Gene expression analysis of combined RNA-seq experiments using a receiver operating characteristic calibrated...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA-sequencing (RNA-seq) platforms are frequently updated, necessitating data integration from diverse sources.
• Combining datasets from different experimental platforms complicates statistical modeling and multiple testing due to complex data distributions.
• Existing methods for batch effect modeling lack general, robust data-driven procedures for RNA-seq analysis.

Purpose of the Study:

• To develop a novel, robust procedure for RNA-seq data analysis that accommodates data from multiple experimental platforms.
• To provide a data-driven simulation (DDS) approach for calibrating significance levels in multiple testing scenarios.
• To enhance the reliability of differential gene expression detection when combining heterogeneous RNA-seq datasets.

Main Methods:

• Proposed a new robust procedure combining popular RNA-seq analysis packages with data-driven simulation (DDS).
• Constructed average receiver operating characteristic curves via DDS to determine calibrated significance levels for multiple testing.
• Calibrated significance levels for specific methods and mean effect models, avoiding adjustments to p-values.

Main Results:

• The DDS procedure successfully calibrated significance levels for multiple testing across different RNA-seq analysis methods.
• Demonstrated the procedure's effectiveness with popular methods like edgeR, DEseq2, and limma+voom.
• The method relaxes stringent data distribution assumptions inherent in many RNA-seq analysis tools.

Conclusions:

• The proposed robust procedure offers a data-driven solution for analyzing combined RNA-seq datasets from multiple platforms.
• This approach improves the accuracy and power of differential gene expression analysis, particularly in complex, heterogeneous data.
• The method is applicable to real-world studies, as illustrated with colorectal cancer data from The Cancer Genome Atlas.