Table 2.
Average expression thresholds
| EE\DEC | limma | edgeR | DESeq2 |
|---|---|---|---|
| Subread | 42.6 | 46.6 | 45.4 |
| r-Make | 42.6 | 48.1 | 46.8 |
| TopHat2/Cufflinks2 | 42.6 | 47.8 | 46.1 |
| SHRiMP2/BitSeq | 41.6 | 47.8 | 45.1 |
| kallisto | 41.3 | 46.4 | 42.8 |
Our benchmark compares the specificity and reproducibility of differential expression analysis for different tools. For a meaningful comparison, all tools are run to give the same sensitivity. For each combination of methods for expression estimation (EE) and differential expression calling (DEC), a threshold for removing the most weakly expressed genes was therefore determined to adjust sensitivity as required. The percentile of genes filtered is shown for which 3,000 genes were found at an average site (q<5% and absolute log-fold change larger than one)