How to estimate total genetic diversity from a sequenced population?

[mmanhart]

Hi all,

I am analyzing whole-population sequencing data of E. coli that has undergone experimental evolution in different environmental conditions --- the goal is determine if one condition produces more genetic diversity than the other.

I initially ran breseq to identify SNPs/indels/etc. in each sample; however, there were no obvious differences in the number, frequency, or type of mutations between samples. However, this analysis is only sensitive to relatively high-frequency mutations (>5%), and the genetic diversity we're looking for may very well be manifested in lots of low-frequency alleles.

For that reason, and since we moreover don't care about identifying specific mutations anyway, I'm wondering if I can estimate total genetic diversity just from the overall number of mismatches in each sample's alignment. Of course the vast majority of these mismatches are noise from the sequencing instrument, but if there are enough real mismatches on top of the noise, perhaps I can detect a statistically significant difference in them between my samples.

Is this a reasonable thing to do, or does anyone have any other ideas? Thanks in advance for any comments or suggestions!

Michael

[Brian Bushnell]

If you're very stringent - only allow properly-paired reads with a mapq of at least X, and ignore snps and indels within Y bp of either end of the read and involving bases with at least a QV of Z - and possibly, any mutation seen only once (which will impact your rate of discover of mutations that are non-viable, but should not have too much effect on the rates of evolutionarily propagated mutations) - and, ideally, sequence everything multiplexed together in the same run(s) of the same instrument, with the same library-prep -

Then, I think, you should probably get a good signal-to-noise ratio.

Studying the ratio of synonymous to nonsynonymous SNPs might also be enlightening. I would expect the high-mutation-rate environments to have a lower ratio. If that's not distinguishable from noise (or does not correlate with the overall mutation rates), then perhaps there's not enough signal to be useful.

[mmanhart]

Thanks for the suggestions, Brian! I think that will definitely help to get a more reliable sample of mismatches to compare.

The ratio of synonymous to nonsynonymous mutations would also definitely be useful here, but unfortunately there just aren't enough to use --- the time scale is only ~1000 generations and the mutation rate isn't that high.

Michael

[SNPsaurus]

My academic lab does overlapping paired-end sequencing with multiple barcodes per sample in order to call genetic variants with allele frequencies well below 1% with no false positives. The overlapped reads remove sequencing errors and the different barcodes identify PCR errors. Can you start over or are you stuck with the sequencing data you've already generated?

[mmanhart]

Thanks for the suggestion, SNPsaurus. This is only a small part of a larger project so I'm probably stuck with the sequencing they gave me, but I'll keep this idea in mind for the future. Is this anything like duplex sequencing, e.g., http://www.pnas.org/content/109/36/14508?

Michael

[SNPsaurus]

It's more like ORP http://www.biomedcentral.com/1471-2164/14/96 but we add extra barcodes to filter PCR dups.