Truth set for NA12878 SNVs

[alonie]

Hi. I wonder if anyone can direct me to a complete 'truth' SNV set for the 1000genomes NA12878 individual (and, ideally, her mum and dad too). There are a number of existing NA12878 vcf files from various sources, with many or most of the variants in common, but I'm thinking that there must be one 'correct' vcf that the Broad refers to for evaluation etc? The 1000 genomes CEUtrio vcf at ftp://ftp.1000genomes.ebi.ac.uk/vol1...notypes.vcf.gz is a couple of years old now and I'm guessing there's a better call set somewhere. Most particularly I'm after one that is comprehensive, i.e. very few false negatives, as we're looking to evaluate FNR for various sequence depths. Maybe there is one generated from one of these newer very deep sequencing sets for NA12878:
ftp://ftp.1000genomes.ebi.ac.uk/vol1...878/alignment/
which is around 60-70x, and

http://ftp.1000genomes.ebi.ac.uk/vol1/ftp/technical/working/20120117_ceu_trio_b37_decoy/

which look to be ~100x. I think the 1000genomes public VCFs are generated from much earlier GAII based runs?

Any help appreciated

Andrew

[terhorst]

I have yet to find a satisfying answer to this question despite a lot of searching. In truth, there does not seem to be any "truth" variant callset for NA12878 -- everything I have seen to date was generated on the same platform and with the same (or inferior) informatics software as that which you find yourself trying to validate. For SNPs with a relatively high MAF you can look at earlier SNP chip genotyping datasets for this individual, but those will not help you with rare variants. To my knowledge the highest-quality and most recent resequencing data for NA12878 (in fact, the whole trio) is the so-called Platinum Genome released by Illumina. When I examined the raw reads, however, I still saw enough noise that I would hesitate to classify as Truth whatever callset you managed to extract from them.

[alonie]

Thanks. I hadn't seen the Illumina platinum data. Where exactly are the VCFs though?
I know Complete Genomics claim an error rate of ~3 SNVs per MB currently, but not sure how they evaluate that.

[alonie]

OK found the VCFs: ftp://ftp.platinumgenomes.org/trio

I will do some comparisons and post here