ddRAD size selection with Ampure beads

[Todd McLay]

Hi there,

I am looking at using a modified ddRAD protocol in a genus of plants with a genome approximately 1000Mbp in size. I do not have a Pippin Prep available, so instead I have been playing around using double-ampure size selection. I use a 0.65-0.75x ratio on the initial digest (with EcoRI and MseI) and after adapter ligation and PCR amplification my library size is between 400-600bp (or close to, only based on gel photos).

Using the available genomes of the most closely related organisms I can find, Allium and Asparagus, I estimate that this would be between 30k-100k ddRAD sites in my genus.

However, I note that size selection in ddRAD is often focused on a +- of about 50bp. I am worried that with the wideness of the AMpure will give more fragments than I expect, and my coverage would be too low.

I am looking to pool ~120 or so samples on a Hiseq 2000 lane and am worried that I might be on the low end of coverage.

Cheers,
Todd

[nucacidhunter]

If we take higher end of your fragment number estimate and assume that by size selection half of the resultant double digested fragments are present in your libraries, there will 50Kx120=6 M fragments. A lane of HiSeq with 150 M reads would give an average 25x coverage. If 25% of reads are not informative, the coverage will be around 18X. Depending on your intended downstream application, ploidy and relatedness of samples this coverage might be enough. However, a tighter gel cut still would be better option than double SPRI size-selection.

[SNPsaurus]

I find people usually underestimate the number of reads needed when multiplexing. A typical issue is that the number of reads per sample may have a 4-fold range (some get 500k reads, some 2M reads). There is also a wide range of depths at different loci (some will have 10 reads, some 200 reads). The locus variation is usually consistent across samples though. And then, as nucacidhunter mentioned, some % of reads are lower quality, don't align or have other issues that prevent them front being used productively.

It may not matter for your analysis, but of the 120 samples, 40 may have 9X depth on average instead of 18X. And of your 50k loci in those 40 samples, 25k of the loci may have 4X depth. Before you start is the best time to check if your statistics are robust to missing alleles and other issues this variation will create.

[Todd McLay]

Thanks for both of your responses.

I have been fiddling with calculations for a while now, it's a daunting task to try and determine the best way to do it when a failed run costs so much.

I think I will take your advice and use a gel cut in the final library to narrow the size range.

I am intending to use ddRAD for phylogenetic purposes. Other papers I have read set the minimum coverage for loci as low as 4x, but I don't find that as satisfying as a coverage greater than 10, which was why I decided to multiplex 120 or so for the 50-100k loci I assumed.

Regards,
Todd