Generalized gene co-expression analysis via subspace clustering using low-rank representation.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene Co-expression Network Analysis (GCNA) is crucial for identifying biologically functional gene modules.
• Current GCNA methods predominantly rely on gene correlation, potentially missing modules with complex relationships.
• There is a need for advanced GCNA algorithms that utilize alternative similarity measures to uncover novel biological insights.

Purpose of the Study:

• To develop a generalized GCNA algorithm that identifies gene modules beyond high correlation.
• To leverage subspace clustering for constructing gene co-expression networks.

Main Methods:

• A novel generalized gene co-expression analysis algorithm employing subspace clustering.
• Construction of gene co-expression networks using low-rank representation.
• Identification of gene co-expression modules via local maximal quasi-clique merger.

Main Results:

• The proposed method successfully identified biologically meaningful gene modules with diverse functions across microarray and single-cell RNA sequencing data.
• Gene modules discovered by this approach exhibited prognostic value.
• The method identified modules containing genes that are not all highly correlated, offering novel biological insights.

Conclusions:

• The developed GCNA method utilizes subspace clustering, offering a powerful alternative to correlation-based similarity measures.
• This generalized approach provides complementary insights to existing GCNA algorithms.
• The findings suggest broader applications in analyzing gene expression datasets for novel biological discoveries.