Method to detect CNV in a large set of coverage profile ?

Question

[Pierre Lindenbaum]

Hi all,

I've got the profile of coverage for >10000 BAM files for the same 16Kb region , all normalized on the median depth of the region. The data are available as a binary file containing an array of integers but I can change this.

Of course, the profile of coverage is more or less close to '1.0' , but if there is a deletion a local part of the coverage will be close to '0.5' , etc...

Do you known any method (machine learning ?) to detect a rare CNV in those files ? something unusual (>50bp) in the profile ?

Thanks!

P

Answer 1

[colindaven]

I don't know if your data are appropriate (short reads?), but this one might fit the bill. https://github.com/PopicLab/cue

I must admit I haven't tried it yet.

Answer 2

[jared.andrews07]

I'd probably just try circular binary segmentation first and see what that looks like. You may have to play with bin size a bit to get the resolution you want, but I don't see why it wouldn't work for your purposes.