Fast iterative gene clustering based on information theoretic criteria for selecting the cluster structure

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression clustering is crucial for understanding gene function from microarray data.
• Accurate selection of the number of clusters is a key challenge in gene expression analysis.

Purpose of the Study:

• To introduce a novel method for selecting the number of clusters in gene expression data using the minimum description length (MDL) principle.
• To develop a fast and efficient estimation method for determining the optimal number of clusters without exhaustive evaluation.

Main Methods:

• The study proposes a Minimum Description Length (MDL) based approach for selecting the number of clusters.
• This MDL estimation method is integrated with clustering algorithms, including a modified "gene shaving" (GS) procedure.
• The modified GS algorithm replaces Gap statistics with the MDL estimation method.

Main Results:

• The proposed MDL-based method provides a fast evaluation of the number of clusters.
• Simulations demonstrate that the modified GS algorithm (GS-MDL) outperforms the original GS-Gap and Classification Expectation Maximization (CEM) algorithms.
• Application to B-cell differentiation data shows comparable or improved clustering results compared to Self-Organizing Maps (SOM).

Conclusions:

• The MDL principle offers an efficient and sound basis for determining the optimal number of clusters in gene expression data.
• The integration of MDL with clustering algorithms like gene shaving enhances analytical performance.
• This method provides a valuable tool for biological information discovery from gene expression datasets.