criteria for filtering SNPs

[BhariD]

Hi,

I have variant (SNP) calls made from the GATK (vcf format). I have a tumor data-sets (No normal data-set). I have filtered out the SNPs that are present in the public domain such as dbSNP as I am interested in looking at the novel SNPs in the tumor data-set. I am also excluding anything that is present in COSMIC as well.

In an attempt to gain more confidence in the novel SNP calls I plan to filter them based on SNP quality controls such as read depth, mapping quality, Fisher p-value, Phred score, or variant confidence/quality by depth. I am afraid of selecting arbitrary thresholds for these variables. So my questions are:

1) What is the best approach to SNP QC filtering?
2) What variables (among the few I mentioned) are the most relevant to determine a SNP quality? how should I set these thresholds?

I would really appreciate your insights and feedback on this. Thank you!

Regards,
BhariD

[ron128]

Hello, The most relevant criteria in my experience is the altered allele depth. I personally do not recalibrate my variant files, so I dont rely on the GATK quality filter score. In general, I can validate a lot of the variants which have a high altered allele depth. What altered allele depth you choose, depends on your sequencing coverage. There is no absolute value for this. Go too high, and you miss out on precious variants, especially heterozygous, or go to low, and you are left with a lot of false positives. If I achieve a coverage of 100X, i optimally filter out all variants which are not covered in atleast 10 reads harbouring this mutant allele. Hope this helps! As far as i know, phred score is not taken into account by the unified genotyper, though I could be wrong on this. phred score as of right now, is not incorporated in any of the variant calling filters/criterias, as far as I know. It is a very good criteria however, and the closest some one has come to integrating it in SNP variant calling, is in the implementation of BWA as BWA PSSM.