Clustering of RNA-Seq samples: Comparison study on cancer data

Area of Science:

• Genomics
• Bioinformatics
• Cancer Research

Background:

• RNA-sequencing (RNA-Seq) is the standard for gene expression measurement, surpassing microarrays.
• RNA-Seq data analysis standards, especially for clustering, are underdeveloped compared to microarrays.
• Clustering is crucial for understanding gene expression patterns in cancer research.

Purpose of the Study:

• To evaluate computational strategies for clustering cancer samples using RNA-Seq data.
• To provide guidelines for practitioners on optimal RNA-Seq data clustering methods.
• To identify best practices for expression estimation, feature selection, and data transformation.

Main Methods:

• Empirical analysis of 15 mRNA-Seq cancer datasets.
• Evaluation of expression quantification strategies.
• Assessment of non-specific filtering effects on gene number.
• Comparison of data transformation techniques (e.g., log-transformation).
• Performance evaluation of four clustering algorithms (e.g., Average-Linkage, k-medoids).
• Assessment of twelve distance measures, including Symmetric Rank-Magnitude correlation.

Main Results:

• Clustering cancer samples based on gene quantification is recommended.
• Non-specific filtering to approximately 1,000 genes generally improves clustering performance.
• Log-transformation of data prior to clustering is beneficial.
• Average-Linkage and k-medoids algorithms show superior performance.
• Symmetric Rank-Magnitude correlation offers consistent results across different scenarios.

Conclusions:

• Optimal RNA-Seq cancer sample clustering involves gene quantification, log-transformation, and feature filtering.
• Average-Linkage or k-medoids algorithms combined with Symmetric Rank-Magnitude correlation are recommended.
• These findings provide practical guidelines for robust gene expression data clustering in cancer research.