Cancer Copy Number Analysis - Multiple Snp Array Platforms

Question

[ed.schwalbe]

Dear all,

I have SNP array data run on three platforms: Affymetrix SNP6 (n=~250) Illumina OmniExpress (fresh frozen DNA) (n=12) Illumina OmniExpress FFPE (n=48)

Since the cancer I work on is comparatively rare, many of our diagnostic samples are available only as paraffin blocks, so the arrival of the OmniExpress arrays which were reported to work with FFPE material was most welcome (just FYI, 40/48 FFPE arrays have passed Illumina's own QC procedures available within GenomeStudio).

What I would like to do is to test for recurrent copy number abnormalities across the three platforms, in a way that minimises platform-dependent bias as much as possible. To aid me in this, I have 6 samples which have been run across all platforms, enabling direct comparison of segmentation algorithms.

I first tried the GLAD segmentation algorithm, but it didn't seem to work too well with the FFPE data, so I am now trying a second approach, whereby I identify a segmentation algorithm that works well with FFPE data, and then apply the same algorithm to the omni express fresh frozen and SNP6 data.

Does anyone have any ideas about the best way to proceed with this? Generally, as you might expect, the FFPE SNP array data seems to have an inherently higher variance than arrays from fresh frozen DNA, so maybe I could control the variance to match the frozen arrays?

Thanks for any comments you may have.

Answer 1

[Chris Miller]

I don't have a great answer for your specific problem, but you might find some inspiration in this paper where they combined array intensities from multiple platforms to get higher resolution calls.

A single-sample method for normalizing and combining full-resolution copy numbers from multiple platforms, labs and analysis methods http://bioinformatics.oxfordjournals.org/content/25/7/861.short

That study also deals with taking platforms with different dynamic ranges and different noise levels and combining them into measurements that are comparable.

[ed.schwalbe]

That seems like a great starting point. Thanks so much for the suggestion!

Answer 2

[emilie.lalonde]

Hi there, I'm faced with a similar situation and wondering how your experience went with aroma.cn? I noticed there are two other packages with similar solutions, MPSS and MPCBS. Any insight would be greatly appreciated!