Science · Plant Clinic

How the Plant Clinic Works

When you bring a pest or a diseased plant to the clinic, we match it to biological treatments that are registered with government agencies around the world. This page explains exactly how that matching works — and where the data comes from.

A — Two Paths Through the Clinic

The Plant Clinic has two entry points, depending on what you’re dealing with.

Insect path

Photo → AI identifies the insect family → we look up which biological treatments are registered against that family → you see matched products.

Plant disease path

Photo or search → AI identifies your plant → we look up known pathogens for that plant genus → you select the symptoms you see → you see matched treatments.

Both paths end at the same place: a grid of biological active ingredients, organised by how they work, with links to buy nationally registered products from retailers in your country.

B — How Insect Matching Works

Step 1: Identify the insect

Your photo is processed by BioCLIP 2, a vision model trained on 214 million organism photographs. We search against 302,000 arthropod species embeddings to find the closest match, returning the taxonomic family and order.

Multiple photos improve accuracy. On the IP102 pest benchmark — 22,619 images of pests in natural agricultural conditions — single-photo identification achieves 83% accuracy at the order level and 64% at the family level. With 3 photos from different angles, order accuracy rises to 90% and family to 66%.

Step 2: Look up treatments by family

We maintain a pest-family routing matrix — a table of 5,998 rows mapping every combination of insect order, family, and biological active ingredient. This table was built by:

  1. Extracting named pest species from 35 EPA-registered product labels (the actual government documents)
  2. Resolving each pest common name to its taxonomic family and order via GBIF (Global Biodiversity Information Facility)
  3. For families with direct EPA label evidence: marking them as “EPA-registered”
  4. For other families in the same order: imputing based on mode of action (e.g. if pyrethrin kills aphids via sodium channel disruption, it likely affects other Hemiptera families the same way)
346 families with direct EPA evidence
5,652 families imputed from order-level mode of action
13 active ingredient categories

Step 3: The 2×2 treatment grid

Matched treatments are displayed on two axes: speed (how fast the product works) and specificity (how targeted vs. broad-spectrum it is). This helps you choose: do you need a targeted preventive (like Bt for caterpillars), or an immediate broad-spectrum knockdown (like pyrethrin)?

C — How Disease Matching Works

Step 1: Identify the plant

You search by name or upload a photo. BioCLIP searches against 39,000 garden-ready species (or 140,000 across our full encyclopedia) to identify yours to genus level.

Step 2: Look up known pathogens

We query GloBI (Global Biotic Interactions) — a database of recorded interactions between organisms. For your plant’s genus, we retrieve all documented pathogen genera: the fungi, bacteria, and oomycetes that have been observed infecting plants in that genus.

Each pathogen is linked to a disease common name from the APS Common Names of Plant Diseases — the standard reference maintained by the American Phytopathological Society.

Step 3: Group by visual symptoms

To help you narrow down what’s affecting your plant, we group diseases into nine symptom categories based on what you can see: leaf spots, wilts, rots, rusts, mildews, cankers, root diseases, viral symptoms, and nematode damage. Each category shows reference photographs so you can visually match what’s happening on your plant.

The categorisation was performed by an LLM (large language model) classifying each of 2,778 disease-pathogen pairs against the nine categories — using whether the pathogen is a fungus, bacterium, or virus to help distinguish diseases with similar names.

Step 4: Match to treatments

For each disease, we identify which biological active ingredients are labelled for use against pathogens in that genus. The matching works through two bridges:

  1. Bacillus / Trichoderma bridge: We extracted target pathogen genera from EPA product labels for Bacillus subtilis, Bacillus amyloliquefaciens, and Trichoderma harzianum products. If your plant’s pathogen genus appears on those labels, the treatment is matched.
  2. Neem oil bridge: Clarified and cold-pressed neem oils have broad foliar fungicide claims. We match these to any foliar fungal disease based on the label’s general efficacy claims.
31,597 plants with disease profiles
407 pathogen genera in database
9 disease treatment categories

D — From Treatment Category to Product

Once a treatment category is matched (e.g. “Bacillus subtilis QST 713”), we look up purchasable products from CABI’s global product database. CABI tracks which biological pest management products are nationally registered in each country — meaning they have passed each government’s safety, and in some jurisdictions, efficacy reviews.

We detect your location and show products registered in your country first. Each product links to Amazon or eBay search results where you can compare prices and availability.

4,387 products across 20 countries
28 active ingredient categories

E — What We Can’t Do (Yet)

We believe in being upfront about the boundaries of this system. Here’s what you should know:

EPA labels are US-specific samples

Our product target extraction comes from US EPA-registered labels. We use these as representative samples for each active ingredient category, then match to products registered in other countries via CABI. The underlying biology is the same, but label claims may differ between jurisdictions.

Pest family imputation

94% of our pest-family matches are imputed from order-level mode of action, not from direct EPA label evidence for that specific family. When a product has been tested against aphids (Aphididae) and we impute it works against plantbugs (Miridae) because both are Hemiptera — that’s an inference, not a proven claim.

GloBI coverage varies

Some plant genera have hundreds of documented pathogen interactions; others have very few. A plant showing “3 known diseases” doesn’t mean it only gets 3 diseases — it means GloBI has documented interactions with 3 pathogen genera for that plant group.

We don’t diagnose

The Plant Clinic shows you what diseases are possible for your plant and helps you visually match symptoms. We don’t confirm which disease you actually have. For definitive diagnosis, send a sample to your local extension service or plant pathology lab.

Always read the label

Product registration requirements vary by jurisdiction. In the US, EPA registration requires safety data but does not require efficacy testing for crop uses — the manufacturer self-certifies efficacy claims. In the EU, Regulation 1107/2009 requires both safety and efficacy evaluation. Registration standards differ between countries — we do not endorse any specific product. Application rates, re-entry intervals, and crop restrictions vary — the label is the law.

F — Data Sources

BioCLIP 2

Vision model for organism identification. ViT-L/14, trained on 214M images. Paper

GloBI

Global Biotic Interactions. Open database of species interactions. globalbioticinteractions.org

APS Common Names

American Phytopathological Society disease nomenclature. apsnet.org

CABI Compendium

Global database of pest management products and national registrations. cabidigitallibrary.org

US EPA PPIS

Pesticide Product Information System. Source of all product label extractions. epa.gov

GBIF

Global Biodiversity Information Facility. Used for pest taxonomy resolution. gbif.org