CNV segmentation issue

[jiadong324]

Hi Tobias,

I am using delly to call CNV from long-read data with the command below:

delly cnv -u -z 100000 -o $out/tumor.bcf -l $vcf -c $out/tumor.cov.gz -g $ref -m $map $tumor

Here are the outputs:

chr16   1980    12153492        CNV00000127     2       2.00551
chr16   12153494        12782279        CNV00000128     2       2.22715
chr16   12782282        54149476        CNV00000129     2       2.02534
chr16   54149479        78570770        CNV00000130     2       1.58412
chr16   78570773        79390610        CNV00000131     2       1.71714
chr16   79390613        79753287        CNV00000132     1       1.18118
chr16   79753290        96183326        CNV00000133     2       1.61651

The depth ratio plot generated by CNVkit is:

As you can tell, most of the q-arm of chr16 is deleted. Therefore, the CN value for CNV00000130, CNV00000131 and CNV00000133 should be 1.

Accordingly, I am wondering how delly converts RDCN to CN? And are there any tuning parameters?

Thanks!

[jingydz]

Are you verifying Delly's output results one by one?
Can CNVkit do this? What does each point represent? If the copy ratio (log2X) is -0.4, shouldn't X be less than 1?
When it's diploid, should the copy ratio be 0?

[tobiasrausch]

Hi Darren,

Looks like delly's raw CN estimates are in agreement with CNVkit. The assignment of integer copy-number status indeed assumes 100% purity, which is not ideal for cancer genomics samples, for instance, with normal cell contamination. I think a useful command-line parameter would be sample purity then these CNs can be adjusted for contamination. I put this on the todo list.

Best, Tobias

[jiadong324]

@tobiasrausch

That would be great! For my long read samples, I found delly is accurate for CN segmentation comparing to Hatchet.

Looking forward to your updates!