TCGA data: CNV SNP 6.0 vs LowPass DNAseq

Question

[Yuri Tolkach]

A have a question regarding the differences in CNV capture between the SNP 6.0 and LowPass DNAseq. I would appreciate when someone could answer it.

I have made a comparison between the results of these two methods and found out the dramatic differences:

-SNP 6.0 captures commonly large segments of genome with different CNV statuses of areas within one analysed segment, which leads to almost uninterpretable small amplification/detection statuses (based on segment_mean parameter). I.e., you have almost no possibility to carry out the real CNV status for selected genes.

-DNAseq seems to be much more precise, but is available only for small number of patients.

Has anyone encountered this problem?

And maybe a more relevant question: Do you generally use SNP 6.0 generated CNV data given the aforementioned unreliability? When yes, how exactly?

Best wishes, Yuri

[Yuri Tolkach]

Dear Cyriac,

thank you very much for answer.

As I have understood, CNV is for global evaluation and is poor at single gene resolution. DNAseq is method of choice for this aim. To the question of the intratumoral heterogeneity and clonality, TCGA does not account for it. This is simply to accept and to do the analyses which would make sense at any degree of heterogeneity.

Best wishes,

Yuri

Answer 1

[Cyriac Kandoth]

Sounds like your goal is to find "CNV status per-gene". Note that other goals of CNV analysis can be to find arm-level gains/losses, which are often used as clinical biomarkers, and SNP-array data is good at that. But I'll compare 3 TCGA platforms only for your specific goal of calling gene-level copy-number per-sample.

1. SNP array - Tells us whether a gene is significantly amplified/deleted relative to the rest of the genome, but has no sense of the actual number of copies of the gene, or the copies of each allele. So it misses events like isodisomy or whole genome duplication.

2. DNA Low-pass Seq - Tells us the actual breakpoints of structural variants like insertions, deletions, translocations, that lead to various gene-level CNVs. But the low read depth restricts our ability to account for purity and tumor heterogeneity, and detect subclonal events.

3. DNA Exome Seq - Tells us gene-level CN fairly accurately, but misses the actual breakpoints of the structural variants, since they are usually intronic or intergenic. Callers like ASCAT and FACETS will properly account for sample purity, ploidy, heterogeneity, and allelic-imbalance.