Figure 2

Comparison of mapping results by category for seven tools. The figure shows the mapping by event category on simulated RNA-seq against the human genome on datasets with short and long reads (left 42M, 75 nt; right 48M, 200 nt) for seven different mapping tools: Bowtie, Bowtie2, BWA/BWA-SW, CRAC, GASSST, GSNAP, and SOAP2. We consider six categories of reads depending on whether they contain an SNV, an insertion, a deletion, a junction, a sequence error, or a chimeric splice junction (a chimera). In each category, the bar is the percentage of those reads mapped at a unique location by the corresponding tool. The red tip at the top of the bar is the percentage of incorrectly mapped reads. With 75 nt reads, CRAC is better than the other tools, reaching a sensitivity >90% and a precision >95% whatever the category. The other tools except GSNAP are below 50% sensitivity for mapping reads in categories where spliced alignments are needed (for which they are not intended) and for reads containing insertions or deletions. With 200 nt reads, CRAC remains by far the most sensitive and specific tool; the difference between CRAC and GSNAP and Bowtie2 increased in all categories. Compared to short reads, the other tools had a better mapping of insertion and deletion containing reads. SNV: single nucleotide variant