Evaluation of de novo transcriptome assemblies from RNA-Seq data

Table 2 Wall-clock runtimes (in hours, minutes and seconds) and memory usage (as measured by the maximum resident set size, in gigabytes) for several assembly evaluation tools

	Assembly T		Assembly O		Assembly S
Program	Runtime	Memory	Runtime	Memory	Runtime	Memory
RSEM-EVAL ^a	1 h 4 m 57 s	2.02 GB	4 h 40 m 36 s	8.18 GB	34 m 57 s	1.23 GB
Genovo	6 d 11 h 54 m 3 s	192.23 GB	>1 week	–	4 d 15 h 3 m 3 s	188.79 GB
ALE ^a	12 h 39 m 36 s	0.67 GB	6 d 23 h 23 m 13 s	2.31 GB	7 h 33 m 1 s	0.59 GB
REF-EVAL, contig ^b	3 s	0.19 GB	8 s	0.33 GB	2 s	0.2 GB
REF-EVAL, nucleotide ^b	8 s	0.39 GB	33 s	1.27 GB	6 s	0.33 GB
REF-EVAL, KC score	1 m 18 s	2.09 GB	1 m 30 s	2.37 GB	1 m 13 s	2.03 GB
Bowtie	15 m 42 s	0.11 GB	1 h 1 m 38 s	0.31 GB	11 m 16 s	0.1 GB
BLAT	35 m 14 s	0.0 GB	1 h 51 m 1 s	0.01 GB	28 m 19 s	0.0 GB

^aPlus time to run Bowtie. We calculate Bowtie statistics separately because ALE takes Bowtie alignments as input.
^bPlus time to run BLAT.
Each tool was run on three different assemblies of the real mouse data. Assembly T was produced by Trinity with its default parameters (52,667 contigs, 33 million nucleotides). Assembly O was produced by Oases with its default parameters (160,455 contigs, 115 million nucleotides). Assembly S was produced by SOAPdenovo-Trans with its default parameters (79,460 contigs, 28 million nucleotides). Multithreaded programs (RSEM-EVAL, REF-EVAL and Bowtie) were run with 16 threads. All programs were run on a compute server with an Intel®; Xeon®; CPU E5-2660 v2 2.20 GHz processor and 500 gigabytes RAM. Genovo had not finished running on assembly O after more than 1 week. KC score, k-mer compression score.

ISSN: 1474-760X