too much CNV detected

[lmanchon]

--Hi,

i have tested your tool and i obtain too much CNV, with GATK pipeline or CNVkit i obtain ~ 15 CNV whereas with Cobalt i have ~ 130 CNV. I don't know why ,i have used default parameters maybe i need to adjust some parameters.

best

[brendanofallon]

Hi there, this can happen for a couple reasons. First, it might be that the background samples used to generate the model aren't a great fit for the query sample (for instance, maybe they had different depths or hyb capture conditions). Second, as you mentioned it could be that the default parameters aren't quite right. For instance, the 'a' and 'b' params could be a bit too high (higher values favor sensitivity, but generate more calls). Reducing a and b to smaller values will probably reduce the total number of calls.

[lmanchon]

Can I suggest you take a look at my input files(my_background_depths.bed & my_sample.coverage.bed) to see if you have any parameter ideas to modify ?

[brendanofallon]

A good first bet is to try setting the 'a' and 'b' params to something lower than the default, try running the calling step with ' -a 0.002 -b 0.002' (the default is 0.005), that should help improve precision

[lmanchon]

I tested it and got 14 CNV less.

[brendanofallon]

A good idea would be to experiment with different values for a and b to find the ones that give you the right balance between sensitivity and precision. Sounds like both 0.002 and 0.005 are too high, try a few smaller values. You can also try increasing the '--min-qual' argument from the default value of 0.8 to eliminate more quality calls. If the first try doesn't look quite right, then you might need to try a few different values to find one that works well

[lmanchon]

ok i'm going to do that. Thank you --