Using Descriptor Counts in Clustering

Overview

• Use descriptor-count information to avoid comparing every possible pair of compounds during clustering.



• Clustering approach is agglomerative.



• First phase: form "small," tight clusters.
• BOOST algorithm.
• Completes in O(n) time.
• Substantially reduces dataset size.


• Second phase: cluster the reduced dataset.
• DiET algorithm.
• Complexity depends on dataset.
• Merge small clusters and single compounds to complete the clustering.