Flat history after adapting HapNe-LD to a non-model species with very low LD

[twooldridge]

Hello,

I've been working with HapNe to try and obtain recent demography for a non-model system I'm working in. After tweaking some of the hard-coded indices/ranges in tools.py and creating a file of genomic regions, I'm able to successfully run everything. However, the output history is entirely flat with narrow confidence intervals:

Obviously this does not seem realistic. I suspect that some attributes of my system are contributing to this - historical Ne is likely quite high (~5e5), and we do see that LD decays VERY rapidly:

I've attached a tarball of the output folder with GENOTYPES, LD, and HapNe, where you can see the ld and ccld calculation outputs. Based on this, do you have any suggestions for further tinkering and exploring why HapNe might be producing this result?

For what it's worth, some other considerations:

1. My recombination rate map (see attached) is inferred from population resequencing via pyrho. In general, cMMb is very high. Whether or not that's real is another question.
2. I'm inferring based on 8 diploid individuals I phased with whatshap & shapeit2.
3. I'm using a small set of genomic regions (see attached) to expedite these tests, although there are still lots of SNPs in this

Thank you in advance for your help, and please let me know if I can provide other info!

output
rmaps
regions_testAB.txt

[mufernando]

@twooldridge you should take the cM/Mb estimates from pyrho with a huge grain of salt. pyrho estimates rho (population recombination rate) and then scales that to r (recombination rate). To do so, it has to make a lot of assumptions about historical Ne and mutation rate.

I would scale down their estimates such that the average is closer to 1cM/Mb, unless your species is super special. Are there any pedigree-based or linkage-based estimates for any species closely related you yours? If so use that average in the rescaling.

I don't know much about HapNe, I was just lurking on the issues here.

[twooldridge]

Hi @mufernando I appreciate the suggestions! Unfortunately there are no map estimates from such sources for even more distantly related species - this is a very non-model case. However, I don't think there's a reason to suspect this species is special, so downscaling to ~1cM/Mb does sound like a nice approach.

Regarding pyrho and inference, I feel pretty good about mu/Ne estimates because of small pedigreed sample sets, but I definitely see your point. Unfortunately these pedigreed sets didn't have F2s, so no estimate of r there.