New resampling method for evaluating stability of clusters

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Hierarchical clustering is crucial for analyzing gene expression data.
• Assessing cluster stability is a significant challenge.
• Existing methods like bootstrapping have limitations.

Purpose of the Study:

• To introduce a new resampling method, continuous weights, for evaluating cluster stability in hierarchical clustering.
• To address limitations of traditional bootstrapping methods.

Main Methods:

• Development of a novel resampling technique using continuous weights.
• Comparison with traditional bootstrapping via simulation and real-world datasets.

Main Results:

• Continuous weights demonstrate advantages over bootstrapping, particularly for datasets with few observations or low fold-change.
• This method retains full data dimensionality, unlike bootstrapping.

Conclusions:

• Continuous weights are recommended for both small and large datasets.
• This method provides comparable or superior results to conventional bootstrapping.