CNV Segmentation density - non-zero?

Question

[dompearcejankowski]

Hi All,

tl;dr my segment means look odd https://imgur.com/a/5yOTodf

I'm analysing Illumina CytoSNP 850k data, that has been pre-processed as a (I believe) combination of Genome Studio and BlueFuse.

Confusingly to me, when the distribution of log2 segment means (i.e. the "copy number") are visualised, it's bimodal around 0.

My expectation would be that most segments display, on average, no copy change and a single peak should be evident at ~0 (this is ignoring any smaller peaks that are assumed to be associated with actual CNV).

Is this a problem with the segmentation algorithm that has been used? Would it be prudent to perform the segmentation again, myself?

Thanks for any help! Dom

The plot itself can be found here - https://imgur.com/a/5yOTodf

Answer 1

[markus.riester]

The normalization algorithms assume that test and control cells contain the same amount of DNA. In your case, the test ploidy is significantly below 2, thus you get a right shift in log2-ratios. Have a look at tools like ABSOLUTE or ASCAT that assign integer copy numbers to these peaks by jointly calculating purity and ploidy. This question also has been asked a few times here, e.g. C: "Shifted" copy number ratios coming from cnvkit?, so you might find more detailed information in the archive.

[dompearcejankowski]

Thanks a lot, really appreciated