8. De novo assembly with Velvet

Velvet is an assembly algorithm developed for genomic assembly, but it can be applied to transcriptome assembly as well. A more recent algorithm from the same authors, called Oases, was developed specifically for un-guided transcriptome reconstruction, but for our purposes the original velvet is enough. Other de novo transcriptome reconstruction algorithms exist, for example Trinity, which is generally more accurate but has very large computational requirements, especially for memory. Trinity offers also modules for gene expression and differential expression estimation, while velvet is limited to transcript reconstruction. We will make a similar usage guide for Trinity in another session.

Velvet is composed by two modules, velveth and velvetg, that need to be run one after the other. The first module analyzes the reads, decomposes them in sub-sequences of fixed length k (called k-mers) and builds an index of all k-mers. The second module takes as input the output of velveth and builds structures in which reads overlap, called contigs, corresponding to transcripts when the input contains RNA-seq reads. Let’s open a script for running velvet, and write the header, load the modules and point to the working directory:

#$ -S /bin/sh
#$ -cwd
#$ -q amd.q,large.q,intel.q
#$ -l h_vmem=16G
#$ -e velveth.e
#$ -N velveth
#$ -o velveth.o
module load velvet/1.2.10
cd /ibers/ernie/scratch/vpl/zebra_fish