LEB

LEB identifies biclusters in gene expression matrices by localizing and extracting correlated submatrices of genes and experimental conditions to reveal patterns not apparent from row- or column-wise clustering.

Key Features:

• Problem definition: Finds subsets of genes (rows) and experimental conditions (columns) that exhibit strong correlation within the subset but not necessarily across the entire matrix.
• Localization: Identifies local neighborhoods within the data matrix that are likely to contain correlated structures.
• Graph-theoretical techniques: Employs graph-theoretical methods to detect localized regions of potential biclusters.
• Extraction: Extracts biclusters from the confined neighborhoods identified by the localization step.
• Search-space reduction: Reduces the overall search space by focusing subsequent extraction on small localized regions.
• Evaluation: Validated on both artificial and real gene expression datasets demonstrating accurate identification of meaningful biclusters.

Scientific Applications:

• Gene expression analysis: Identification of condition-specific co-regulated gene modules in gene expression datasets.
• Biological process inference: Reveals patterns that can suggest underlying biological processes and pathways.
• Disease mechanism investigation: Identifies gene–condition associations relevant to disease mechanisms.
• Regulatory network inference: Supports discovery of genetic interactions and regulatory relationships.

Methodology:

Two-step process: Localization using graph-theoretical techniques to identify local neighborhoods likely to contain correlated submatrices, followed by extraction of biclusters from those reduced search-space neighborhoods.