Svdetect Taking Too Long To Run Or Crashing For Finding Links

Question

[Jordan]

Hi,

I'm trying to find SV's using SVDetect on a whole genome bam file. I was doing the linking step on the entire bam file (60GB), which was generated after BAM preprocessing. But after sometime the program got crashed. So, I divided the whole genome bam file and into individual chr bam files and ran SVDetect again using the following shell script:

for i in {1..22}
do
      echo "SVDetect linking -conf Conf/chr$i.conf 2>Logs/chr$i.log"
      SVDetect linking -conf Conf/chr$i.conf 2>Logs/chr$i.log
done

It ran well for chr 1&2. But when I came to chr3, it threw me an error saying that,

can't open chr3.bam: Cannot allocate memory.

I have given a memory allocation of 120GB. Each chr bam file is approximately 3GB in size. Does it still need more memory than this? I wanted to know if anyone else had memory issues like me and how others perform their SVDetect analysis or Structural variant analysis.

I did look at BreakDance too, but we decided to use SVDetect because it gives Circos plots in the end.

Thanks!

Answer 1

[Charles Warden]

I successfully ran SVDetect on 8-9 GB .bam files for mate-pair data without using the pre-processing script (although I know that was supposedly required). Took about a day to run the initial linking step (per sample).

Have you tried:

1) Running the analysis without the pre-processing script (with a sufficiently large number of threads, this should be splitting up the job into manageable pieces automatically)

2) Splitting up the reads by chromosome prior to the filter and/or only running the analysis on a small chromosome (say, chr18). Obviously, the chromosome splitting strategy isn't ideal, but maybe this can at least help establish the cause of the problem (if a very small .bam file still gives the error message, maybe it is a configuration problem rather than true memory problem)

[Samarth Kulshrestha]

@Charles Warden: Thanks for the explanation.

I am also facing some memory related issue on my BAM files (Mate-pair, ~3GB, after pre-processing step) during my analysis. I tried to run SVDetect on a 64GB as well as 128GB Machine with multiple threads but program always crashes during "links2compare" process.

Any recommendation will be helpful for my SV analysis.

Thanks Samarth Kulshrestha

Answer 2

[Charles Warden]

Also, SVDetect creates a circos format input file, but it didn't automatically create the image for me. I needed something like RCircos to do that:

http://cran.r-project.org/web/packages/RCircos/index.html

http://circos.ca/software/download/