Questions about creating my_antitargets.bed

Question

[Hällyss]

Hello,

I'm trying to search CNV on patients with my somatic panel (15 genes) and I'm interesting about CNVkit.

The first steps are to creating target and antitarget bed. And I have few questions about it :

1- In the manual : " CNVkit uses a cautious default off-target bin size that, in our experience, will typically include more reads than the average on-target bin." I don't understand, the mean of RC in the off-target must be higher than the mean of RC in the in-target or the total of RC in the off-target must be higher than the mean of RC in the in-target ?

2- I think it's the first probability, so i create a antitarget bed whilch the mean RC is equal to the mean RC of in-target. The split of my_target.bed is 100 (like your recommandation for deep sequencing) and for my_antitarget.bed is 1 200 000. Is there a problem if the segments of target and antitarget have not the same length (x 12 000 is maybe too big ?)

3- In the antitarget we have lots of part with 1X or 0X for the depth of coverage (maybe 80-90% of the antitarget). Is it a problem ? Does CNVkit make differencies between antitarget with coverage and antitarget without coverage or specificity ?

thank you for your answer and sorry for my bad english

Alice

Answer 1

[Eric T.]

Thanks for trying out CNVkit.

1. You can ignore this for now. The on- and off-target bins ideally should contain roughly the same number of reads. It's better for a bin to contain too many reads than too few (coarser resolution vs. reduced accuracy), so the default antitarget bin size is a "safe" value that may be larger than optimal (as you determined in #2).

2. That's fine. An antitarget bin size of 12,000bp (12kbp) is reasonable for a target bin size of 100. If your results look noisy (e.g. if the source is somewhat degraded FFPE tissue), then try increasing the target bin size to the default (267bp) and the antitarget bin size accordingly to somewhere between 24-32kbp (since most targets will be limited to the size of the exon, ~200bp in human). CNVkit normalizes the bin coverages separately so that they're comparable.

3. Not a problem at all, that's typical and the reason we use large off-target bins.