more discussion about GONE

[mufernando]

It is really surprising that GONE is doing so poorly in our simulations. GONE does really well in the papers: HapNE, GONE, GONE structure, GONE empirical.

I propose that we:

• Cite all these papers in the results/discussion
• Ensure that in the GONE plots the time scale is not much greater than say 200 generations (because this is unfair to the method)
• Add a potential explanation to our discrepant results: the CEU/YRI/CHB demography we simulate has considerable migration, which could impact LD patterns in dramatic ways.

What do you think @andrewkern @silastittes?

[stsmall]

GONE doesnt look too bad from 1 - 100 gens in the past. Hard to see with the current scale of the figs containing msmc2, smc++, and stairwayplot2

[igronau]

I'd suggest to show two versions of the GONE inference - one with the complete time range, to allow direct comparison with other methods, and one with a focused range up to ~200 generations back in time. @silastittes - can you prepare these two versions?

[mufernando]

Hi Ilan,

I was wrong, GONE automatically clips the inferred trajectories at 200 generations. But the two other points I made are still useful. I can make a pass later on adding the citations and a bit of discussion.

[igronau]

I saw this now. I'll have a placeholder in the end of this results section for discussing the relative accuracy of each method and you can add details there about GONE

[igronau]

I'm now rewriting the description of the results on Ne(t), and I'm starting to wonder if we even want to show GONE in these comparisons. It seems like both scenarios that we simulated (human OOA and Vaquita) are poorly suited for this method, which is targeted for recent demographic shifts. So, when summarizing the results I end up saying this, to explain why the GONE inference is a bit wacky in both cases. But then it begs the question: if we show it, why don't we consider a scenario in which it's expected to produce useful inference? If I were a reviewer, I would likely ask this.

[mufernando]

it is still useful to know that GONE can predict stable population sizes, I think. The wonkiness of the results likely has to do with migration.

[igronau]

I'm not sure it's (just migration). The GONE estimates look quite noisy also in the Vaquita simulations, which have a single population. We can discuss this next meeting after the updated text and figs are in place