best-practices for assigning read-groups for snp-caller

[brentp]

Hi, We have samples from 4 individuals from 2 tissues each == 8 total:

tissue is blood/lung
individuals are control/disease

so, let's (ignore that 2 samples is too few and) say
lc1,lc2,bc1,bc2,ld1,ld2,bd1,bd2
where l == lung, b==blood, c == control d == disease, and the last number is the individual of that disease category.

If we want to call lung-specific, disease-related SNPs, is it better to send to the caller with ld2,ld1 vs all others by giving ld1,ld1 a read-group of disease and the rest a read-group of control?
Or, is it better to just call snps where each of the 8 groups has it's own read-group and pull out the SNPs by intersecting post-hoc?

I'm not asking wether it's better to send in subsets, but how to divvy up the subsets.
thanks.

[Dameon]

I would use GATK available from the Broad's Genome Analysis Toolkit and indel realign, recalibrate, and call and filter genotypes following the Broad's recommended best practices. Then I would take the finished and annotated vcf file of just the SNPs (-glm of SNP in the Unified Genotyper) and use vcftools to convert the vcf file into Plink or Beagle format. Plink or Beagle would be better tools to conduct your association testing rather than try to perform a series of intersects. If you are familiar with SAS and JMPGenomics, I believe the latest version of JMPGenomics can readily import vcf format SNP files and SAS is an excellent platform to conduct any number of test you wish to perform. Good luck.

[brentp]

Thanks Dameon, do you know of an example of this in Plink or Beagle?