Hybrid coexpression link similarity graph clustering for mining biological modules from multiple gene expression...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Genomic technologies generate vast gene expression datasets across species and conditions.
• Integrating these datasets overcomes limitations of single-dataset analysis for protein annotation and module discovery.
• Single-dataset approaches often yield spurious coexpression, hindering accurate biological insights.

Purpose of the Study:

• To develop a novel joint mining algorithm for integrating multiple gene expression datasets.
• To improve protein functional annotation and biological module discovery.
• To address the challenge of spurious coexpression in individual datasets.

Main Methods:

• A joint mining algorithm constructs a weighted hybrid similarity graph where nodes represent coexpression links.
• Edge weights in the graph combine topological and co-appearance similarities between coexpression links.
• Clustering of the weighted hybrid similarity graph identifies recurrent coexpression link clusters (modules).

Main Results:

• The algorithm successfully identifies recurrent coexpression link clusters (modules).
• Experimental results on human gene expression datasets demonstrate functional homogeneity of the identified modules.
• Enrichment analysis confirms modules are associated with Gene Ontology (GO) terms and KEGG pathways.

Conclusions:

• The proposed joint mining algorithm effectively integrates multi-species gene expression data.
• This approach enhances the accuracy of protein functional annotation and biological module discovery.
• The identified modules exhibit significant functional coherence, validated by GO and KEGG pathway enrichment.