BicSPAM: flexible biclustering using sequential patterns.

Area of Science:

• Bioinformatics
• Computational Biology
• Data Mining

Background:

• Biclustering is vital for biomedical applications, identifying local regularities using order-preserving submatrices.
• Existing methods lack robustness, scalability, and flexibility, failing to handle symmetries or noise effectively.

Purpose of the Study:

• To introduce novel biclustering algorithms (BicSPAM) for flexible, exhaustive, and noise-tolerant bicluster discovery.
• To enable the detection of biclusters with symmetries and robustness to noise.

Main Methods:

• Developed biclustering algorithms based on sequential patterns (BicSPAM).
• Incorporated strategies for symmetry detection and efficiency gains using item-indexable properties and partitioning methods.
• Evaluated performance on synthetic datasets with varying noise and missing values, and on gene expression data.

Main Results:

• BicSPAM effectively captures symmetries and handles noise in bicluster discovery.
• Demonstrated scalability in memory and time, achieving superior recovery of hidden biclusters in synthetic data.
• Identified novel biclusters in gene expression data, corresponding to biologically relevant modules.

Conclusions:

• BicSPAM offers an exhaustive method for discovering flexible, order-preserving biclusters.
• This is the first approach to address order-preserving biclusters with symmetries and noise robustness.