Have you heard about this tool?

Kofler R, et al. (2011) PoPoolation: A Toolbox for Population Genetic Analysis of Next Generation Sequencing Data from Pooled Individuals. PLoS ONE http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0015925

Here's a link to the google code site:

https://code.google.com/p/popoolation/

Terrible name, I know,

-Mike

@Michael

Yes I have. Even had a play with it but couldn't figure out how it calculated allele frequencies. Do you have any experience with PoPoolation?

Not personally, but I'm asking friends who have used it.

Will get back to you asap.

I'm still waiting to hear back from someone, but one person actually gave up on Popoolation and created their own custom Perl script to count the number of reads per allele and weight allele counts by population size to get a relative allele frequency estimate.

So going back to your question: at its simplest, to move from allele "counts" in your reads to allele "frequencies", one needs to account for the different sizes of the populations being compared.

A friend recommended this paper - but perhaps you have already read it:

Konczal et al. 2014 Accuracy of allele frequency estimation using pooled RNA-seq data. Molecular Ecology Resources 14: 381-392

I get the feeling calculating allele frequencies from pooled data is inherently error-prone because of potential biases during PCR or sequencing, so its best to recognize you are generating a relative estimate, with some amount of error.

Sorry I can't be of more help

Cheers Michael.

If you find out, I would be keen to know what your friend has to say about it and perhaps learn more about their script? Though I wonder how it might be affected when population size isn't readily estimated? Or is the "population" the sample pool?

I haven't come across that paper, but will have a look at if for sure.

I just find it a bit odd that these population genomics pooled techniques haven't got more information around them!

Cheers,

~ Josh

Check out this publication and the associated software:

Gautier M, Foucaud J, Gharbi K, Cezard T, Galan M, Loiseau A, Thomson M, Pudlo P, Kerdelhue C and Estoup A (2013). Estimation of population allele frequencies from next-generation sequencing data: pooled versus individual genotyping. Molecular Ecology. 22(14):3766-79.

http://www1.montpellier.inra.fr/CBGP/software/PoolSeqUtils/

http://doi.org/10.1111/mec.12360

Yeah - by "population" size I mean "sample pool" size

It is difficult because different alleles can have different relative levels, and this can change with each locus and even with the extended haplotype. This is true for RAD, ddRAD, etc. As a simple case, imagine that for a SbfI library, a particular allele at a locus is linked to another SNP 2 kb away that creates a new SbfI site. This affects the shearing of the DNA and that allele is now 3-fold under-represented compared to the other allele. But in other samples that haplotype is broken, and the locus is more evenly represented. Same thing in ddRAD but if the second-enzyme cut site moves and the fragment is a little longer making it PCR less effectively. 

Second, one sample may be more degraded and so is less represented in the pool so it's allele count is missing. Now, you may be thinking, well we don't have to be so precise, but then why not just sequence 10 individuals instead of a pool of 50 that has the precision of 10? The amount of sequencing would be about the same, and that is what is going to drive the bulk of the cost.

@Eric, thanks for the answer. I suppose the question is does a sample of 10 individual sequences provide a better estimate of population genetic properties vs a large pool? The literature I've been reading seems to favour pooling because results do not differ much from individual samples (if not better) and you are more likely to find shared SNPs between populations. But then, there are some things you just cannot do with pooled data (kinship, assignment, parentage) which is kinda annoying - unless you barcoded each individual in a pool differently I suppose?

Could you just elaborate on the following please Eric: 

"imagine that for a SbfI library, a particular allele at a locus is linked to another SNP 2 kb away that creates a new SbfI site. This affects the shearing of the DNA and that allele is now 3-fold under-represented compared to the other allele. But in other samples that haplotype is broken, and the locus is more evenly represented."

Say each cut site (allele) is represented by an 'X':

X1------------------X2------------------X3 Individual 1

X1----------------------------------------X3 Individual 2

Say X1 is linked to the X2 in your example: how does the addition of a new cut size under-represent the X2 allele? Please correct me if I am wrong in my interpretation.

Joshua, That's a reasonable image of what I was thinking. The question is how does the X2 site alter the representation of X1 alleles? The issue for RAD is that the shearing efficiency is dependent on the size of the restriction fragment. So the bottom X1-X3 fragment will shear more efficiently than the X1-X2 or X2-X3 fragments. If the top X1 has a different allele than the bottom X1, you might get a 4-fold decrease in top X1 allele reads. This was measured nicely by John Davey in http://www.ncbi.nlm.nih.gov/pubmed/23110438

You can also imagine X1-X3 is a ddRAD fragment. The appearance of X2 will remove the top allele completely from the ddRAD library, since X1-X2 is not the size selection range.

If individuals are sequenced, you can still identify that X1 is heterozygous despite the lower representation for RAD. You'd miss it either way for ddRAD.

On the other hand, you'd probably get good-enough info either way!