	Cross-ontology coordination venue	No `LICENSE`	PROCESS (files last pushed 2023-07-14)
PTFI	Multi-omics characterisation of ~1,650 foods	`OPEN` — no licence stated on either public site	WATCH — namespace risk, not data risk
VMH	Recon3D / AGORA / whole-body GEM host	`OPEN` — see the VMH position	ADJACENT, and already a redistribution exposure
COBRApy	Constraint-based metabolic modelling	GPL-2.0 — matters, see below	ADJACENT
Tellurium	SBML/Antimony modelling environment	Apache-2.0	ADJACENT
libSBML	SBML reference implementation	LGPL-2.1 (majority)	ADJACENT
PySB	Rule-based systems-biology modelling	BSD-2-Clause	ADJACENT
Open mHealth schemas	JSON schemas for mobile health measures	Apache-2.0	ADJACENT — but read the IEEE 1752.1 note
awesome-nutrition-tracking	Curated list of consumer trackers and food APIs	CC0-1.0	ADJACENT — it catalogues the problem
anatomy	three.js 3D human anatomy explorer	No `LICENSE` — all rights reserved	WATCH
gao-lab guideline	Lab training curriculum, largely in Mandarin	No `LICENSE`	PROCESS
biocode	Genomics utility scripts (GFF, FASTA, BLAST)	MIT	UNRELATED

The ontology layer — the one that matters¶

The issue said these are "higher leverage than any repo in the screenshot." That is correct, and the reason is narrow: the claim vocabulary, schemas/aca.ttl already names FoodOn, CDNO, FOBI, ChEBI, MONDO, UBERON and NCBITaxon as the normalisation targets for claim subjects and objects. They are not candidates. They are the vocabulary this repository has already committed to.

The work left is not choosing them. It is resolving every accession before shipping it — the rule the claim shapes (schemas/claim-shape.ttl) already state as never ship a guessed accession.

FoodOn¶

The clean case. dcterms:license in the release header, the repo LICENSE.txt, the OBO Foundry registry and Bioregistry all say CC BY 4.0. Release 2025-12-30, repo active. Consume it, attribute it, do not fork it.

CDNO¶

CDNO is the resolution target for FDP-1's nutrient_ref field, so its licence is load-bearing for anything downstream of a nutrient value. It is CC BY 3.0, not 4.0, and not CC0 — see Licence traps for why three sources disagree.

FOBI¶

Food–biomarker links — the layer between "ate this" and "measured that", which is the join our biomarker claims need. Two cautions: the repository has not been pushed since 2022-05-03, and the registries carry a stale licence for it. Treat FOBI as a frozen but usable artefact, and read its licence off fobi.owl.

CompTox / DSSTox¶

US public domain, no attribution obligation, no commercial restriction — the only one of the four with no rights friction at all. DTXSID is the right identifier for food-contact and additive substances, which is exactly what FDA_ingredients/ is full of. The API needs a free key obtained by emailing ccte_api@epa.gov, and bulk data refreshes roughly every six months.

joint-food-ontology-wg¶

A coordination venue, not an artefact. Repository files were last pushed 2023-07-14; issue traffic since is sparse. There is nothing here to vendor. It is the right place to ask whether an alignment exists before building one — which is precisely the question in the next section.

Two load-bearing claims, tested¶

Both of these are assumptions this repository already relies on. If either is wrong, something downstream is wrong. Both were attacked rather than confirmed.

Claim A — "no official FoodOn → FDC mapping exists"¶

CONFIRMED. Reproduce it:

curl -sL https://raw.githubusercontent.com/FoodOntology/foodon/master/foodon.owl -o foodon.owl
grep -oE "<oboInOwl:hasDbXref[^>]*>[A-Za-z_]+:" foodon.owl \
  | grep -oE ">[A-Za-z_]+:" | sort | uniq -c | sort -rn | head -20

Across the full 40 MB release, the cross-reference namespaces FoodOn actually carries are taxonomic and regulatory:

Prefix	Count
`itis:`	2,457
`PLANTS:`	834
`Europe:`	411
`Codex:`	399
`MANSFELD:`	324
`langual:`	49

There is no FoodData Central namespace. FDC appears exactly once in the whole file, as a URL inside a prose definition (fdc.nal.usda.gov/fdc-app.html#/food-details/173711/nutrients), plus four links to the retired NDB interface. Fifty-three langual: references exist — LanguaL, not FDC, is the alignment FoodOn actually ships.

The trap worth naming: FoodOn does carry USDA cross-references — 834 of them — but they are USDA PLANTS, a plant-taxonomy database. A careless check greps for "USDA", finds hundreds of hits, and reports a mapping that is not there. USDA PLANTS is taxonomy; FoodData Central is composition. They are different databases.

The gap is real. Anything joining a FoodOn term to an FDC id is our own work and must carry its own provenance.

Claim B — the VMH position¶

SUBSTANTIALLY CONFIRMED, one detail now stale.

THIRD-PARTY-DATA.json records the VMH licence as OPEN with licence_confidence: OPEN, on the grounds that the NAR 2019 database paper names no licence and this project's own notes record CC BY-NC 2.0 for Recon3D. That position stands, and is independently corroborated: Bioregistry carries full entries for vmh.metabolite, vmh.reaction and vmh.gene — resolving https://www.vmh.life/#metabolite/$1 — with the license field null. A metaregistry that resolves the identifier still declines to assert the terms.

One detail has changed. The register's note says "vmh.life returns 403 to automated retrieval." On 2026-08-29 the site returned 200, with and without a browser user-agent. It is reachable. What it does not have is a licence: there is no /terms, /license or /about page (all 404), and the 27 KB front-page shell contains no occurrence of licence, license, Creative Commons, copyright, terms or non-commercial.

So the conclusion is unchanged and the reasoning is now stronger: VMH is silent, not unreachable. The licence enquiry is still the only way to close it, and MASTER_CROSSWALK.tsv stays non-commercial until it does.

Licence traps¶

Three sources disagreed about two ontologies. The disagreements run in opposite directions, which is why "check the registry" is not a method.

Ontology	Repo `LICENSE`	OBO Foundry	Bioregistry	The artefact's own header	Authoritative
CDNO	CC0-1.0	CC BY 3.0	CC BY 3.0	CC BY 3.0 (`dcterms:license` in `cdno.owl`)	CC BY 3.0
FOBI	CC BY 4.0	CC BY 3.0	CC BY 3.0	CC BY 4.0 (`fobi.owl`)	CC BY 4.0

For CDNO the repository is wrong; for FOBI the registries are. GitHub's licence API reports CDNO as CC0-1.0, because it reads the LICENSE file and the ontology's own declaration is somewhere else entirely. Bioregistry inherits OBO Foundry's stale FOBI value, so the two "independent" registries are one source.

The rule this produces: the licence of an artefact is what the artefact declares. A repository's LICENSE file governs the repository. A registry entry is a summary and can be years old. A paper's CC BY covers the paper, not the database it describes.

Verify with:

curl -sL http://purl.obolibrary.org/obo/cdno.owl | grep -m1 dcterms:license
curl -sL http://purl.obolibrary.org/obo/fobi.owl \
  | grep -m1 -oE "creativecommons.org/licenses/[a-z-]+/[0-9.]+"

This is the same failure mode THIRD-PARTY-DATA.json exists to catch, and the reason check_third_party.py runs in CI.

The modelling layer — why we stay out of it¶

The GEM primer already argues the architectural case: a GEM answers which fluxes are feasible across the whole network; this package models what happens to a meal through named stages with provenance on every value. That argument is unchanged.

There is a second reason, and it is not architectural.

COBRApy is GPL-2.0. This package is Apache-2.0 with a patent non-assertion covenant, and advertises itself as zero-dependency. Taking COBRApy as a dependency would put a copyleft obligation on top of a permissive licence that cannot carry it. The decision not to become a GEM was made on modelling grounds; the licence makes it structural. If a flux calculation is ever genuinely needed, it belongs behind a process boundary — a separate tool, its own repository, its own licence — never as an import.

The SBML stack is cleaner: Tellurium is Apache-2.0, PySB is BSD-2-Clause, libSBML is LGPL-2.1. The issue's own condition is the right one — only if a LAW-SPEC ever needs a declared SBML hook. Today none does. A LAW-SPEC declares a Gate, a Bound and Conditions; it does not declare kinetics. When one does, libsbml at a module boundary is the cheapest correct move, and LGPL-2.1 permits that use.

The composition-data layer¶

PTFI — a namespace question, not a data question¶

PTFI characterises foods at molecular scale: untargeted metabolomics, lipidomics, ionomics, fatty acids, with glycomics and proteomics in development. They operate a layer below this repository. A complete molecular inventory of a lentil does not tell you that co-ingested ascorbate widens the non-haem iron bound — that is a claim with conditions and a citation, which is what the 47 LAW-SPEC cards and the claims register hold.

So "complementary upstream source" is the comfortable answer. The uncomfortable one is about identifiers. An initiative that characterises foods at scale does not only produce values; it produces the ids everyone downstream joins on. If PTFI becomes the default food-composition namespace, the question is not whether it competes with the laws — it does not — but whether our crosswalk gains a column or gets routed around. The cheap hedge is to map to them early.

Shipped versus claimed, as of 2026-08-29:

Data access requires account registration on the PTFI Research Hub. There is no open bulk download.
No licence or terms-of-use statement appears on either public site. Position: OPEN. Not asserted, not assumed permissive.
No identifier namespace is publicly named. Whether PTFI mints its own food ids or reuses existing ones could not be determined from public pages.
The flagship paper (Nat Food, 2024; PMID 38459394, with a correction at PMID 38499749) is not in PMC, so the data-availability statement could not be read without a subscription.

An initiative whose stated mission is data "shared openly and equitably as a global resource" currently publishes no licence and gates access behind registration. That is not an accusation — young initiatives ship late — but it is the reason PTFI is WATCH and not UPSTREAM. Nothing should be built against it until the terms exist in writing.

Bioregistry — and the two prefixes we owe it¶

Bioregistry (MIT, 2,768 prefixes, active daily) is a metaregistry: it registers namespaces, not entities. MASTER_CROSSWALK.tsv maps entities. They do not compete; the crosswalk should declare its prefixes in Bioregistry terms.

The nearer neighbours are in the same organisation, and they are the ones worth watching: Biomappings (MIT) is curated entity-level equivalences, and SeMRA (MIT) assembles and reasons over them. That is much closer to what the crosswalk does than Bioregistry itself is. Before the crosswalk grows a new mapping lane, check whether Biomappings already carries it.

Two findings that are directly actionable:

USDA FoodData Central has no Bioregistry prefix. Searching all 2,768 records for fdc.nal, fooddata or FoodData Central returns only false positives (holofooddata.org, an unrelated project) and one unrelated veterinary code, usda.cvb.pcn. The most-used nutrition identifier space in the world is not in the metaregistry.
INFOODS has no prefix either. FDP-1 accepts both fdc and infoods as alternate keys for nutrient_ref, so this repository already depends on two namespaces that nothing standardises.

Submitting those two prefix requests is a few hours of work, benefits everyone downstream, and is the clearest available example of contributing upstream instead of building a private table.

Note also that Bioregistry's license field is null for vmh.metabolite, comptox and langual — even where the underlying data is unambiguously public domain, as CompTox is. Bioregistry resolves identifiers; it does not assert rights. That is exactly why THIRD-PARTY-DATA.json exists and cannot be replaced by pointing at a registry.

The person layer¶

This repository models what a body does with a meal. It does not measure the body, and it does not render it.

Open mHealth schemas¶

Apache-2.0, active. JSON schemas for mobile-health measures, with a Java and Kotlin SDK. The right answer if a downstream product ever needs to exchange measurement data — do not invent a glucose or step schema.

Two things to know before adopting any of it:

Its sleep and physical-activity schemas are superseded by IEEE 1752.1. The repository says so itself. Adopt the IEEE schemas for those measures, not the Open mHealth originals.
It has no food-composition schema. The closest are calories-burned and temporal-relationship-to-meal — a meal is a timestamp in that model, not a composition. There is no overlap with a food packet, and nothing there to reuse for one.

awesome-internet-of-the-body¶

The sibling list (murffious/awesome-internet-of-the-body, CC0-1.0): the sensors and apps that measure the body, curated privacy-first. Kept deliberately separate — that list is about instruments, this repository is about mechanism. See its page here.

awesome-nutrition-tracking¶

CC0-1.0, active, 146★. Worth reading precisely because it catalogues the problem this repository was written against. Its Food Databases section — FDC, Open Food Facts, Edamam, Spoonacular, Nutritionix, FatSecret and the rest — is a list of places to get a number. Not one entry ships an evidence grade or a provenance chain with that number. The list is an accurate census of a layer where values circulate with their origins already stripped, which is the opening paragraph of this repository's README restated as a directory.

There is no overlap. biology-as-code is not a tracker and should not appear on that list until it has something a tracker could consume.

anatomy¶

thebuggeddev/anatomy — an interactive three.js 3D human anatomy explorer, 2,870★, TypeScript. murffious/anatomy is a fork of it.

Why it is WATCH rather than UNRELATED: our LAW-SPEC cards carry an Organ field and our claim schema resolves anatomical sites to UBERON. A renderer keyed on UBERON identifiers would turn a law card into a picture. That is a real, attractive convergence — and it is also the fastest way to acquire a 3D-rendering dependency this project has no business maintaining. If it ever happens, it belongs in a separate repository that consumes this one, and only after the licence question is answered.

Process, not product¶

gao-lab guideline¶

The issue's note — "read their checklist; don't vendor it" — is right, with one correction to what it is. This is not a reproducibility standard. It is the Peking University Gao lab's internal training curriculum for new interns: Linux, Python, R, writing, statistics, machine learning, with linked PDFs and courses, written largely in Mandarin. Repository files were last pushed 2025-05-06 and it carries no LICENSE.

Useful as a skills reading list. It defines no schema, ships no data, and overlaps with nothing here.

biocode¶

MIT, 573★, active. A curated grab-bag of genomics utility scripts — GFF, FASTA, BLAST, Chado — explicitly framed by its author as "a collection of bioinformatics scripts many have found useful."

There is no nutrition content and no shared layer. It appears on this page because it surfaces in the same GitHub searches, and it is listed so the question does not need asking twice. UNRELATED.

What would change these verdicts¶

Tripwires. If one trips, this page is wrong and needs revisiting.

Tripwire	Verdict it overturns
PTFI publishes a licence and an open bulk download	PTFI moves WATCH → UPSTREAM; map to their ids immediately
PTFI mints a food-identifier namespace that others adopt	The crosswalk needs a PTFI column, or gets routed around
VMH answers the licence enquiry	`MASTER_CROSSWALK.tsv` either clears for commercial use or migrates to Human-GEM
An official FoodOn → FDC alignment ships	Our own join becomes redundant; adopt theirs and delete ours
A LAW-SPEC needs declared kinetics	libSBML enters at a module boundary — never COBRApy
FOBI resumes releases	Re-check its licence header; the registries will still be stale
Biomappings covers a lane our crosswalk holds	Contribute the lane upstream instead of maintaining it here

Nothing here is vendored¶

To be explicit about the thing the issue was actually worried about: this repository redistributes rows from exactly four sources — VMH/Recon3D, FDA inventories, Open Food Facts and USDA FDC. All four are declared, with licence, conflict and remedy, in THIRD-PARTY-DATA.json, and check_third_party.py fails the build if that register drifts. No project on this page is vendored, and none should be.

This video is "PTFI Science Symposium: Molecules to Meaning with PTFI's First 500 Foods," published by the Food EDU channel.

While the full transcript isn't currently available to extract specific quotes, the presentation centers on the Periodic Table of Food Initiative (PTFI) and their milestone of mapping their first 500 foods. Based on the symposium's broader context, here is an overview of the core concepts covered:

Foodomics and Molecular Diversity: The initiative moves beyond traditional macronutrient tracking to map the "dark matter" of nutrition. PTFI uses high-resolution untargeted metabolomics and glycomics to capture a vastly broader array of small molecules, carbohydrates, and proteins across diverse food sources.
Contextual Metadata: The project integrates molecular data with environmental, agricultural, and cultural contexts, treating a food's nutritional profile as something dynamically linked to its ecosystem rather than a static metric.
Standardized Data Architectures: A major focus of PTFI is establishing standardized analytical protocols so that global researchers can contribute to and leverage a unified, comprehensive database.

The PTFI's approach represents a significant step forward in computational nutrition science. By establishing comprehensive, machine-readable schemas that account for deeper molecular profiles and environmental metadata, their work aligns closely with the goal of transforming biological and nutritional evidence into structured, AI-ready data ecosystems—moving beyond the constraints of traditional, flat databases like USDA FoodData Central or Open Food Facts.

Both inventories exist for the same job — “what is in food?” — and they sit at different layers of that job.

PTFI 1,650 is a priority specimen list: foods the field decided are worth measuring next.
FAO INFOODS is a standards-and-tables network: how nations compile, name, and share the ~150 nutrient values we already use for guidelines and surveys.

PTFI checked its list against INFOODS and USDA and found most of it missing. That is the relationship.

1. The 1,650 priority foods¶

Published March 2024 with the Nature Food comment. Dataset: doi:10.7910/DVN/M9TA1H (Harvard Dataverse; guestbook required to download). Compilation: ~40 regional experts + existing lists, then de-duplicated. Critically endangered / endangered / vulnerable wild taxa were dropped unless already domesticated.

How the list was built (the interesting part)
Not “top calories.” Facets used to nominate:

biology (plants, animals, fungi, bacteria, algae, insects, one lichen)
tissue (leaf, root, seed, fruit, muscle, egg…)
geography (origin vs where it is eaten)
domestication vs wild / underutilized
processing (raw → fermented → cooked)
whether it is a meal center or a garnish
staple vs festive
affordability

About one-third (542) were nominated by more than one source — those are the consensus foods.

Shape of the 1,650

Slice	Count (approx.)
Plants	~1,260
Animals	~376
Vegetables (broad)	417
Leafy vegetables	175
Roots & tubers	108
Nuts & seeds	125
Herbs & spices	63
Aquatic plants	10
Land animals	126
Aquatic animals	115
Insects	73
Invertebrates	114
Vertebrates	262

By region of importance

Scope	Foods
Global (widely cultivated/consumed)	476
Asia	355
Pacific	311
Americas	290
Africa	176
Europe	42

Fruits ~30%, vegetables ~25%, nuts/seeds ~8%, land animal products ~8%, aquatic ~7%. Half of world calories still come from wheat, maize, and rice; this list is the opposite bet.

Coverage against existing tables (the finding that launched the project)

~22% (362) appear in USDA FoodData Central
~25% (405) appear in at least one of the three major FAO/INFOODS nutrient databases they checked
~62% are absent from those global public tables

So the 1,650 is not “foods we already have rows for.” It is a gap map. Over 1,000 of the world’s culturally important foods have no internationally comparable nutrient profile, let alone a metabolome.

What the list is not
It is not composition data. No protein, no lycopene, no dark-matter features. It is a queue. The first measured slice (500 foods / ~250 species, ~24k metabolomic features) is a different product, sitting on MarkerLab and the AHA Precision Medicine Platform.

Dataverse package (if you fill the guestbook): codebook, food-product table, category table, IUCN-excluded foods, references.

2. FAO INFOODS¶

INFOODS = International Network of Food Data Systems. Founded 1984 under FAO / UNU. It is not one database. It is the operating system national food-composition tables are supposed to plug into.

What it actually is

Standards
Food description guidelines (single vs mixed foods, facet strings)
Tagnames — unique IDs for components so VITA and “vitamin A” and “retinol equivalents” do not collide
Data interchange format
Food-matching guidelines (how to map a survey item onto a table row)
Unit / denominator conversion guidelines
Pre-publication data-check guidelines
Evaluation framework for published tables
A global directory of national and regional food composition tables (started 1988, still updated). FAO’s current catalogue lists tables country-by-country — Albania 2022, Angola 2025, AUSNUT, ÖNWT, Bangladesh, Vietnam, Zambia, etc. Inclusion is not endorsement.
FAO/INFOODS’s own compiled databases (Excel + user guides, free):

Database	Role
AnFooD 2.0	Archival analytical-only values. ~1,940 foods, ~364 components. No imputed gaps.
BioFoodComp 4.0	Biodiversity: cultivars, landraces, wild foods. Earlier versions compiled thousands of biodiverse entries and hundreds of components. Closest INFOODS cousin to PTFI’s mandate.
WAFCT 2019	West Africa regional user table (update of 2012).
uPulses / PulsesDM	Pulses, including dry-matter basis for standards work.
uFiSh 1.0	Fish and shellfish, raw/cooked/processed.
PhyFoodComp 1.0	Phytate + Fe/Zn/Ca + molar ratios. ~3,377 entries.
Density DB, supplement DB	Utility tables.

Compilation Tool — an Excel system with tagnames, recipe calculation, yield and retention factors. Built so countries without custom software can still compile to INFOODS rules.
Regional centres — AFROFOODS, LATINFOODS, EUROFOODS, etc. — the human network that actually updates national tables.

What a typical INFOODS-shaped table contains
Proximates, minerals, vitamins, sometimes amino acids and fatty acids. On the order of 50–150 named nutrients, occasionally a few hundred if a specialist table (phytate, biodiversity). Values are per 100 g edible portion, documented by method and source. Recipes are calculated from ingredients + retention factors. That is the data behind dietary guidelines, food-balance sheets, and most intake surveys.

Hard limits (why PTFI exists next to it)

Components stop at the nutrient panel. No untargeted metabolome, no 20k unnamed features.
Most national tables average a commodity (“tomato, raw”), not cultivar × soil × process.
Coverage is uneven: staples and European/US commodities are dense; orphan crops, insects, fermented regional foods, and wild foods are thin — which is exactly what BioFoodComp tried to patch and what the 1,650 list still found missing.
Harmonization is tagname-level, not instrument-level. Two labs can both report FE and still not be comparable the way PTFI’s 33-compound retention standard is designed to be.

Portal: fao.org/infoods
FAO-built DBs: fao.org/infoods/.../faoinfoods-databases

How they fit together¶

INFOODS / national FCDBs     PTFI
─────────────────────────    ─────────────────────────
Policy + survey layer        Measurement + discovery layer
~150 nutrients               20k+ features + proteome/lipidome
Country tables, recipes      Standardized multi-omics specimens
Tagnames, food matching      Internal standards, MarkerLab
“Tomato, raw, 100 g”         “This tomato, this farm, this assay”
1984–ongoing                 2021–, first 500 foods public

PTFI’s own overlap check is the punchline: of 1,650 foods chosen as globally or regionally important, only about a quarter have an INFOODS-class nutrient row. BioFoodComp is the INFOODS attempt to close the biodiversity gap with the old analyte list. PTFI is the attempt to change the analyte list.

For a master nutrition schema that is the clean split:

INFOODS tagnames + national table rows → NutrientAmount (energy, Fe, folate, phytate…) with method and denominator.
PTFI specimen + feature → MolecularFeature (annotated or dark) with assay, abundance, and production metadata.
The 1,650 list → coverage backlog: foods that should exist as FoodConcept objects even before they have either a tagname row or a metabolome.

If you want a next concrete step: download AnFooD 2.0 and BioFoodComp from FAO (no guestbook), pull the 1,650 table after filling Dataverse’s form, and join on scientific name. The non-matches are the food-genome gap.