CNV calling from exome sequencing data with a low number of samples

Question

[FKM]

Hi all,

I have trio data from exome sequencing on which I would like to do CNV calling. I have used different software since now, but they all require a large number of samples. I was wondering then if there is any software specific for CNV calling on a very limited number of samples, I remember I read something about some time ago, but I cannot find anything. Any idea either on software or other way of doing it?

[Ram]

When we faced this problem, we would use "background" samples. I'm not entirely certain about the criteria for these samples, but they were principally used to make up for the low actual sample count. I'll get in touch with my colleagues and get back to you.

[FKM]

Thanks for your answers. The main problem is that these samples were sequenced with a different enrichment kit respect the one we usually use, hence we do not have a reasonable number of controls/samples. I was already skeptical regarding the possibility of doing this, because without a reference it would be very complicated to do an accurate CNV calling, but I posted anyway in case someone had a smart alternative solution.

[Ram]

Please do not add an answer unless you're answering the top level question. This should be a comment on the appropriate person's post (or a reply to their comment). I'm moving it to a comment now, but please be more careful in the future.

Answer 1

[gab]

I used cnvkit for this purpose. It does the job for one sample only at the time, if you have pairs of control-abnormal samples. You'll need many samples only if you do not have a control sample, and you want to build a reference as solid as possible.

Answer 2

[Eric T.]

If you're looking for de novo CNVs in the proband, you can use CNVkit to build a reference from the two parent samples. The results will still be noisy, so see the "calling" and "germline" pages in the documentation for suggestions to clean up the results.

Answer 3

[Tm]

CNVKIT is a good pipeline to consider for CNV anaysis. You can also give try to VarScan