Understanding projecting down

[Zach N]

Hello all!

I am trying to understand projecting down a 1 population SFS to deal with missing genotype information. From what I understand, when you project down, you use all SNPs with calls for at least as many individuals as you project down to, and SNPs with fewer are discarded.

When I sum the SFS produced by projecting to the actual sample size, I get as expected an SFS with the number of the SNPs in our dataset that are called in all individuals. However, if I project down to 60% of the actual sample size, I get more SNPs, but not as many as if I counted up how many SNPs in our dataset had genotype calls for 60% of individuals. Is this expected and all sites with 60% calls are still being used, but the method doesn't necessarily produce an SFS with that number of SNPs?

Thank you all for your help!

Zach

[Ryan Gutenkunst]

Hello Zach,

Yes, this is expected. The explanation is that when you project down, SNP frequencies get “smeared out” by binomial sampling. For example, imagine you have a SNP observed in 1 out of 10 sampled chromosomes. If you were to subsample 6 chromosomes, often that SNP won’t be segregating in the subsample. A similar thing happens in projection, where we’re averaging over all possible subsampling.

Best,
Ryan

> --
> You received this message because you are subscribed to the Google Groups "dadi-user" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to dadi-user+...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/dadi-user/2f71e4bb-d72b-4ed2-8b5a-afc3f516667bn%40googlegroups.com.

[Zach N]

Thank you for the explanation, Ryan! That makes a lot of sense and clears things up.

Best,

Zach