Overlapping clustering of gene expression data using penalized weighted normalized cut.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Clustering gene expression data is crucial for understanding complex diseases and gene functions.
• Existing disjoint clustering methods are limited as genes can have multiple functions.
• Small sample sizes and uncertainty in genetic studies hinder definitive gene clustering.

Purpose of the Study:

• To develop an effective overlapping clustering approach for gene expression data.
• To address the multiplicity of gene functions and uncertainty in gene clustering.
• To improve the analysis of complex diseases using gene expression data.

Main Methods:

• Proposed a penalized weighted normalized cut (PWNCut) criterion.
• Based on the normalized cut (NCut) technique with an L1 norm constraint.
• Developed the 'pwncut' function within the R package NCutYX for implementation.

Main Results:

• The PWNCut approach demonstrated superior performance compared to existing methods in simulations.
• Analysis of The Cancer Genome Atlas (TCGA) breast and cervical cancer data yielded novel, biologically relevant findings.
• Results differed significantly from those obtained using alternative clustering methods.

Conclusions:

• The proposed overlapping clustering method effectively handles multiple gene functions and analytical uncertainty.
• PWNCut offers a more biologically sensible approach to gene expression data analysis, particularly for complex diseases.
• The R package NCutYX facilitates the practical application of this advanced clustering technique.