The science behind what pairs.

Every ingredient contains dozens to hundreds of volatile aroma compounds — molecules that evaporate and hit your nose when you eat. Two ingredients that share many of the same compounds tend to taste coherent together because your nose is already tuned to them. CompKitchen ranks 857 ingredients across 5,240 compounds and scores every possible pair by how much molecular overlap they share. High overlap + low recipe co-occurrence = a scientifically-backed pairing nobody's cooking yet.

We merge FlavorDB 2.0 (IIITD's public molecular flavor database), FooDB, PubChem, ChemTastesDB, Dr. Duke's USDA phytochemical database, and extractions from 44,648 peer-reviewed GC-MS papers mined via PubMed. Recipe co-occurrence data comes from a curated corpus of 612,557 recipes combining OpenRecipes (CC BY), Fandom Recipes Wiki (CC BY-SA), and public-domain cookbook scans from Project Gutenberg and the Internet Archive. Every compound-food link is cross-referenced against at least one published source. See the methodology page for the full data pipeline.

A pairing is flagged novel when two ingredients share a high fraction of compounds (compound_score ≥ 0.40) but almost never appear together in recipes (recipe_score ≤ 0.10). The chemistry says they work; the culinary tradition hasn't caught up. Examples: tomato × black currant shares 329 compounds but appears in only 7 of 195,676 tomato recipes. Pineapple × beans shares 281 compounds and appears in zero recipes.

Two layered scores. Compound score is the IDF-weighted Jaccard similarity of the two ingredients' volatile compound sets — shared compounds divided by total unique compounds, weighted so rare distinctive compounds count more than ubiquitous ones. Recipe score is the Normalized Pointwise Mutual Information (NPMI) of how often the two co-occur in real recipes. The final score blends them 50/50 compound-to-recipe with a synergy bonus when both are strong.

Flavor wheels are qualitative — someone drew them based on intuition. Foodpairing is a commercial tool using similar principles to ours but not fully open about its data. CompKitchen publishes every source, every scoring formula, and every ingredient profile. You can see exactly why two ingredients score high together — which compounds they share, at what concentrations, and how perceptually impactful each one is.

Two reasons. (1) Classic pairings often work on contrast (heat × cream, acid × fat) rather than compound similarity — those fall outside the compound-overlap framework but still work because balance matters. (2) Our category-diversity filter prevents the top results from all being herbs or all being berries, which can push true soul-mate pairings down the list if they're in the same family. The comparison tool at /compare shows raw compound overlap without diversity filtering.

Each ingredient is tagged with its dominant flavor family — fruity, green & herbal, roasted & savory, sulfury & pungent, sweet & caramel, spicy & warm, floral, woody & earthy, dairy & creamy, or sour & tangy. These labels come from analyzing each ingredient's compounds, weighted by IDF (rarity) and odor threshold (perceptual impact). You'll see the cluster tags at the top of every ingredient profile page.

Yes — every ingredient has dietary flags (vegan, vegetarian, kosher, halal, gluten-free, dairy-free, nut-free, egg-free, alcohol-free) plus Big-9 allergen tags (milk, eggs, fish, shellfish, tree nuts, peanuts, wheat, soy, sesame). These render as pills on every ingredient profile. Filtering by diet in the pairing tools is coming — right now it's displayed but not yet filterable.

Yes. Start with the flavor tools — type an ingredient and see its top pairings, or find a molecular substitute when you're out of something, or describe a mood ("smoky", "citrusy") and get matching ingredients. Every result links to a full compound profile so you can see why it's suggested.

Yes. Pro is automatically granted to anyone who signs in with an academic email (.edu, .ac.uk, .edu.au, .ac.jp, .ac.in, .ac.nz, .edu.sg, .ac.za, and most international academic TLDs) for 12 months, auto-renewing on each sign-in. Course instructors, research labs, and extension/incubator programs can apply for program-level access at /education. We ask (but don't require) that you cite the methodology paper if CompKitchen contributes to published research.

Pro ($12/mo or $99/yr) unlocks the full ranked pairing list (free shows top 5), full chemistry on every result (OAV where measured, odor threshold, aroma class, chemical class, CAS, key-aroma flag, volatility, odor description), the inverted compound index at /compound/{name} showing every ingredient that contains a given molecule, side-by-side ingredient comparison at /compare, and CSV export on every pairing run. Also includes saved pantries with notes, opt-in search history, and priority email support (24hr weekday reply). Free accounts get the top 5 pairings, all ingredient profiles, and the full recipe browser.

Yes — see the Enterprise page. The JSON API exposes every engine method (pair, novel, substitute, bridge, mood, pantry) plus batch endpoints for up to 50 ingredients per request. Used internally for the site and available to recipe platforms, product developers, and research teams.

Yes — if you see an ingredient missing or a compound profile that looks off, email [email protected] with the citation and we'll review. Every added compound-food link needs a peer-reviewed source.