Performance evaluation on WGBS data.

Performances comparison on real-life libraries among GPU-BSM, Bismark, BSMAP, BS-Seeker2, and segemehl. Two directional libraries are analyzed: SRR019597, which consists of 5.943.586 reads of length 76 bp, and SRR019048, which consists of 15.331.851 reads of length 87 bp. The first and second column of the table report the library and the name of the tools, respectively. The third column reports the time required to analyze the libraries. Columns 4 to 9 report the percentage of uniquely mapped reads according to the number of mapping differences. Differences are mismatches when the tools are used to look for ungapped alignments, whereas they may be mismatches and/or indels when the tools are used to look for gapped alignments. Computing time for GPU-BSM has been reported running it on a single and on two GPUs. As for multi-threading based tools, computing time has been reported for 12 cores. Tools settings: i) GPU-BSM -m 5 –ungapped -l 1, GPU-BSM -m 5 –e2e -l 1, GPU-BSM -m 5 -l 1; moreover for all experiments with GPU-BSM the following settings have been used: -L 76 for SRR019597 and -L 87 for SRR019048, -g 0 to run the experiment on a single GPU (-g 0 -g 1 to run the experiment on two GPUs); ii) Bismark -q ––directional, Bismark -q –directional –bowtie2 -p 6; iii) BSMAP -v 5 -w 2 -r 0 -p 12; iv) BS-Seeker2 -m 5 –aligner = bowtie -f sam, BS-Seeker2 -m 5 –aligner = bowtie2 -f sam –bt2–end-to-end –bt2-p 6, BS-Seeker2 -m 5 –aligner = bowtie2 -f sam –bt2-p 6v) segemehl -F 1 -H 1 -D 0 -A 70 –threads 12.GPU-BSM run on a single GPU.GPU-BSM run on two GPUs.Bismark and BS-Seeker2 run in parallel two instances of Bowtie2. To ensure that both tools use 12 core we used the option -p 6/–bt2-p 6 so that each Bowtie2 instance runs with 6 threads.