Joint variant calling in trios

[Kath]

Hello,

I have exome sequencing data for a few sets of trios in which the offspring have phenotypically related but probably genetically distinct diseases. The parents are unaffected so I am particularly interested in identifying de novo mutations in the offspring.

My plan at present is to analyse each sample separately up to the variant calling phase then call variants jointly in each trio. I was going input all three (mother, father, child) analysis-ready BAM files to the GATK UnifiedGenotyper to obtain one vcf per trio and then use GATK's PhaseByTransmission and ReadBackedPhasing tools to perform phasing and finally filter.

I know you can pass a ped file to the UnifiedGenotyper but does anyone know if it uses pedigree information to infer the correct genotypes? If not, can anyone suggest a good tool (preferably one that takes in and outputs a vcf) for joint calling in trios? I have heard of a few: FamSeq, PolyMutt and DeNovoGear, but I'm not sure of their relative merits.

Finally, is it better to call variants with the UnifiedGenotyper on all the trios combined, or each trio separately?

Best wishes!

[Zaag]

I would read this: http://gatkforums.broadinstitute.org...igree-analysis

I would call each trio seperate with the HaplotypeCaller.

ReadBackedphasing is not very useful, but PhaseByTransmission is where you find mendelian violations, this really helps a lot in trios.

[Kath]

Thanks Zaag. I have given the GATK website a read and also watched all the videos from their workshop on best practices. I realise HaplotypeCaller is supposed to be better than UnifiedGenotyper so I might update that later when I have a working pipeline. I just don't think they make it clear whether the information in the ped file is used by their calling algorithms to improve accuracy. I have now posted this question to the GATK forum.

[Zaag]

It doesn't always use the ped, but in general I think it is better to perform realignment etc with the whole trio (or family) at once. Specially when using the unifiedgenotyper it can be handy if the realigner realigns all the reads to the same location (for repeats). For the recalibrator I think the more events (SNP or InDel) the merrier.

As far as I know the ped file only is required with PhaseByTransmission, but it would be interesting to hear what other walkers use the information.

[Kath]

So, UnifiedGenotyper does not use pedigree information. I guess I'll need to use a third-party tool.