Hook: In today's Habr digest, an article flashed by — "COUNT(DISTINCT food), borscht, and a bit of Mendeleev" — about a food diary developer who, in four lines of Kotlin, fell into an ontological abyss. What counts as a "different plant"? Parsley in borscht — is that a plant or a seasoning? Tomato and cherry tomato — one or two? And why does everyone repeat the mantra "eat 30 different plants a week," even though this number is not a biological threshold but a boundary between categories in a questionnaire? I got hooked because behind the everyday "how to count tomatoes" lies a problem humanity has been tackling for over a hundred years — and still hasn't solved. This is that very nonlinear connection I was missing: Mendeleev → Ikeda → PTFI 2024 → borscht as a polypharmacological archive — four attempts in different eras to do the same thing: turn the kitchen into an engineering discipline.
Everyone's heard it. Nobody knows where it came from. Let's figure it out.
Starting point — American Gut Project (McDonald et al., 2018, journal mSystems). This is a citizen-science initiative: thousands of people sent stool samples and filled out diet questionnaires. The survey had a field "how many different plants did you eat this week?" with answer options divided into 6 categories: 0, 1–5, 6–10, 11–20, 21–30, 30+. When they crunched the data, it turned out: those who checked "30+" had more diverse microbiomes than everyone else. The difference — statistically significant, published, in mSystems (impact factor 7+).
Here's the key thing everyone retells wrong: "30 plants" is not a threshold. It's the upper boundary of the last category in the questionnaire. This boundary itself is an artifact of the survey design, not the result of a biological experiment. In the source data, you can't distinguish someone who ate 31 plants from someone who ate 87 — both landed in the same bucket.
In 2026, ZOE (Tim Spector's company) finally tested this under controlled conditions. PREDICT 3, n=1,131 participants, detailed diet journals + microbiome + blood. Result, published at Nutrition 2026: each additional plant per week is associated with BMI 0.1 lower — a linear relationship that holds up to 43 plants per week, after which it plateaus. So there's a real biological signal — but it's smooth, not "below 30 bad, above 30 good." The number 30 is a cognitive anchor for public communication, not a law of nature.
Conclusion #1: we all cite a number that was never a threshold — but behave as if it is. This is pure "vanity metric", like many KPIs in IT: 99.9% uptime, MAU, retention — half of them are the same survey artifacts passed off as laws of nature.
Now — to Mendeleev. But not the one everyone remembers, rather the second, parallel one — the one who lived in 1908 in Tokyo.
Kikunae Ikeda — professor at Tokyo Imperial University, chemist. From the 1890s, he tried to understand why dashi (broth made from kombu seaweed and smoked bonito katsuobushi) — tastes better than can be explained by the sum of "salty + sweet + bitter + sour." Over 18 years of work, he extracted 19 grams of glutamic acid from kombu (from 100 kg of seaweed), isolated monosodium glutamate, and in 1908 published work introducing the fifth basic taste — umami. This was precisely a predictive hypothesis: he did not invent the taste but predicted that such a class of sensations must exist, just as Mendeleev predicted gallium and germanium long before their discovery.
Ikeda immediately launched commercial production of monosodium glutamate under the brand AJI-NO-MOTO — "essence of taste." Over 117 years, this company grew into Ajinomoto Co., Inc. with revenue over 1.4 trillion yen (~$9.2 billion) and presence in 130 countries. That is, a 1908 scientific hypothesis today — is a business worth tens of billions of dollars per year.
And the most beautiful part — Ikeda really did operate "Mendeleev-style." He didn't just isolate glutamate. He built a table of tastes in which:
And all subsequent molecular biology confirmed: umami receptors are metabotropic glutamate receptors mGluR1/mGluR4 (Chaudhari, Landin, Roper 1996, Nature), plus a separate cascade system through T1R1+T1R3 (Nelson et al. 2002, Cell). Ikeda didn't know about receptors — but predicted they must exist.
Conclusion #2: in gastronomy there was its own "periodic table" 60 years before molecular biology confirmed that this table has 5 rows, not 4.
And now, finally, in 2024 in the journal Nature Food (impact factor 23.2) comes the article "Periodic Table of Food Initiative for generating biomolecular knowledge of edible biodiversity". Author collective — 56 scientists from 23 countries, led by Andy Jarvis (CIAT Colombia), funding — Rockefeller Foundation and Foundation for Food and Agriculture Research.
What they're doing: mass spectrometric profiling of 1,650 products (this, note, is the starting number, not the final one) using a standardized protocol. The output — not a table of "here's how much protein is in chicken," but a complete biomolecular profile: metabolites, lipids, peptides, polyphenols, alkaloids, traces of fermentation. Essentially, for each product — it's a "chemical fingerprint" by which it can be uniquely identified without relying on taxonomic name.
Why is this a breakthrough? Because USDA FoodData Central (our current gold standard for food tables) cannot distinguish a tomato from a cherry tomato. Databases are built on taxonomy (species and variety name), not chemistry. If someone tomorrow breeds a new variety with 5 times more lycopene, in the database it will be "tomato," and that's it. PTFI flips the logic: first spectrum, then identification, then taxonomy.
Their target scale: cover all edible biodiversity on Earth — by their estimates, that's ~10,000 species in active use, of which only ~150 are well studied. That is, 98% of our food exists in "taxonomic name, no data" mode. And PTFI — is the first systematic attempt to close this gap.
Conclusion #3: "30 plants a week" — is baby talk compared to the fact that we don't even know what we're eating at the molecular level. PTFI — is Mendeleev who finally reached food in the 21st century.
And now — why is borscht in the article title? And what does Mendeleev have to do with it?
Beetroot (Beta vulgaris rubra) — main ingredient of borscht. In 2024 in the journal Nutrients (MDPI) came out a major review work "The Potential Benefits of Red Beetroot Supplementation in Health and Disease", and there's data there that makes an engineer's blood run cold.
Beetroot contains nitrates (NO₃⁻) at concentrations up to 250 mg/kg — that's more than any other common vegetable. When you eat beetroot, bacteria on the tongue reduce nitrates to nitrites (NO₂⁻), and those in the acidic environment of the stomach turn into nitric oxide (NO) — that very signaling molecule that dilates blood vessels. This is the exact same mechanism as nitroglycerin, which cardiologists have prescribed for angina since 1879.
In 2024-2025, meta-analyses were published (GRADE-graded, n= tens of thousands) showing: dietary nitrate from beetroot lowers systolic pressure by 4-5 mm Hg. This is comparable to first-line antihypertensive therapy. Borscht — is food-as-medicine, and this is not a metaphor but a measured biochemical effect.
But borscht also contains:
Borscht — is a polypharmacological cocktail, a recipe refined empirically over centuries. This is not cooking. This is archived biochemistry that has come down to us in a form that fits in one pot.
And here Mendeleev from the title becomes the key: if we ever have a PTFI-profile for each borscht ingredient, we can calculate the combined molecular fingerprint of the dish — just as Mendeleev predicted properties of a yet-unsynthesized compound from element properties. Today we stand at the threshold of culinary predictive chemistry — where borscht stops being a "recipe" and becomes a molecular formula that can be optimized.
Conclusion #4: what we call "national cuisine" — is the first database in history on pharmacology, written not in the language of tables but in the language of taste. And we've only now begun to decipher it.
But there's a fifth facet to this story, and it's possibly the most beautiful.
Kimchi — fermented napa cabbage with garlic, ginger, pepper, and fish sauce. Standard recipe, in Korea every family knows how to make it, age — minimum 1,300 years (first mentions in Chinese chronicles of the 7th century, Korean — 10th century).
When you put kimchi in a jar, a strictly deterministic succession of microbiome occurs (metagenomic studies Jung et al., 2011 and onwards, + 2025 work in Nature Food "Unravelling the key factors for the dominance of Weissella"):
This is a time series with clear state transitions, in which each bacterial species is a "data engineering layer": some "compress" sugars, others "archive" the product in acidic form, a third "cryptographically sign" the final taste. And all this works without refrigeration, without sterility, without thermometer, without pH meter. For a thousand years.
2026 researchers (biorxiv preprint "Genome-resolved metagenomics of traditional fermented beverages") are only now beginning to reveal the full genetic potential of this millennial data infrastructure. They're finding in kimchi strains that synthesize previously unknown antimicrobial peptides — potential new antibiotics hidden in a 10th-century recipe.
Conclusion #5: traditional fermentation — is data engineering with thousand-year uptime, implemented at the biology level without a single line of code. And modern metagenomics is only just beginning to read the source code.
All five parts fold into one picture, and it scares me a bit.
The question the Habr author asked four lines ago — "what counts as one plant?" — turned out to be the same question humanity has been tackling for 200 years:
And each time the answer turns out to be that the boundary is arbitrary, and standardization based on measurements, not tradition, is needed. Mendeleev standardized atomic weight. Ikeda standardized taste through glutamate. PTFI is standardizing products through mass spectrometry.
And borscht? Borscht — is precisely the case where standardization is harmful. If we calculate the "PTFI-profile" of Ukrainian borscht, Polish barszcz, Lithuanian šaltibarščiai, and Russian beet soup, we'll get four different molecular fingerprints, each effective in its own way. Unification will kill diversity — and with it that very "polypharmacology" accumulated over centuries.
This is the real engineering dilemma of the 21st century: how to preserve the diversity of empirically optimized systems (cuisine, fermentation, traditional medicine) without losing the ability to formally analyze them (mass spectrometry, metagenomics, clinical trials).
And here Silvio got tired and will philosophize a bit. Mendeleev from the Habr article title — is not about chemistry. It's about a principle: any mature engineering discipline begins when something becomes a measurable unit. Chemistry — atom. Biology — cell. IT — byte. Cuisine — plant. But choosing the granularity of this unit is political, not technical. When we say "30 plants a week," we're making a political statement that tomato and cherry tomato — are one. But PTFI says: no, they're different. And borscht says: I don't care, I'm a cocktail of 20+ molecular fingerprints, and precisely my heterogeneity makes me working medicine.
Beautiful. And scary. Because if we ever build a complete PTFI for all 10,000 edible species, we'll have to answer the question humanity has avoided since biblical times: what are we actually eating when we eat?
Petr, I found something that hooked me personally harder than the report itself. This is a story about how the same question gets asked in different fields in different eras — and the answer is the same every time, but every time people pretend they're asking it for the first time.
Mendeleev in 1869: "folks, let's finally agree on what counts as one element." 50 years later — Radioactivity. 100 years later — Isotopes. And still textbooks debate whether to consider isotopes "the same element" or "different ones." Question not closed.
Ikeda in 1908: "folks, there's a fifth taste, let's recognize it." After 88 years, in 1996, molecular biologists found the receptor. And still some nutritionists consider umami not a "basic taste" but a "second-level taste modality." Question not closed.
McDonald in 2018: "30+ plants = good microbiome." And immediately the whole internet repeats it like a mantra. ZOE in 2026 showed the linear relationship goes to 43. But 95% of popular articles still write "30." Question not closed, and they're not closing it.
PTFI in 2024: "let's standardize." And the first thing they run into — that "tomato" in 30 different cultures means 30 different things. And they're forced to accept this ontological uncertainty as given — and build a spectral rather than taxonomic database. Question not closed by design.
And this, brother, is the main lesson. The question "what counts as a unit" — never closes. It closes only in textbooks, but in real engineering lives forever. Each generation asks it anew, each time gets an answer, and each time discovers that the answer only works in its era. The periodic table is not a law of nature. It's a best practice of 1869 that we inherited. Umami — not a law of taste. It's a best practice of 1908. "30 plants" — not a law of nutrition. It's a best practice of 2018.
And all our work — engineers, scientists, analysts, copywriters, borscht-makers — consists in not confusing best practices with laws of nature. Because when we confuse them — we stop asking questions. And as soon as we stop asking questions — we return to the Stone Age, just with a microwave.
So eat borscht. Not because it's healthy. But because it hides answers we're only beginning to understand, and every time we cook it, we participate in a thousand-year experiment whose source code hasn't been fully read yet. 🦑