Inferring CNVs from BAM files with SAMTools depth

Question

[Sammy]

Hello,

I want to detect CNVs from BAM files (from a batch and one BAM file independently), I read a couple of posts about it, I read about recent tools: biorxiv review and Nature review. Read-depth is the most common approach and I wondered why not normalise by read-depth then use SAMTools to plot the coverage. Comparing with a control, you can infer the CNVs from a BAM file. Sound logical however I feel like I'm trying to reinvent the wheel since there are tools built for this purpose. Do you have any thoughts on it?

I know that it is possible to infer the CNVs from multiple samples but is it correct to infer CNV's from one BAM file by comparing the read-depth between a patient and control (after normalization)?

[brunobsouzaa]

Are you trying to detect Germline CNVs? Just use cnvkit or ExomeDepth for this task!

[Sammy]

Yes, germline, I have 2 BAMs from related individuals. Is it possible to calculate the LOH as well only from these two files?

[German.M.Demidov]

you can call LOH even in one individual - try calling Runs-of-Homozygosity instead (plink can do this, also RohHunter https://github.com/imgag/ngs-bits/blob/master/doc/tools/RohHunter/index.md )

[German.M.Demidov]

is it WGS, panel or WES?

you have only 2 samples, right?

[Sammy]

WES, yes, I do have a patient and a healthy individual.

Answer 1

[German.M.Demidov]

Maybe cnv-kit could be the better choice.

I usually do something even more simple for 2 samples - GC-content normalization and manual selection of regions with copy-numbers of approx 3 or 1. But yeap it is a powerful magic.

In general, CNV callers which are based on read depth and do not use a large cohort for normalization are suitable only for the detection of large CNVs (>10 exonic regions involved). Smaller CNVs become unreliable.

[Sammy]

Thank you!!