Sensitivity, specificity, and reproducibility of RNA-Seq differential expression calls

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Objective benchmarks are crucial for evaluating microarray and RNA-sequencing analysis tools.
• The MAQC/SEQC consortium developed a benchmark for expression profiling analysis.
• Extended benchmarks with common experimental effect strengths are presented.

Purpose of the Study:

• To present extended benchmarks for RNA-sequencing (RNA-seq) analysis tools.
• To evaluate the reproducibility and performance of different analysis pipelines.
• To demonstrate the benefits of using reference standards for data analysis.

Main Methods:

• Utilized a benchmark dataset for microarray and RNA-seq expression profiling.
• Applied factor analysis and additional filters to remove artefacts.
• Evaluated differential expression calls across various tool combinations.
• Analyzed results in the context of experiments with reference standard samples.

Main Results:

• Reproducibility of differential expression calls exceeded 80% for genome-scale surveys after artefact removal.
• Reproducibility for top candidates with strong expression changes ranged from 60% to 93%.
• Factor analysis identified and removed hidden confounders, improving empirical False Discovery Rate (eFDR).

Conclusions:

• Analyzing RNA-seq data with reference standards improves empirical False Discovery Rate (eFDR).
• Computational identification and removal of hidden confounders enhance data analysis robustness.
• Appropriate filtering significantly improves agreement of differentially expressed genes across studies and pipelines.