Question about SGS computational analysis for Cancer Genomics

[changjpark]

I am currently working on a project where we have cancer and normal patient samples, in which we developed a pipeline for variant calls for SNPs and Indels. Before I can analyze it, I need to make sure the pipeline is suited for cancer genomics.

As for the pipeline, I've used BWA for alignment and GATK's Unified Genotyper for mutation calling. I understand that this method is appropriate for non-cancer sample analysis, as commented in Nature Review: http://www.nature.com/nrg/journal/v1...l/nrg2841.html

My problem comes from needing to call the somatic genome alteration detected in tumour and not detected in the normal data, as a differential analysis. From GATK, I've taken a look at "SomaticIndelDetector" (http://www.broadinstitute.org/gsa/ga...torWalker.html), which seems to call Indel, but ignores all the other possible SNP mutations. Beyond this, I have not had luck in finding online sources for appropriate Cancer Genome SGS analysis with Tumour-Normal paired analysis.

Can I ask those who have insight regarding to this matter for the advices and latest tools that I can use? Thanks in advance.

Regards,

CJ

[adaptivegenome]

Well one problem is that you want compare normal to tumor. The second is most mutation callers make assumptions about the number of alleles (in humans, the expectation is 1-2 alleles per locus). However in cancer samples, there is heterogeneity and thus there could be any number of alleles. You don't want to employ a method that incorrectly assumes the sample contains only one genome.

[aurelielaugraud]

have a look at ...

http://www.broadinstitute.org/gsa/wiki/index.php/MuTect

http://gmt.genome.wustl.edu/genome-music/current/

[changjpark]

http://www.broadinstitute.org/gsa/wiki/index.php/MuTect
http://gmt.genome.wustl.edu/genome-music/current/

Thanks for the link. MuTect is not public, but I can definitely use the MuSIC.

The thing about MuSIC is that it will work with a candidate set of mutated gene list, but I'm not yet in this stage. Is there a standard or published protocol/tools for obtaining candidate mutations in normal/tumour pair samples from same patients? I've checked out SNVMix and UnifiedGenotyper, and would like your opinion in their appropriateness for the scenario.