NNN

NNN clusters genes with similar expression profiles by constructing nearest-neighbor-based networks to identify overlapping cliques that represent functionally coherent gene modules from microarray data.

Key Features:

• Graph-based clustering: Uses a graph-based algorithm that represents genes and their relationships to define clusters.
• Mutual nearest neighborhoods: Builds an interaction network derived from mutual nearest neighborhoods rather than relying on absolute distance measures.
• Overlapping cliques: Identifies overlapping cliques within the interaction network to define clusters that can share genes.
• Nearest-neighbor focus: Emphasizes mutual nearest neighbors to detect clusters with high connectivity even when members are spatially separated in expression space.
• Unclustered gene handling: Leaves genes without sufficient mutual nearest-neighbor support unclustered to improve specificity.
• Functional coherence: Produces functionally coherent gene clusters with high precision when applied to microarray gene expression data.
• Comparative performance: Demonstrated superior performance relative to eight other clustering methods in generating functionally coherent clusters.
• Large-scale applicability: Applicable to large microarray datasets involving thousands of genes across many conditions.

Scientific Applications:

• Microarray gene expression analysis: Clustering genes across thousands of genes and multiple biological conditions measured by microarrays.
• Functional module and pathway identification: Identifying functionally coherent gene modules and individual biological pathways.
• Benchmarking clustering methods: Comparing clustering approaches (e.g., hierarchical, K-means) based on functional coherence of resulting clusters.
• Broad process discovery: Detecting clusters that span a wide array of biological processes rather than being skewed toward dominant processes such as ribosomal genes.

Methodology:

Construct an interaction network from mutual nearest neighborhoods, identify overlapping cliques within that network to form clusters, use nearest-neighbor-based connectivity instead of absolute distance measures, and leave genes without sufficient mutual nearest-neighbor support unclustered.