BitBIRCH

BitBIRCH clusters large molecular libraries using a tree structure and the instant similarity (iSIM) formalism to achieve O(N) time scaling and reduced memory requirements for processing binary fingerprints.

Key Features:

• Time and Memory Efficiency: BitBIRCH achieves O(N) time scaling and reduces memory requirements compared to standard clustering methods.
• iSIM Formalism: The method leverages the instant similarity (iSIM) formalism to process binary fingerprints and compute Tanimoto similarity.
• Tree Structure: BitBIRCH uses a tree structure similar to the Balanced Iterative Reducing and Clustering using Hierarchies (BIRCH) algorithm.
• Clustering Refinement Options: The package includes multiple options for controlling the tree structure to improve the quality of final partitions without compromising efficiency.
• Post-processing Tools: New post-processing tools help dissect clustering information.
• Parallel/Iterative Approximation: Strategies enable handling of billion-sized sets through parallel and iterative BitBIRCH approximation.

Scientific Applications:

• Large Molecular Library Clustering: BitBIRCH clusters million- and billion-sized molecular libraries, including clustering of one billion molecules in under 5 hours using parallel/iterative approximation.
• Chemical Space Analysis: The tool dissects chemical space through clustering of extremely large molecular libraries for chemical applications.

Methodology:

BitBIRCH employs a tree structure similar to the BIRCH algorithm with O(N) time scaling, leverages the iSIM formalism to process binary fingerprints using Tanimoto similarity, and uses parallel/iterative approximation strategies for handling large sets.