Performance assessment of kernel density clustering for gene expression profile data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Kernel density smoothing is established for supervised learning of gene expression profile (GEP) data.
• Its utility in unsupervised learning, specifically clustering GEP data, remains unexplored.

Purpose of the Study:

• Introduce and evaluate a novel kernel density clustering method for GEP data analysis.
• Compare its performance against established clustering techniques: hierarchical clustering, K-means, and multivariate mixture models.

Main Methods:

• Applied kernel density clustering to both simulated and real GEP datasets.
• Utilized metrics such as Adjusted Rand Index, Pseudo F test, and r(2) test for performance assessment.
• Evaluated cluster recovery, isolation, coherence, and robustness against noise.

Main Results:

• Kernel density clustering demonstrated excellent performance in recovering clusters from simulated GEP data.
• It effectively grouped large real GEP datasets into compact, well-isolated clusters.
• The method proved to be the most robust against noise compared to hierarchical clustering, K-means, and mixture models.

Conclusions:

• Kernel density clustering is a highly effective unsupervised learning method for GEP data.
• It offers superior robustness and performance, particularly for noisy datasets.
• This technique advances the analysis of gene expression data through improved clustering capabilities.