How to kill a mouse?

[Baseless]

Dear forum,

I have to deal with a xenograft experiment, where a human tumor was bred in a mouse system. I have performed exome sequencing (Agilent capture, 2x100bp Illumina Sequencing), ran most of a BWA/GATK pipeline and finally compared the xenograft tumor with the original patient sample with varscan2 and mutect.

I find that I have a lot more (like 100x) putative somatic SNVs than I am used to see comparing a normal reference and a tumor, most likely because I have some mouse reads in my sample.

Now I would like to remove the mouse reads from my sample. The most straightforward thing I can think of is making up a combined mouse/human reference and filter out everything that aligns to a mouse chromosome.

On second thought, I'd rather like to ask if anyone has done something similar before and what the most acceppted way to deal with this.

So, how do I kill a mouse?

Kind regards,
Baseless

[sBeier]

So you are working with data derived from laboratory animals which were bred for developing a disease and thus, at some point had to be sacrificed.
Don't get me wrong, I have worked and will work with this kind of data as well. I do not want to start an animal rights discussion here, as well. There's just the fact that at least you should be aware of it. And never, never again, in this context use this kind of thread title. Thank you.

[mcnelson.phd]

I just attended a short bioinformatics gathering with people from Jackson labs, and I know a few people from Jax discussed that they were working on similar things to what you're looking for. I can't remember the names of anyone specifically that you might want to contact directly, so your best bet will be to look on their website for people that have done work in that area and contact them.

[Baseless]

Dear sBeier,
It was not my intention to disrespect any laboratory animal. Therefore I changed the thread title.

[sBeier]

Thank you. I did not mean to be picky. But, especially in this context found here, imagine how someone might react to it that just had to sacrifice mice.

[rhinoceros]

I would just bowtie2 the reads against the mouse genome. Refer to the manual for optimal settings..

[Björn_P.]

To answer your question:
The filtering approach wold be best. You take any mouse reference and map your reads against this reference. A very strict mapping condition should only align "mouse reads" to it. The leftover reads are those you want to keep

[rflrob]

I've done a similar thing in flies, pooling, then disambiguating reads: http://www.plosone.org/article/info%...l.pone.0071820

The strategy I used was to make a combined reference, map reads to that combined reference using a pretty permissive set of options, then look at reads that multiply map, and if there are more than N differences between the best mapping read for species A and species B, assign specifically to A, otherwise put it in a pool of "ambiguous" reads.

For the evolutionary distances I was using, N=3 was a pretty good tradeoff between being strict and actually getting enough data. There's a poorly commented script for doing it: https://github.com/petercombs/EisenL...ssignReads2.py . It assumes that in the reference files, all the chromosomes are preceded with species_, so I had chromosomes like dmel_2, dmel_3, ..., and dvir_2, dvir_3, ... etc.