ChristofferFlensburg
commented
6 years ago Haha yes! So for anyone reading, Gil sent me an email with this question and I asked him to post it here, as there is a solution.
There is support for different sets of reference normals, although it isn't very well documented, sorry. I added it mainly for myself in a similar case, so I’m happy to see that others use it as well. There are definitely cases where this is the best option, as the coverage reference normals ideally should have similar biases in as the studied samples in terms of coverage. So the coverage reference normals should ideally share all biases affecting coverage, such as PCR, fragment length, which lab, the person doing the library prep, capture kit, and so on. You may not always have many samples like this. The reference normals for the SNV calling are mainly picking up rare germline variants and alignment artifacts, which you can use a much wider range of samples for.
To do this in superFreq, use the "normalCoverageDirectory=“ option in the superFreq() function. If not entered it defaults to “normalDirectory=“ and it runs with the same reference normals for coverage and SNV filtering, but if entered (and they are different), then the “normalDirectory=“ will be used for SNV filtering and “normalCoverageDirectory=“ will be used for the coverage part in the CNA calling.
gilhornung
Hi Christoffer,
Here is an unexpected question for you :P
As I understand it, superFreq uses the reference normal samples for two purposes: a) To calculate the differential coverage over capture regions: The multiple normals are important for estimating the coverage variance using limma-voom. b) To filter variants: If a certain variant is present in any of the normal it is flagged as "Nnc" or "Nnn".
I am working on a case in which I feel I need to use different pools of normals for these two purposes. I will briefly describe it:
I have data from different batches of FFPE samples, which all passed the same exome capture and sequencing protocol. Each batch has a different quality (possibly due to degradation etc.). When I try to use all normals for the limma-voom step, then I get terrible results with the estimation of fold-changes in coverage. It makes sense, because the large variation between batches distorts the variance estimation of limma-voom. The solution is to use only the normals from the same batch for the differential coverage step. However, since all samples underwent the same exome capturing and sequencing, I still want to use all the normal samples to flag suspicious variants.
Is there a way to use different sets for normals for the differential coverage and variant filtering?
Thanks!
Gil