Robust principal component analysis for accurate outlier sample detection in RNA-Seq data.

Area of Science:

• Genomics
• Bioinformatics
• Statistical analysis

Background:

• High-throughput RNA sequencing (RNA-seq) is crucial for gene expression studies.
• Technical variations or biological differences can cause outlier samples in RNA-seq data.
• Detecting outliers in high-dimensional RNA-seq data with few replicates is challenging.

Purpose of the Study:

• To apply robust statistics, specifically robust principal component analysis (rPCA), for outlier detection in RNA-seq data.
• To evaluate the performance of rPCA compared to classical principal component analysis (cPCA).
• To assess the impact of outlier removal on differential gene expression analysis.

Main Methods:

• Utilized two rPCA methods: PcaHubert and PcaGrid.
• Tested methods on simulated and real RNA-seq datasets with known outliers.
• Compared rPCA and cPCA performance on a mouse cerebellum RNA-seq dataset.
• Analyzed differentially expressed genes before and after outlier removal.

Main Results:

• PcaGrid demonstrated 100% sensitivity and specificity in detecting positive control outliers.
• Both rPCA methods identified two outlier samples in a mouse RNA-seq dataset where cPCA failed.
• Outlier removal without batch effect modeling yielded the best results for detecting biologically relevant differentially expressed genes.

Conclusions:

• Robust principal component analysis (rPCA) provides an accurate and objective method for outlier detection in RNA-seq data.
• rPCA is suitable for high-dimensional data with small sample sizes.
• Outlier removal enhances differential gene detection and downstream functional analysis.