Cnv Calls On Infinium Omni Express Chips

Question

[Yunling]

We have run about 60 samples on Infinium Omni Express chips and 24 samples on Infinium Omni Express plus chips. We performed cnv analyses with Penncnv. The Infinium Omni Express plus chips gave us beautiful CNV calls which look real when we looked at the plots. The Infinium Omni Express chips set generated poor CNV calls half of which are false positive when we looked the plots.

When we looked at QC parameters, LRR-SD in Infinium Omni Express chips is higher (<= 0.14) than Infinium Omni Express plus chips (> 0.16). Is it possible to lower the LRR-SD and how?

Any suggestion or experience to share? Thanks!

Yunling

Answer 1

[Matt Shirley]

To answer your main questions: no, there is not a way to lower the LogR ratio standard deviation. It is a fundamental property of your data. I think you might have reversed your LogR ratio standard deviations in your explanation, but I would be wary of calling half of your Omni Express CNV calls "false positives" without some more qualifications. You should enable GC model adjustment, which will normalize the LogR ratio in regions of local abnormal GC content, but I do not think this will affect your overall LogR ratio SD.

What number of markers are you evaluating as a CNV? Generally, if you are looking at CNV calls involving a small number of markers, then you will observe many more spurious CNV calls than if you evaluate CNV calls involving a larger number of markers. Also, the reproducibility across your arrays will be limited. See this comprehensive analysis of CNV calling platforms and algorithms. In summary, don't set your expectations too high. At least you aren't using Affymetrix arrays where it is common to see LogR SD > 0.5 (in my experience).