Had some email exchange with the author of cn.mops about somatic variants detection. Think this may also be interesting to other users, the author suggest that I post some of the conversation here.
1.) Germline CNV detection: This is the default setting for "cn.mops" and it was developed/published for germline CNVs. You should use as many normal samples as possible. I would suggest at least 10.. the more the better. Use a standard cn.mops run. There should be "few" CNVs detected that cluster at specific regions ("cnvr" slot of the result object) - you can quickly check for overlaps with previously detected CNV-regions (e.g. Hapmap).
2.) Extract the CNVs of ONE individual: The "cnvs"-slot contains the CNVs of all individuals. You should use only the CNVs affecting the individual (let's call it "A") with which you continue the analysis.
3.) Somatic CNV detection: The function "referencecn.mops" of the package. Use the tumor sample of "A" for the argument "cases" and use as many "normal" samples as possible for the slot "controls". referencecn.mops constructs an "average" genome from the controls you submitted. The resulting CNVs are both germline and somatic. You can subtract the germline CNVs (step 2.) from these and you will end up with only the somatic CNVs.
Some problems which could affect analysis: a) Too few samples for the germline CNV detection -> results could be very unreliable. b) Used all CNVs from the germline cn.mops run. You should only use the CNVs affecting sample A. The "cnvs" slot contains a column "sampleName". c) The parameters for "germline CNV detection" are a bit more sensitive than the parameters for "referencecn.mops". Germline CNVs are typically smaller and rarer than tumor-CNVs. This is why there are different parameters.
These are the three reasons which may result in many germline CNVs.