Oases assembly merge times out on a cluster

[RamakrishnanRS]

Hi,

I have run velvet-oases on k-mers 21 thru 55, and I am trying to merge the assemblies now. I tried using the velveth --> velvetg --> oases approach from the "Assembly Merging" section of the manual, but velvetg runs for an abnormally long duration (>48 hours), and the cluster on which I'm running the script times out.

I have tried providing 244 GB of RAM, but it makes no difference - the "Log" file from velveth is written to once when the script starts running, and no other file(including the redirected stderr/stdout) is touched for the entire duration of the run.

Any idea how I could tackle this?

--
Thanks,
Ram

[jgibbons1]

Small K-mer assemblies are pretty computationally heavy. How many reads are you working with? Have you done any quality filtering on your reads?

If you are using a massive amount of reads, you may want to consider implementing some sort of digital normalization (Trinity can do this independent of assembly and Titus Brown has a version of digital normalization as well http://arxiv.org/abs/1203.4802).

Reducing reads and reads with errors will greatly reduce memory and run time.

[RamakrishnanRS]

Hi,

Thank you so much for the response!

I'm working with Titus Brown's version of digital normalization right now - that brought the reads down from around 289 mil to 12 mil (6X2) PE reads and 7 mil SE reads (after a strip-and-split ).

I asked my sequencing provider - the only QC step was the pass filter run on raw reads. Is there a specific QC filter I can use for transcriptome reads?

Originally Posted by jgibbons1

Small K-mer assemblies are pretty computationally heavy. How many reads are you working with? Have you done any quality filtering on your reads?

If you are using a massive amount of reads, you may want to consider implementing some sort of digital normalization (Trinity can do this independent of assembly and Titus Brown has a version of digital normalization as well http://arxiv.org/abs/1203.4802).

Reducing reads and reads with errors will greatly reduce memory and run time.

[jgibbons1]

No problem.

One thing you may want to try BEFORE digital normalization is trimming reads based on quality score. There are quite a few pieces of software that can do this (trimmomatic and trim galore to name a couple).