liveBook · Manning

Similarity metric

As explained in Build AI Drug Discovery Pipelines

A similarity coefficient is a numerical measure that quantifies the resemblance between two sets or binary representations.

A similarity metric numerically measures how much two molecules resemble each other through shared features. In ligand-based screening, it ranks library compounds against a molecule with known activity, helping prioritize candidates for experimental validation. section 2.1.3

The result depends on molecular representation as well as the metric. Molecular fingerprints act as binary feature vectors, so similarity compares their shared non-zero features rather than treating structural resemblance as proof of matching biological function. section 2.1.3 section 2.5.1

Common coefficients for binary fingerprints include:

  • Tanimoto, also called Jaccard similarity, divides the number of shared non-zero features by the number of non-zero features in either vector. section 2.5.1
  • Dice similarity doubles the shared-feature count and divides it by the sum of the non-zero feature counts in both vectors. section 2.5.1

From feature space to molecules

An apple at coordinates (5, 3) and an orange at (3, 4) illustrate how nearby points share feature values. Molecular fingerprints provide an analogous feature-space representation for comparing compounds. section 2.5.1

Figure 2.3 Structure-property relationships are nuanced. Discovering similar molecules doesn’t guarantee desired, similar function with respect to a query molecule.

Use the structure–property relationship figure to see why finding structurally similar molecules does not guarantee similar functionality. section 2.1.3

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Defines feature-space similarity and gives Tanimoto and Dice formulas, but defers broader representation and search-strategy choices.
Examples
Uses an apple–orange feature-space analogy, binary molecular fingerprints, coefficient calculations, and antimalarial compound-library searching.
What it assumes
Introduces feature spaces, molecular fingerprints, and binary-vector coefficients before applying similarity to screening.
What you can do after
Evaluate molecular resemblance with Tanimoto or Dice coefficients and interpret its limits as a proxy for biological function.
How it connects
Connects molecular fingerprints, feature representation, similarity searching, structural similarity, diversity, and clustering.

Also explained in

Compare how 14 books explain similarity metric →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Build AI Drug Discovery Pipelines. The section links point to the parts of the book it draws on.