A generalized clustering problem, with application to DNA microarrays

Area of Science:

• Data Science
• Bioinformatics
• Statistics

Background:

• Traditional cluster analysis assumes a single, unique clustering structure for observations.
• This assumption is limiting as data can exhibit multiple valid clusterings based on different variable subsets.

Purpose of the Study:

• To develop and present an algorithm for identifying multiple, distinct clustering structures within a dataset.
• To generalize cluster analysis beyond the assumption of a unique underlying structure.

Main Methods:

• Proposes a novel algorithm that simultaneously clusters variables and observations.
• Variable clustering uses a dissimilarity measure based on nearest-neighbor graphs.
• Observation clustering employs weighted distances, with weights derived from variable clusters.

Main Results:

• The algorithm can identify multiple clustering structures.
• The number of identified clustering structures is determined by the number of variable clusters.
• Demonstrates applicability to gene expression data analysis.

Conclusions:

• The proposed method effectively estimates multiple clustering structures, offering a more comprehensive data analysis approach.
• This generalized clustering framework is particularly relevant for complex datasets like gene expression data.