How to run velvetoptimiser on paired-end FASTQ datasets

Question

[whatup]

Running Velvet has always been confusing for me. So, I have always been running it using VelvetOptimiser in order to avoid dealing with searching for optimal parameter values.

If I run VelvetOptimiser to assembled a set of paired_end FASTQ sequences, then does this mean I do not need to run Velveth and Velvetg at all? (I am asking this because I am not so satisfied with the result. Most of the contigs from running only VelvetOptimiser seem too short)

Here is the basic steps that I take to VelvetOptimiser.

1. I convert two FASTQ files into one using shuffleSequences_fastq.pl since Velvet requires that paired-end FASTA and FASTQ datasets come in a single merged file.

2. then I run velvetoptimiser as following:

   VelvetOptimiser.pl \
   -s 27 \
   -e 31 \
   -t 8 \
   -d '/outputfolder' \
   -f '-fastq -shortPaired theMergedFile.fastq'
   

#2 run without any error. However, the resulted contig length are too short. I know they are too short because I ran it also using Trinity assembly. Maybe it's the default parameter values -s 27 -e 31 that is causing this problem. Can you spot what I might be doing wrong?

Answer 1

[Philipp Bayer]

It looks like you assemble RNASeq, is that right? You might want to use Oases with your Velvet results, have a look at the paper comparing Oases to Trinity: http://bioinformatics.oxfordjournals.org/content/28/8/1086.long

It corrects and scaffolds your contigs so you should see longer ones. As far as I know, Velvet was built mostly for genomic contigs, so it assumes a few things that Oases corrects, Trinity was built for RNASeq so it's no wonder you see longer contigs.

For your other questions:

Yes, VelvetOptimiser runs velvetg and velveth internally, so you can use the final results. Which optimal k-mer value does it report? If it reports 31, your range might not be inclusive enough, try something larger for the end value.

[whatup]

Thank you for your advice! It's already clearing things up for me. I'll have a look at Oases as you suggested.

[whatup]

yes, it is RNASeq