Normalization of gene counts affects principal components-based exploratory analysis of RNA-sequencing data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression count data normalization is vital for RNA-sequencing analysis.
• Statistical analysis often focuses on univariate differential expression, overlooking multivariate relationships.

Purpose of the Study:

• To investigate the impact of various normalization methods on Principal Component Analysis (PCA) models.
• To assess how normalization affects the interpretation of gene expression data using PCA.

Main Methods:

• Applied twelve widely used normalization methods to simulated and experimental RNA-sequencing data.
• Utilized Principal Component Analysis (PCA) and Covariance Simultaneous Component Analysis.
• Evaluated PCA model complexity, sample clustering, and gene ranking.

Main Results:

• PCA score plots showed similarity across different normalization methods.
• Biological interpretation derived from PCA models varied significantly based on the normalization technique used.
• Gene enrichment pathway analysis highlighted normalization-dependent interpretations.

Conclusions:

• While PCA visualizations may appear consistent, the biological conclusions drawn from RNA-sequencing data analysis are sensitive to the normalization method employed.
• Selecting appropriate normalization is critical for reliable multivariate analysis and biological interpretation.