Calling CNVs using a single tumor sample

Question

[vctrm67]

If I only have a tumor sample to call CNVs from, is it possible to exclusively call somatic or germline CNVs? It seems to me that the only way to call CNVs is to compare it to a reference and call that way, but it seems that would just give both somatic and germline CNVs.

[brunobsouzaa]

I don't think this will work but you can try using cnvkit with a flat reference. Usually, you need more samples to build your baseline but doesn't hurt to try!!!

[Ram]

If you only have tumor BAMs, try using the CopyWriteR package by supplying the same file as both sample and control. The output will have sample_name.vs.sample_name and sample_name.vs.none entities, and you can use the vs.none entities. IMO this is the closest you can get without matched/pooled normals.

[Richard]

If you only have a tumour bam file the CNVs you detect will be a mix of germline and somatic variants as you expect. Ideally you would have a matched normal to subtract from your tumour during CNV analysis. To find CNVs in your tumour bam file that are larger than about 50Kb you might find that control-freec does a good job at this, though other tools may be better. For very small CNVs tools like manta or delly might be your best bet.

To separate the somatic variants from the germline variants you could try comparing to the common germline variants listed in DGV. Similarly there may be some variants (like focal gains of MYC) that you know are very likely to be of tumour origin.