Similarity search
Ranks a set of candidate compounds by fingerprint similarity to one query compound, using Tanimoto similarity over a Morgan or MACCS fingerprint. Use to find which of several saved candidates is closest to a reference structure, or to triage a batch of hits.
Similarity search
What it does
Takes one query compound and up to five candidates, all saved in the project, and ranks the candidates by how similar each one is to the query — the standard "find me something like this" question in cheminformatics.
Each candidate gets a Tanimoto similarity score between 0 and 1, where 1 would mean an identical fingerprint and 0 means no shared features at all. The result is a ranked list, most similar first.
How similarity is measured
A molecule is turned into a fingerprint — a compact encoding of the substructures it contains — and two fingerprints are compared with the Tanimoto coefficient, the fraction of features they share out of all the features either one has.
Two fingerprint types are offered, and they can genuinely disagree:
- Morgan (ECFP4), the default — encodes circular neighborhoods around each atom out to a radius of 2 bonds. Sensitive to the general shape and connectivity of a molecule.
- MACCS keys — a fixed set of 166 predefined structural features (rings, functional groups, and so on). Coarser, and sometimes closer to what a chemist means by "similar" when the Morgan ranking feels too literal.
If the two disagree on which candidate is "most similar," that's not a bug — they're measuring genuinely different notions of similarity, and which one matches the question being asked depends on the project.
Reading the output
The ranked list and each candidate's score are what to read. A candidate that couldn't be parsed is skipped rather than failing the whole run, and is named in a warning rather than silently dropped.
Input
One query compound and one to five candidates, each saved as an SDF.
Important limitations
This tool ranks, it doesn't generate. It can only compare compounds already saved in the project — it has no way to propose a new molecule that isn't one of the candidates given to it. If what's needed is a new scaffold rather than a ranking of existing ones, this isn't the tool for that.
A high score is not a claim about biological activity. Structural similarity is a starting point for prioritizing candidates, not a prediction of how either compound will actually behave.