Critical Assessment of RNA-Seq Differential Expression

Warden and Wu Preprint: v1

In general, this primarily focuses on the following types of comparisons:

Cell line experiments with over-expression or knock-down to define a known causal gene, with processing starting with public reads.

Processed TCGA (The Cancer Genome Atlas) data for breast cancer (BRCA) to compare gene expression by immunohistochemistry status (ER/ESR1, PR/PGR, or HER2/ERBB2).

Differential expression methods include the following:

edgeR (GLM)

edgeR-robust (GLM)

edgeR (QL)

edgeR-robust (QL)

DESeq1

DESeq2

limma-voom

limma-trend (CPM)

limma-trend (FPKM/RPKM)

ANOVA (log2 FRPKM/RPKM)

The most common preprocessing strategies include STAR, TopHat2, and Salmon.  However, a limited amount of additional processing with HISAT2, kallisto, Bowtie2 (+eXpress), and Bowtie1 (+RSEM) is also provided.

Most STAR and TopHat2 alignments use htseq-count for quantification, as well as running cuffdiff (for single variable 2-group comparisons).  However, a limited amount of additional processing with featureCounts is also provided.

Most STAR and TopHat2 alignments start with the public forward reads, even if paired-end data was available.