Removing unwanted variation from large-scale RNA sequencing data with PRPS.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate biological interpretation of RNA sequencing (RNA-seq) data necessitates the identification and removal of unwanted technical variation.
• Large and complex studies, such as those from The Cancer Genome Atlas (TCGA), present significant challenges due to multiple sources of noise.
• Uncontrolled variation can compromise critical downstream analyses, including cancer subtyping, gene expression-survival associations, and gene co-expression network construction.

Purpose of the Study:

• To examine sources of unwanted variation in TCGA RNA-seq data.
• To propose and validate a novel normalization strategy, removing unwanted variation III (RUV-III) combined with pseudo-replicates of pseudo-samples (PRPS), for mitigating these variations.
• To demonstrate the efficacy of RUV-III with PRPS compared to standard TCGA normalization methods.

Main Methods:

• Utilized RNA-seq data from The Cancer Genome Atlas (TCGA).
• Developed and applied the removing unwanted variation III (RUV-III) method.
• Implemented a pseudo-replicates of pseudo-samples (PRPS) strategy for normalization.
• Compared RUV-III with PRPS against standard TCGA normalization techniques on multiple datasets.

Main Results:

• Demonstrated that library size, tumor purity, and batch effects are significant sources of unwanted variation in TCGA RNA-seq data.
• Showcased that RUV-III with PRPS effectively removes these confounding factors.
• Illustrated improved performance of RUV-III with PRPS over standard TCGA normalization methods in downstream analyses.
• Validated the approach on several TCGA RNA-seq datasets.

Conclusions:

• RUV-III with PRPS provides a robust strategy for normalizing large-scale transcriptomic datasets.
• This method is effective in removing technical variations originating from library size, tumor purity, and batch effects.
• The approach facilitates more reliable biological discoveries from complex, multi-laboratory, or multi-platform transcriptomic data integration.