Variation calling, pool samples or not?

[dan]

I seem to remember from talks and manuals that it's good practice to pool reads from all available samples when calling (per-sample) variations, as the added depth improves call statistics.

Can anyone point me at literature or discussion on this specific point, as I can't find anything concrete.


Cheers,
Dan.

[severin]

GATK best practices

I seem to remember from talks and manuals that it's good practice to pool reads from all available samples when calling (per-sample) variations, as the added depth improves call statistics.

Can anyone point me at literature or discussion on this specific point, as I can't find anything concrete.


Cheers,
Dan.

I was literally just reading about that here.

http://www.broadinstitute.org/gatk/guide/topic?name=best-practices

Though not sure what is recommended for the Haplotype Caller as it is still a little experimental.

[dan]

Nice link. It's very clear, but there isn't much detail, i.e. Why are samples called together?

Not surprised that haplotype calling is up in the air ;-)


Cheers,

[dan]

Here is a good answer :-)

http://www.biostars.org/p/10926/

[dgscofield]

Especially if you're after allele frequency spectra, estimators are generally better for pooled samples. The 2010 Genetics article by Futschik and Schlötterer will get you started.

One common-sense statistical issue for calling SNPs in individuals is that there are stochastic allele-specific coverage biases up and down around expected coverage for any sample, nothing you can do about that. Pooling samples reduces the influence of this error term relative to the detection threshold for reasonably-frequent alleles. If there is weak evidence for a SNP in a single individual considered alone but that same SNP is segregating at reasonable frequency within the population, that prior knowledge strengthens the evidence for the SNP in the individual.