Topological biclustering ARTMAP for identifying within bicluster relationships

Area of Science:

• Computational biology
• Data mining
• Machine learning

Background:

• Biclustering is vital for exploratory data analysis in gene expression and social networks.
• Topological learning excels at identifying complex, connected data regions.
• Existing methods struggle with simultaneous data reduction and topological feature detection.

Purpose of the Study:

• To introduce TopoBARTMAP, a novel hybrid biclustering algorithm.
• To enhance module extraction and data reduction capabilities.
• To leverage topological associations for improved biclustering.

Main Methods:

• Combines adaptive resonance theory (ART)-based biclustering ARTMAP (BARTMAP) and topological ART (TopoART).
• Develops a graphical representation for gene bicluster associations.
• Benchmarks TopoBARTMAP on 35 real-world cancer datasets and 12 synthetic datasets.

Main Results:

• TopoBARTMAP demonstrated statistically significant improvements over existing biclustering methods.
• The method excelled in identifying various bicluster types (constant, scale, shift, shift scale) in synthetic data.
• Refined graphical output effectively represents gene bicluster associations.

Conclusions:

• TopoBARTMAP offers superior performance in biclustering and module extraction.
• The integration of topological learning enhances the detection of complex data structures.
• This method provides a powerful tool for analyzing large-scale biological datasets, such as gene expression data.