search

Bioconductor / copy-number-analysis

Explore, compare, and evaluate Bioconductor packages related to genomic copy number analysis
21 stars 12 forks source link

one CNV was split into multiple parts,how to tune it #6

Open tamadezhenshiyuncai opened 4 years ago

tamadezhenshiyuncai commented 4 years ago

cn.mops is helpful for me

i met that one CNV record (validated by other golden standard) was split into multiple parts, the result confused the clinical interpretation, is there some parameters could change this situation?

chr15:23679001-28547200 image

while we plot it by depth, we can see it is sure a long CNV, not many parts image

any advice, i appreciate it very much, thank you !

gklambauer commented 4 years ago

Thanks for bringing this up. This is indeed a problem of the segmentation algorithm that can be handled by changing the parameters of the cn.mops function to "segAlgorithm='DNAcopy' " and -- most importantly -- the parameter "minWidth" should be increased (default 3; increase to 5,8, 10, or 20). A different approach would be to risk to oversee a couple of short CNVs, but avoid the splitting of the longer CNVs -- this can be handled by decreasing the resolution of the readcounts: strongly increase the parameter WL (window length) of the function getReadCountsFromBAM.

tamadezhenshiyuncai commented 4 years ago

Thanks for bringing this up. This is indeed a problem of the segmentation algorithm that can be handled by changing the parameters of the cn.mops function to "segAlgorithm='DNAcopy' " and -- most importantly -- the parameter "minWidth" should be increased (default 3; increase to 5,8, 10, or 20). A different approach would be to risk to oversee a couple of short CNVs, but avoid the splitting of the longer CNVs -- this can be handled by decreasing the resolution of the readcounts: strongly increase the parameter WL (window length) of the function getReadCountsFromBAM.

thank you very much, i'm trying these parameters, they all worked to a certain extent(didn't try WL), but no one output the CNV as an complete one. i will try the WL parameter, ALSO try different combinations of these parameters . at the same time i run the cnvnator software(also for CNV calling ), and it output the CNV as one, but this tools output too many other CNVs, and high FP，maybe these two tools combination will work well