Cluster analysis on high dimensional RNA-seq data with applications to cancer research - An evaluation study

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression data clustering is a key method for discovering novel cancer subtypes.
• Existing clustering approaches lack comparative analysis regarding their performance and data characteristic influence.
• Understanding these factors is crucial for advancing cancer subtyping research.

Purpose of the Study:

• To evaluate how clustering choices impact performance in human cancer gene expression data.
• To investigate the influence of sample size, subtype distribution, and sample heterogeneity on clustering accuracy.
• To provide insights into optimizing clustering strategies for cancer subtyping.

Main Methods:

• Analysis of four publicly available human cancer datasets (breast, brain, kidney, stomach).
• Systematic evaluation of clustering performance across different sample sizes.
• Assessment of the impact of subtype distribution and sample heterogeneity (including sex) on clustering outcomes.

Main Results:

• Increasing sample size generally had a limited effect on clustering performance.
• Disproportionate subtype distribution significantly hindered accurate clustering.
• Clustering performance varied based on the chosen method and data set, with sex-specific analysis often yielding superior results due to reduced heterogeneity.

Conclusions:

• Sample size has a minimal impact on clustering performance; data heterogeneity, especially sex, is a critical factor.
• Analyzing genders separately can mitigate performance loss from reduced sample size by creating more homogeneous datasets.
• Gender-specific analysis may enhance the detection of novel cancer subtypes.