normalization when calling CNV areas on a genome-wide scale in tumor-germline samples

Question

[Bogdan]

Dear all,

please would you advise : what is a more reliable method for normalization when calling CNV areas on a genome-wide scale on tumor-germline paired samples ?

And : shall we apply the normalization on a chromosome basis, or on a whole-genome ? Thanks a lot !

-- bogdan

[Chris Miller]

I'd encourage you to use an existing package that rolls these normalizations in, rather than trying to reinvent the wheel. As a self-link, copyCat (https://github.com/chrisamiller/copyCat/) does mapability, GC correction, as well as normalizing global read depth. There are other tools that do similar things.

[Bogdan]

Dear Chris, thank you for your suggestions.

I would be very glad to use copyCat . A question though : I have seen distinct packages doing 1) global (genome-wide) normalization or 2) local (chromosome-wide) normalization ?

Would you recommend to use 1) or 2) or perhaps I shall use both normalization methods and compare the results ? Sometimes, on some tumor samples that we have, depending on how we do the normalization, we can call a region CN1 or CN3. Thanks !

[Chris Miller]

If given samtools input, copy cat will exclude chromosomes from depth normalization that appear to differ substantially from CN2, based on allele frequency information. This information is more coarse than what you'd get from programs like ABSOLUTE or TITAN_CNA, that use allele frequency but is a quick way to do it that works reasonably well on most samples.

[Bogdan]

Thank you Chris for your gracious help ..