Pairwise gene GO-based measures for biclustering of high-dimensional expression data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Biclustering algorithms identify genes with similar expression patterns across samples.
• Gene Ontology (GO) provides functional information to guide biclustering.
• GO semantic similarity measures quantify functional relationships between genes.

Purpose of the Study:

• To evaluate a scatter search-based biclustering algorithm integrated with GO information.
• To analyze the impact of different GO measures on algorithm performance.
• To identify functionally coherent gene biclusters relevant to biological insights.

Main Methods:

• A scatter search algorithm was developed to optimize a merit function incorporating GO semantic similarity.
• Two distinct gene pairwise GO measures were investigated.
• The algorithm was tested on yeast and human cancer gene expression datasets.

Main Results:

• GO-driven biclustering successfully identified groups of genes with shared functionality.
• Both yeast and human datasets showed improved biclustering results with GO integration.
• One specific GO measure demonstrated superior performance on large-scale, high-dimensional human cancer datasets.

Conclusions:

• Integrating biological knowledge, specifically GO information, significantly improves biclustering.
• The choice of GO measure is critical, especially for large and complex gene expression datasets.
• This approach offers a valuable tool for exploring clinically relevant gene functions in cancer research.