low theta + equal estimates of M

[andersgs]

Hello all.

I am running Migrate 3.1.7. I have two populations plus a ghost. I ran
some smaller jobs before getting to the run below, but these values
seemed to me to be giving a good estimate of the posterior. To obtain
a final estimate, I am ran 50 replicates, with a Bayesian approach and
4 heated chains with adaptive swapping, exchanging every 10 steps,
with default temperatures. I am used random values to start each
chain, picking from a uniform distribution that is within the bounds
of the priors. Both theta and M have uniform priors (0,0.25,0.025 for
theta, and 0,30000,3000 for M). I am running with a 200K burnin,
sampling a total 5K steps spaced 100 steps apart. I get decent ESS and
autocorrelation values. But, my theta values are hugging zero, and the
posterior distribution of all M values are almost exactly the same. I
find that very hard to believe. I wonder if there is something going
on. I would appreciate any thoughts on this matter.

I am using only one mtDNA locus. Since I don't have access to PAUP, I
have used both Mega 5.0 and jModelTest to estimate Ti/Tv ratio, but I
get widely different values (15 and 55 respectively) for a HKY-G,
which seems the closest to an F84-G of Migrate.

I should also mention that I am running this for a species of
mosquito. So, I am not entirely sure the 0.1 max for theta would be
sufficiently large. Nevertheless, the posterior has a peak on zero
(with over 25% of the HPC including zero) - I have tried increasing
the number of bins to 10000, but to no avail. On a more general note,
should we be using 0 as part of the prior in Theta?

Thank you.

Anders.

PS: These runs are all on the Mac SL version of migrate, on an iMac
running 10.6.4.

[Peter Beerli]

Anders,
I suggest not to use the adaptive swapping as this is often not helping because on some datasets it seems to converge on temperatures that are all the same (1.0),
With mtDNA it seems really hard to believe that it hugs zero. I would like to check
this, could you send me your infile, parmfile, outfile.pdf (to beerli at fsu dot edu)

thanks,
Peter

> --
> You received this message because you are subscribed to the Google Groups "migrate-support" group.
> To post to this group, send email to migrate...@googlegroups.com.
> To unsubscribe from this group, send email to migrate-suppo...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/migrate-support?hl=en.
>

[YC Tay]

Dear all,

This is quite an old thread, but was this issue ever resolved? I am running migrate on a small subset of my data of 100 loci, and my theta also hugs zero, while M values are very similar between all my 4 populations, which I also find hard to believe. The posterior probability histograms mostly look unimodal and the acceptance ratios are 1.0. I've run this several times with increasing sampling, but the theta is still close to zero. I wonder if it is due to the starting prior setting of a very low theta (settings below)?

theta=OWN:0.001

migration=PRIOR:10 

rate=PRIOR:50

split=PRIOR:10

splitstd=PRIOR:10

Should I be running theta=PRIOR:10 instead? But it would be a little disconcerting that the results should depend so heavily on the starting prior!

Attached is my output file for reference.

Any suggestions and advice on tweaking the settings for my data would be greatly appreciated!

With best regards,
yujie

[Christine Ewers]

Dear Peter & all,

now this is a really old threat, but has this ever been resolved? I work with similar data to Anders' (single mitochondrial sequence locus). I am working on a comparative study of 28 marine invertebrates, and for 9 of them, I have the same issue. I am attaching the outfile, as well as the haplotype network for the species. I have increased the range of theta to ten, but still, theta hugs zero. In the majority of these cases, the migration rates are around 400, which also seems oddly similar between very different species.

Thanks in advance,

Christine

[Christine Ewers]

Looking at Beerli 2008, especially fig. 3.5, could the issue be that I did not run it long enough or the prior is too diffuse? Running it longer right now, and will try an exponential prior next. 

[Peter Beerli]

Christine,

one of the key things if you have these triangle posteriors for theta is not to increase the prior range but to shrink it! Make also sure that you datafile really contains some variable sites.

I suggest to try a prior for Theta in the range of 0.0 to 0.01 or even 0.0 to 0.001. This goes often hand in hand with an increase of the prior for M

because if the mutation rate is really low then we expect very small Theta and most likely a large M, but for now the range of M looks fine .

The large prior range spends a lot of time on large values and once in a while accepts them, the closer you go to zero the more often, the binning then does not help.

Could you show mw a run with the lower prior settings?

Peter

To unsubscribe from this group and stop receiving emails from it, send an email to migrate-suppo...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/migrate-support/9adaac40-6ecb-4fda-9e30-eb3b0c42ad40n%40googlegroups.com.

[Christine Ewers]

Dear Peter,

That solved the problem! Thank you! Attached the improved run. 

Christine

[Peter Beerli]

Christine,

Looks great now, you may consider to run another mutation model than jukes cantor and/or look into unidirectional migration.

Peter

On Feb 18, 2021, at 13:45, Christine Ewers <ewers.c...@gmail.com> wrote:

Dear Peter,

That solved the problem! Thank you! Attached the improved run. 

Christine

--

You received this message because you are subscribed to the Google Groups "migrate-support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to migrate-suppo...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/migrate-support/EF684FD5-AC9E-4480-95B7-F188583EC581%40gmail.com.

<Cyanea_capillata_unconstrained.pdf>

You received this message because you are subscribed to a topic in the Google Groups "migrate-support" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/migrate-support/H6bTW3AYcBU/unsubscribe.
To unsubscribe from this group and all its topics, send an email to migrate-suppo...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/migrate-support/B9BD0694-2284-4D12-BA45-0FE7484DBF22%40gmail.com.

--
You received this message because you are subscribed to the Google Groups "migrate-support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to migrate-suppo...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/migrate-support/EF684FD5-AC9E-4480-95B7-F188583EC581%40gmail.com.