Minimum entropy clustering and applications to gene expression analysis

Area of Science:

• Bioinformatics
• Computational Biology
• Information Theory

Background:

• Clustering is crucial for analyzing gene expression data.
• Existing clustering algorithms face challenges with unknown cluster numbers and outliers.

Purpose of the Study:

• To introduce a new clustering algorithm based on information theory.
• To evaluate its performance against established methods, particularly in challenging scenarios.

Main Methods:

• Proposing a minimum entropy criterion using Shannon and Havrda-Charvat's alpha-entropy.
• Developing an efficient iterative algorithm to minimize the entropy.
• Utilizing a non-parametric approach for estimating a posteriori probabilities.

Main Results:

• The algorithm significantly outperforms k-means, hierarchical clustering, SOM, and EM based on the adjusted Rand index.
• It performs well even when the number of clusters is unknown.
• The method effectively identifies outliers while preserving data structure.

Conclusions:

• The proposed information-theoretic clustering algorithm offers superior performance and robustness.
• It provides a valuable tool for gene expression data analysis, especially in the presence of noise and unknown parameters.