Robust complementary hierarchical clustering for gene expression data analysis by β-divergence.

Area of Science:

• Bioinformatics
• Statistical genomics
• Computational biology

Background:

• Hierarchical clustering (HC) is a common unsupervised method for gene expression data analysis.
• Standard HC algorithms may be misled by highly expressed, irrelevant genes, masking important biological signals.
• Existing complementary hierarchical clustering (CHC) methods lack robustness to data contamination and outliers.

Purpose of the Study:

• To introduce a robust complementary hierarchical clustering (RCHC) method.
• To enhance the reliability of gene expression clustering by mitigating the impact of outliers.
• To improve the identification of biologically significant genes in complex datasets.

Main Methods:

• Developed the robust CHC (RCHC) method by maximizing a β-likelihood function.
• Sequential extraction of gene sets and grouping of individuals.
• Tuning parameter β controls the balance between robustness and efficiency.

Main Results:

• RCHC demonstrates robust performance against data contaminations in gene expression clustering.
• The proposed method overcomes limitations of the original CHC algorithm.
• Successfully predicted critical genes from breast cancer data, showcasing practical utility.

Conclusions:

• RCHC offers a more reliable approach to gene expression data analysis compared to existing methods.
• The method effectively handles noisy data and identifies biologically relevant gene clusters.
• RCHC has significant implications for cancer research and personalized medicine.