Discovery of error-tolerant biclusters from noisy gene expression data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Microarray analysis seeks biclusters (coherently expressed genes across conditions).
• Existing algorithms struggle with noise, small biclusters, and overlapping patterns.
• Association mining for biclusters is limited by exact matching and binary data requirements.

Purpose of the Study:

• To develop a novel error-tolerant biclustering (ET-bicluster) model.
• To propose a bottom-up heuristic algorithm for discovering biclusters from real-valued gene expression data.
• To address limitations of existing methods in handling noise and real-valued data simultaneously.

Main Methods:

• Developed the ET-bicluster model for error tolerance.
• Implemented a bottom-up heuristic mining algorithm for sequential bicluster discovery.
• Validated the approach on real-valued S. Cerevisiae and Breast Cancer microarray datasets.

Main Results:

• ET-bicluster recovered larger gene sets with higher functional coherence compared to the RAP approach.
• Functional enrichment analysis using GO demonstrated biological significance.
• Statistical significance was confirmed through randomization tests.
• Biomarker discovery using MSigDB gene sets showed promising results.

Conclusions:

• The ET-bicluster approach is effective for functional module discovery.
• The method demonstrates the importance of incorporating error tolerance in biclustering.
• ET-bicluster enhances biomarker discovery from gene expression data.