classifying a variant as het or hom

[lre1234]

Hi all,
I have a general question. Assume I have a genomic position which had nice sequence coverage and quality at the position (i.e. all metrics to classify the position as a believable mismatch to the reference are met). At what ration of ref basses to alternate bases would you consider the position to be heterozygous vs homozygous.

As an example if the coverage at the position is a total of 50 reads, and 48 are of the alternate allele and 2 are of the reference allele. Would this be classified as a homozygous for the alternate or heterozygous? What about even if it was more extreme with a 49:1 ratio (alt to ref)? Also what if there was even higher depth of coverage and a single read was different (99:1 ration, alt to ref)

Also, assume this is not a tumor and the DNA from whole blood.

This is just something I have been struggling with.

[dpryan]

I doubt most packages use a set threshold, since they can just assign a probability to a call and let the user filter by that. I know GATK uses a bayesian model to get the probability and determine the call (I think they describe this in their paper). I suspect that samtools does the same sort of thing. I'd be very surprised if freebayes didn't do this as well.

[lre1234]

I guess my question was a little more general, in the sense of what most people would consider to be a true variant, het or hom variant, rather than what the caller is saying it is. Also, I do realize that it might also be study specific. But I'm just trying to get a general sense of what people may do, I know there are no hard set rules on this.

[dpryan]

Ah, pretty much the only time I look at SNPs manually is with sanger sequencing, so hopefully others can chime in here.

[vivek_]

I'd think it'd be hom alt unless there is some kind of extreme strand bias with the 48 reads supporting the alternate allele. It would make it lot easier if you have the parental lines sequenced as well.