CNV detection via BAM depth-of-coverage

[quinlana]

Hello all,
I am looking for a reliable, published tool for CNV discovery by differences in depth of coverage (DOC). Preferably something that accepts BAM as input. Our SV tool / pipeline (Hydra) works well for discovering breakpoints via paired-end mapping, but we (and others) find that certain classes of rearrangements are not easily detected by PEM yet are by DOC. In a previous study (Quinlan et al, 2010), we used an HMM on GC-corrected sliding windows, but would prefer to use a published method for a current study.

Could someone help me out with suggestions?

Best,
Aaron

[henry.wood]

Hello
Have you tried CNV-seq http://tiger.dbs.nus.edu.sg/CNV-seq/
I had a play around with it a while back. It seems easy enough to use, but you might have to convert the BAM data beforehand. It doesn't really work well with the complex changes you see in tumours, so I had to develop my own method (far too messy to share at the moment), but is good for discreet changes in constitutional DNA.
I also tried segseq http://www.broadinstitute.org/cgi-bi...w&paper_id=182 but I couldn't get it to work. Probably more a reflection on me than the program.

[krobison]

Which of the listed programs in the Wiki might fit your need? Or which have you tried & found lacking?

[zlu]

Hello
Have you tried CNV-seq http://tiger.dbs.nus.edu.sg/CNV-seq/
I had a play around with it a while back. It seems easy enough to use, but you might have to convert the BAM data beforehand.

I'm trying to use CNV-seq on the bam files I created with BWA and this may be really obvious but how do you decide which is ref and which is test. If I have 2 bam files generated by mapping Illumina reads from 2 individuals against the NCBI reference genome, does it mean either one can be ref or test? I have also been playing around with the p-value and the bigger-window parameters but I get either no CNV or more than 10k CNVs in a single chromosomes. Any suggestion on the tweaking the parameters? Thank you.

[henry.wood]

Choosing which is normal depends what you are doing. If you are looking at a diseased and healthy person then it is easy. If you are looking at variation between two ordinary people then you might want to get hold of a third person, so any changes you seen can be validated. I'm led to believe that there is an artificial sequence read generator somewhere in samtools. Or you could use one of the published genomes. Then you will know whether the gain you see in person A is actually a loss in person B.
I didn't play around too much with the parameters. Nothing to do with the ease of using the program, which was very good, it's just that cnv-seq wasn't so useful for the tumours I was dealing with. It ignored focal amplifications within a chromosomal gain, calling it all one gain. It also didn't like homozygous deletions.