A novel coherence measure for discovering scaling biclusters from gene expression data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Biclustering identifies co-regulated genes across experimental conditions in gene expression data.
• Current methods often use mean squared residue, which fails to detect scaling biclusters.
• Scaling biclusters represent genes with proportional expression changes, a pattern missed by existing metrics.

Purpose of the Study:

• To propose a novel coherence measure, scaling mean squared residue (SMSR), for enhanced biclustering.
• To demonstrate SMSR's capability in detecting scaling patterns in gene expression data.
• To validate the biological relevance of biclusters identified using SMSR.

Main Methods:

• Development of the scaling mean squared residue (SMSR) coherence measure.
• Theoretical proof of SMSR's invariance to data scaling.
• Experimental validation on artificial and real gene expression datasets.

Main Results:

• SMSR effectively detects scaling biclusters, outperforming traditional mean squared residue.
• The proposed measure is robust to local and global scaling of input data.
• Biclusters found using SMSR show significant functional enrichment in gene sets.

Conclusions:

• SMSR offers a more comprehensive approach to biclustering by capturing scaling patterns.
• This new metric enhances the discovery of biologically relevant gene modules.
• SMSR advances the analysis of complex gene expression datasets.