0.4 Available workflows
At the present time are available the following workflows:
- mRNAseq, which allows:
- adapter trimming with skewer
- mapping with STAR
- counting genes and isoforms with RSEM
- ENSEMBL gene annotation.
- organizing the output of RSEM in tables to be used for differential expression analysis
- visualizing experiment data with PCA
- evaluating experiment power and sample size
- detecting differentially expressed genes/isoforms
- miRNAseq, which executes the workflow described in Cordero et al. PLoS One. 2012;7(2):e31630, embedding the following steps:
- ChIPseq, which allows:
The most computing expensive steps of the analyses are embedded in the following docker4seq functions: rnaseqCounts, mirnaCounts, chipseqCounts. These functions are also the only having RAM and computing power requirements not usually available in consumer computers. Hereafter it is shown the time required to run the above three functions increasing the number of sequenced reads.
0.4.1 rnaseqCounts performances
The count generation from fastq files is the most time consuming step in RNAseq data analysis and it normally requires high-end server. We tested rnaseqCounts on different hardware:
+ SeqBox: NUC6I7KYK CPU i7-6770HQ 3.5 GHz (1 core, 8 threads), 32 Gb RAM, HD 250 Gb SSD
+ SGI UV200 server: CPU E5-4650 v2 2.40GHz (8 cores, 160 threads), 1 Tb RAM, RAID 6, 100 Tb SATAWe run respectively 26, 52, 78, and 105 million reads using different number of threads, vaules shown in parenthesis in Figure 1. It is notable that SeqBox, mapping in 5 hours more than 100 milion reads, it is able to handle in 20 hours the throughput of the Illumina benchtop sequencer NextSeq 500, which produces up to 400 milion reads in a run of 30 hours.
0.4.2 mirnaCounts performances
We run resepctively 3, 6, 12, and 24 miRNA samples in parallel using mirnaCounts, with different number of threads, vaules shown in parenthesis in Figure 2.
0.4.3 chipseqCounts performances
We run respectively 37, 70, 111, and 149 million reads using different number of threads, vaules shown in parenthesis in Figure 3.
From the point of view of parallelization the rnaseqCounts is the one that embeds the most computing demanding tools: i) mapping with STAR and ii) quantifying transcripts with RSEM. Both these tools were design to take advantage of multiple cores hardware architecture and they also require massive I/O. On the basis of the results shown in Figure 1 parallelization does not improve very much the overall performances, even if it can mitigate the gap w.r.t. SeqBox due to the poor I/O performance of the SATA disk array. On the other side the presence of a SSD with very high I/O performance can remedy the limited amount of cores of SeqBox.
In the case of mirnaCounts and chipseqCounts the parallelization is very little and it is only available for the reads mapping procedure. Moreover, both functions have a massive I/O. The reduced parallelization of these two analyses combined with the higher I/O throughput of the SSD with respect to the SATA array makes SeqBox extremely effective even with very high number of reads to be processed, Figure 2 and 3.
