Defining clusters from a hierarchical cluster tree: the Dynamic Tree Cut package for R

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Hierarchical clustering is a standard technique for analyzing genomic data.
• Traditional cluster detection relies on a fixed height cutoff for dendrograms, which can be suboptimal for complex data structures.

Purpose of the Study:

• To introduce novel dynamic branch cutting methods for improved cluster detection in dendrograms.
• To provide a flexible and automated approach for identifying clusters based on their shape.

Main Methods:

• Development of the Dynamic Tree Cut R package, implementing dynamic branch cutting algorithms.
• Application of methods to protein-protein interaction network data and simulated gene expression data.
• Optional integration with partitioning around medoids for enhanced outlier detection.

Main Results:

• Dynamic Tree Cut successfully identifies nested clusters, outperforming constant height cutoffs on complex dendrograms.
• The R package offers tunable parameters for flexibility across different applications.
• Demonstrated utility in analyzing biological network and gene expression data.

Conclusions:

• Dynamic Tree Cut provides a more robust and flexible alternative to constant height cutoffs for hierarchical clustering.
• The R package facilitates automated and accurate cluster detection, including nested structures and outliers.
• This approach enhances the analysis of complex genomic datasets.