liveBook · Manning

Redundancy ratio

As explained in Build AI Drug Discovery Pipelines

A redundancy ratio is the fraction of unique compound pairs whose similarity exceeds a specified threshold.

Redundancy ratio measures how many unique compound pairs are highly similar: count pairs with Tanimoto similarity above a threshold, then divide by all unique pairs. Lower values indicate less redundancy. section 7.2.3

Calculate it for a sampled compound set by:

  • Counting each pair once while excluding self-pairs. section 7.2.3
  • Counting pairs whose Tanimoto similarity exceeds 0.85 and dividing by n(n−1)/2. section 7.2.3

Comparing sampling methods

Taylor–Butina sampling produced a much lower redundancy ratio than agglomerative-clustering sampling, indicating fewer highly similar compound pairs. The 0.85 cutoff is a dataset-dependent heuristic, not a universal activity rule. section 7.2.3

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Defines the ratio and gives its pair-counting implementation, but defers universal threshold selection and activity guarantees.
Examples
Compares Taylor–Butina and agglomerative-clustering samples using a 0.85 Tanimoto threshold.
What it assumes
Uses molecular fingerprints, Tanimoto similarity, and pairwise compound comparisons.
What you can do after
Apply it via RDKit to compare redundancy across sampled compound sets.
How it connects
Connects molecular fingerprints, Tanimoto similarity, compound diversity, and clustering-based sampling.

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.