Improving the sensitivity of sample clustering by leveraging gene co-expression networks in variable selection

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Existing gene selection methods for clustering gene expression data often overlook gene interactions.
• Leveraging gene networks is crucial for utilizing relationship information and preserving clustering structures.

Purpose of the Study:

• To develop a novel gene selection method based on gene connectivity within co-expression networks.
• To improve the reliability of selected genes and the accuracy of sample clustering.
• To propose a module analysis approach for uncovering higher-order gene organization and novel sample partitions.

Main Methods:

• Gene co-expression network construction using hard and soft thresholding.
• Variable selection based on gene connectivity (expression similarity).
• Module analysis to identify groups of genes with topological similarity.

Main Results:

• Soft thresholding-based networks yield superior variable selection and clustering compared to hard thresholding and other filter methods.
• The proposed module analysis reveals biologically meaningful sample partitions missed by other approaches.
• The network-based gene selection method demonstrates enhanced reliability and clustering performance.

Conclusions:

• Gene co-expression network connectivity offers an effective strategy for variable selection in gene expression data.
• The developed method improves gene selection reliability and sample clustering outcomes.
• Module recovery facilitates the discovery of novel, biologically relevant sample partitions.