liveBook · Manning

Agglomerative clustering

As explained in Build AI Drug Discovery Pipelines

Agglomerative clustering is a bottom-up hierarchical clustering method that starts with singleton clusters and successively merges the closest clusters.

Agglomerative clustering supports compound selection by starting with one cluster per molecule and repeatedly merging the closest clusters until all compounds form one hierarchy. Molecular similarity can use Morgan fingerprints with Tanimoto distance, while intercluster distance can use single, complete, average, or Ward linkage. section 7.2.3

A dendrogram exposes alternative cluster cuts, allowing the selection strategy to match the number of compounds or the desired diversity of a screening library. section 7.2.3

Fit and sample diverse compounds

Convert molecules to fingerprints, fit scikit-learn AgglomerativeClustering with Ward linkage, choose a cluster count, then sample molecules closest to each cluster center. Compare the resulting library using fingerprint similarity and redundancy.

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Explains bottom-up merging, four linkage choices, dendrogram cuts, implementation, sampling, and diversity evaluation.
Examples
Demonstrates RDKit fingerprint conversion, Ward fitting, dendrogram truncation, cluster-center sampling, and redundancy measurement.
What you can do after
Apply it via RDKit and scikit-learn to cluster molecules, sample representatives, and assess library diversity.
How it connects
Connects hierarchical clustering, divisive clustering, molecular fingerprints, Tanimoto similarity, dendrograms, and diversity sampling.

Also explained in

Compare how 2 books explain agglomerative clustering →

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.