Spectral clustering strategies for heterogeneous disease expression data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression data clustering is crucial for biomarker discovery and personalized medicine.
• Traditional methods like K-means and hierarchical clustering have limitations with complex, heterogeneous human disease data.
• Spectral clustering offers a promising alternative by utilizing pairwise comparisons and eigen techniques.

Purpose of the Study:

• To develop a novel recursive K-means spectral clustering (ReKS) method tailored for human disease gene expression data.
• To evaluate the performance of ReKS against existing clustering techniques.

Main Methods:

• Developed a recursive K-means spectral clustering (ReKS) algorithm.
• Benchmarked ReKS on three large-scale cancer gene expression datasets.
• Compared ReKS with hierarchical and K-means clustering regarding execution time, background models, and biological knowledge integration.

Main Results:

• ReKS demonstrated superior performance compared to hierarchical clustering methods.
• ReKS showed comparable results to K-means but with significantly faster execution times.
• ReKS effectively clustered heterogeneous human disease gene expression data without requiring a priori knowledge of the number of clusters (K).

Conclusions:

• The recursive K-means spectral clustering (ReKS) method provides an efficient and effective approach for analyzing human disease gene expression data.
• ReKS offers a valuable alternative to existing methods, particularly for complex datasets where parameter selection is challenging.