Recommended window size for whole genome CNV detection

[Shankar Ajay]

Hello,

I was wondering if anyone has a recommended window size for whole-
genome CNV detection. I'm currently using 5kb non-overlapping windows.
Is it reasonable to try 1kb/2kb windows?

Thanks,
Shankar

--

Postdoctoral fellow
NIH/NHGRI

[Sebastian Waszak]

Dear Shankar,

This depends mainly on your sequencing depth, sequencing noise, and on
the read length (because of mappability). My recommendation for whole-
genome runs and different sequencing coverages is:
1x: 25-50kb non-overlapping windows
2x: 10-20kb non-overlapping windows
4x: 5-10kb non-overlapping windows
10x: 2-5kb non-overlapping windows
20x: 1-2kb non-overlapping windows
40x: 0.5-1kb non-overlapping windows

The only caveat with small windows sizes is mappability, i.e., many
500bp-2kb windows will have a low or very low effective mappable
fraction in repetitive regions and segmental duplications. Windows
with a low mappability can be filtered out in the final output file
(=field "numUnMapKmers"; number of uniquely mappable k-mers). Also a
whole-genome-1kb-tiled run will take many hours. Here I would
recommend to split up the CopSeq run into per-chr jobs.

Cheers,
Sebastian