How does normalization impact RNA-seq disease diagnosis?

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• RNA sequencing (RNA-seq) data is crucial for disease diagnosis.
• Normalization is a standard procedure to ensure sample comparability in RNA-seq.
• The impact of normalization on RNA-seq disease diagnosis remains unevaluated.

Purpose of the Study:

• To investigate the impact of normalization on RNA-seq disease diagnosis.
• To evaluate whether normalization enhances or diminishes diagnostic performance.
• To compare the diagnostic utility of raw versus normalized RNA-seq data.

Main Methods:

• Analysis of RNA-seq data from The Cancer Genome Atlas (TCGA) portal.
• Development of a novel normalization effect test algorithm (diagnostic index, d-index) and data entropy.
• Application of machine learning models and visualization analysis to compare raw and normalized data.

Main Results:

• Normalized RNA-seq data often yields equivalent or lower diagnostic accuracy compared to raw data.
• Certain normalization methods (e.g., RPKM) can negatively impact disease diagnosis.
• Raw data may better capture pathological transcriptome patterns, and high entropy correlates with better diagnosis.

Conclusions:

• Normalization may not be an indispensable procedure for RNA-seq disease diagnosis.
• Raw RNA-seq data can be as effective, or even superior, to normalized data for disease diagnosis.
• High-dimensional imbalance (HDI) data is resistant to normalization effects and poses challenges for machine learning models.