Correcting scale distortion in RNA sequencing data

Area of Science:

• Genomics and Bioinformatics
• Molecular Biology
• Computational Biology

Background:

• RNA sequencing (RNA-seq) is a standard method for measuring gene expression across entire genomes.
• Accurate gene expression data is crucial for identifying genetic factors in diseases through population studies.
• Existing normalization methods may not fully correct for all sources of error in RNA-seq data.

Purpose of the Study:

• To identify and correct expression level-dependent biases in RNA sequencing data that persist after standard normalization.
• To improve the accuracy of gene-gene correlation estimations and statistical tests in population studies.
• To enhance the reliability of RNA-seq data for clinical and research applications.

Main Methods:

• Analysis of multiple RNA-seq datasets from TCGA, SU2C, and GTEx databases.
• Application of local averaging to detect sample-specific, expression-dependent biases.
• Development and application of two novel nonlinear transforms to correct identified biases.
• Utilized a new simulation methodology to assess the impact of corrections on statistical tests.

Main Results:

• Expression level-dependent biases were detected in all analyzed RNA-seq datasets, varying between samples.
• These biases were shown to negatively impact gene-gene correlation and differential expression analyses.
• The proposed nonlinear transforms effectively removed per-sample biases, reduced variance, and improved correlation distributions.
• Data correction led to a 3-5% improvement in sensitivity and specificity for two-population tests.

Conclusions:

• Novel nonlinear transforms can accurately correct for expression level-dependent biases in RNA-seq data.
• Bias correction enhances the reliability of gene-gene relationship analysis and statistical power in population studies.
• These findings offer improved methods for utilizing clinical RNA-seq data, potentially leading to new discoveries.