Problems using cnvkit for ctDNA targeted sequencing

Question

[ingrid.schulman]

Hi, I have tried to use cnvkit to detect copynumbers in circulating tumor DNA samples. I used the "batch -m amplicon" command with both cancer and normal samples followed by "call" to get absolute copy numbers. The results doesn't correlate much with the "actual" copyumbers in my data (I also have whole genome samples of the same patients to compare with). I get a lot of false CN calls of 0 and some really high, like 35 (that should be 2)

So my question is that is there any additional steps to batch and call that I could use to get better results? And is there anyone that has experience using cnvkit for ctDNA samples?

Answer 1

[Eric T.]

For ctDNA samples, if it's at all possible for your lab to use a hybrid capture protocol instead of targeted amplicon sequencing, that would give you a much better chance of detecting somatic CNVs accurately. The TAS protocol just doesn't retain much information for a copy number caller to use, so it's expected that the WGS results would be much better even on the captured genes.

One method to reduce false positives is to run segmetrics --ci and then call --filter ci to calculate confidence intervals for segment means and automatically drop the segments where the CI includes neutral copy number.

If you have access to SNV calls, then detecting loss of heterozygosity and allelic imbalance via SNP b-allele frequencies should help distinguish true calls from the false positives. Accurate estimates of tumor purity are important here to calculate the absolute integer copy number values (cn, cn1, cn2 from the call command).