How to merge CNV of multiple samples from CNVkit

Question

[cc]

Hi, everyone,

I just got CNVkit results from multiple samples seperately, but it seems that there's no function to merge CNV results from multiple samples in CNVkit. Or, any command I haven't noticed?

Thanks for your interest.

[cc]

Another question about controls.

I have only 2 controls sequenced in the same strategy with 20 cases. Considering the limited number of controls, I tried to run CNVkit with two strategy, which results in obvious different outputs.

1. reference constructed with pooled normal controls (2 samples) : the .cns of sampleA reports 93 segments.
2. reference constructed with no control samples ：the .cns of sampleA reports 9 segments, and the last segment is about 5.3M. (The whole target region is 8M in my data)

In this case, what can I do to get a more accurate results？ Or, can I use the reference constructed with no control samples, then call CNV for all cases and 2 controls, and compare the results of cases and controls at last ?

My data are from targeted sequencing of a 8M region (hybrid capture), if I use WGS data of normal controls from other paper to construct the reference of pooled samples. Is it acceptable ?

Thanks in advance

Answer 1

[Eric T.]

What type of result to you want from merging samples?

• If you're interested in spotting recurrent copy number alterations, try the heatmap command with all of your .cns files.
• If you want to quantify the significance of recurrent alterations, use the export seg command to create a SEG file from each .cns file of interest, then use the output SEG files with a recent version of GISTIC2. (The "markers" file is no longer needed in recent versions of GISTIC, I'm told. Just the SEG files should be enough. I haven't tried this myself.)

If you build a reference from control samples, the control samples should be prepared and sequenced with the same protocol as the test samples -- NOT WGS if the test samples were sequenced with a target panel. If you do have process-matched controls, then a pooled reference built from those controls is usually better than a control-free reference. Otherwise, just use a control-free reference.

Don't worry too much about the number of segments; instead, use the segmetrics and call commands to do further filtering if you need it. You can also repeat segmentation with a more stringent p-value threshold (-t) to reduce the number of segment breakpoints in the .cns files.

[cc]

Thanks! Your suggestions are very helpful. Considering the obvious difference between the 2 controls, I tried to use a control-free reference, then call CNV for both cases and the only 2 controls, and compare cases and controls at last. It seems to be a good choice for me.