A dynamically growing self-organizing tree (DGSOT) for hierarchical clustering gene expression profiles

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Microarray technologies generate vast gene expression datasets requiring efficient analysis.
• Traditional hierarchical clustering methods have limitations such as fixed structures and inability to reevaluate mis-clustered data.

Purpose of the Study:

• Introduce a novel hierarchical clustering algorithm to address drawbacks of existing methods.
• Improve the analysis of gene expression data for better biological pattern discovery.

Main Methods:

• Propose the dynamically growing self-organizing tree (DGSOT) algorithm, a tree-structure self-organizing neural network.
• Employ a new cluster validation criterion using Minimum Spanning Tree (MST) for efficient cluster number optimization.
• Utilize a K-level up distribution (KLD) mechanism for reevaluation of data and improved clustering accuracy.

Main Results:

• The DGSOT algorithm constructs hierarchies effectively from top to bottom.
• It optimizes cluster numbers at each level, revealing the dataset's hierarchical structure.
• Applied to yeast cell cycle data, DGSOT extracted meaningful gene expression patterns with high biological functionality enrichment and reasonable hierarchical structure.

Conclusions:

• The DGSOT algorithm offers an improved approach to hierarchical clustering for gene expression data.
• It provides accurate and interpretable results, enhancing biological discovery from complex datasets.
• The algorithm's ability to reevaluate data leads to more robust clustering outcomes.