Six public APIs, no keys and no paywall. Queries are scoped to title and abstract rather than full text, because a full-text hit on Curcuma in a reference list is not evidence of anything. Everything retrieved is then passed through an interaction-vocabulary gate: a paper is admitted only if it actually describes an interaction, not merely because it names a plant and a drug in the same abstract. What survives is ranked against an eight-level evidence hierarchy, so a randomised trial outranks a review that cites it.
Every returned record is classified by regex against its title, abstract and publication type, then scored. Rank 1 is the strongest.
| Rank | Study type | Tier points | Recognised by |
|---|---|---|---|
| 1 | Systematic review / meta-analysis | 80 | (meta[- ]?analys systemat c rev ew cochrane) |
| 2 | Randomised controlled trial | 70 | (random [sz]ed random sed controlled dou le[- ] l nd plac… |
| 3 | Cohort / pharmacovigilance study | 60 | (cohort prospect ve study retrospect ve study cross[- ]se… |
| 4 | Case report / series | 50 | (case report case ser es a case of f rst report of) |
| 5 | Animal / in vivo study | 40 | (rats? m ce mouse rodent n v vo w star sprague) |
| 6 | Narrative review | 30 | (rev ew overv ew update on current perspect ves) |
| 7 | In vitro / in silico | 20 | ( n v tro m crosome cell l ne hepatocyte dock ng n s l co… |
| 8 | Unclassified | 10 |
Composite score = evidence tier + recency decay + a log-scaled citation term + a small bonus for open access, plus 3 points per interaction term matched. Recency never outranks study design by itself.
In Focused mode a record must use at least one of these 23 terms in its title or abstract, or it is rejected before ranking. Journal titles and author lists are excluded from the test on purpose — a journal named Drug Metabolism would otherwise admit everything it ever printed.
Terms shown in blue are specific enough to admit a paper on their own. The plain ones are general verbs, and they carry a condition: inhibits and induces are the two commonest words in natural-product literature — an extract inhibits an enzyme, an alkaloid induces apoptosis, an oil inhibits a bacterium, and none of those is a drug interaction. A paper matching only these is admitted just when the abstract also discusses metabolism, a transporter, a CYP enzyme, plasma levels or co-administration.
Each term is a regular expression rather than a literal, so potentiates also catches potentiated and potentiation, and CYP450 inhibitor catches CYP3A4 inhibition. Every result shows which terms admitted it, and the rejected count is reported alongside the results rather than hidden.
The engine builds better queries when it can pull marker compounds and synonyms out of the record.
|
Senna alexandrina
Senna |
|
Hypericum perforatum
St. John's Wort |
|
Camellia sinensis
Green Tea |
|
Piper methysticum
Kava |
|
Achillea millefolium
Yarrow |
|
Hydrastis canadensis
Goldenseal |
|
Echinacea purpurea
Echinacea |
|
Rubia cordifolia
Manjistha |
|
Physostigma venenosum
Calabar Bean |
|
Uncaria tomentosa
Cat's Claw |
10 cached result sets, 1 re-serves. Entries expire after 72 hours.
Caching keeps the app usable when NCBI rate-limits an unauthenticated key, which it will under classroom load.