ChemOnt

The predicted chemistry of a cluster’s product

ChemOnt is the ontology the platform uses to describe what kind of molecule a cluster is predicted to make. It complements the BGC class: the class describes the enzymatic machinery; ChemOnt describes the product.

What ChemOnt is

ChemOnt (Chemical Ontology) is a hierarchical classification of chemical compounds from the ClassyFire project. It organises molecules by structural features — rings, functional groups, carbon frameworks. The full ontology contains roughly 4,825 terms; the platform uses the subset relevant to microbial natural products.

The platform’s chemical predictions come from CHAMOIS, which assigns ChemOnt classes to a cluster’s proteins. These per-protein predictions are aggregated up to the whole cluster (see below).

How ChemOnt appears

On an iBGC

The iBGC detail panel’s “compound features” chip lists the cluster’s predicted ChemOnt classes, aggregated across its genes and shown as a small tree. Each node shows the number of contributing genes and a confidence percentage. (For validated clusters, curated compound names appear alongside.)

On a gene

The protein panel shows the ChemOnt class predicted for an individual protein, with its probability and weight — the per-gene predictions that roll up into the cluster’s overall chemistry.

Term format and confidence

  • Identifier: each term has a CHEMONTID:XXXXXXX identifier; the human-readable name is shown, with the identifier on hover.
  • Probability: each prediction carries a confidence between 0 and 1 — how sure the model is that the cluster makes a compound of that class.

Hierarchical structure

ChemOnt is a tree, broad at the top and specific at the bottom:

Organic compounds
  Lipids and lipid-like molecules
    Prenol lipids
      Terpene lactones
        Sesquiterpene lactones

A cluster can be annotated at several levels. Broader categories are generally predicted with higher confidence than specific ones (predicting “a lipid” is easier than “a sesquiterpene lactone”).

How ChemOnt differs from BGC class

BGC class ChemOnt
Describes The enzymatic machinery The chemical product
Based on Protein domains in the cluster Predicted product structure (CHAMOIS)
Example Polyketide Macrolides and analogues

One BGC class can map to many ChemOnt categories — Polyketide clusters can make macrolides, aromatics, or linear polyketides, each a different ChemOnt class.

Novel clusters and missing predictions

Highly novel clusters often have weak or absent ChemOnt predictions: the model was trained on characterised clusters and struggles to predict products for clusters unlike anything it has seen. A missing or low-confidence prediction on a high-novelty cluster is expected — and is itself informative, reinforcing that the product is unlike known chemistry.

Filtering by ChemOnt

The ChemOnt class filter is a searchable, hierarchical chip. Type a class name (e.g. “macrolide”), select one or more classes, and the catalogue narrows to clusters predicted to make them. Selecting a parent class includes its descendants. There is also the Chemical Search, which starts from a SMILES structure rather than a class.

Tips

  • Filter by a ChemOnt class when you have a target product type in mind — more precise than filtering by enzyme class alone.
  • Combine ChemOnt with the BGC class filter for precision (e.g. Polyketide × Macrolides).
  • For pure novelty hunting, lead with the novelty score and treat ChemOnt as supplementary — novel clusters often have weak predictions.