Interpreting histogram of events through time and skyline plot

[Mark]

Hi Peter and all,

I was wondering how to interpret the histogram of 'events through time'. At first I interpreted it as plotting the size of theta or M over time (somewhat like skyline plot), but now I think this interpretation is wrong. The manual states that a histogram that looks like exponential decay indicates the parameter has been constant over time, but I thought it would indicate recent expansion. Should I think of the histogram as plotting the number of times the most likely estimate was estimated from the genealogies over time? Is there any way of inferring changes to population size or migration over time from the histograms?

As for the skyline plots, I have very large error margins which makes me think the results are not very reliable. I used around 1000 nuclear SNPs for the analysis.

Thanks for your help,

Regards
Mark

[Peter Beerli]

Mark,

the skylines show ‘expected’ changes through time, whereas the event histograms show the number of events ( coalescences and migration events) at particular times.

For example in a neutral coalescent, you expect more coalescences closer to today than far in the past. If you look at the migration events and you do not see a pattern with more migration today than in the past the risk of migration was not even over time: for example, A Neanderthal and human population mt dataset for example delivered migration events that peaked some time in the past suggesting there was either a divergence event or a short time of gene flow.

Skyline plots are unreliable in general because commonly you should only consider the time frame from zero to about the time of the size of Theta, e.g. if Theta=0.002 and your skyline plot goes from 0 to 0.2 only the time from 0 to 0.002 will be (if at all) estimated with some confidence. To get good results with skyline plots you will need ancient samples.

 Peter

--
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 post to this group, send email to migrate...@googlegroups.com.
Visit this group at http://groups.google.com/group/migrate-support.
For more options, visit https://groups.google.com/d/optout.

[Mark]

Hi Peter,

Thanks that really clarifies things. By a neutral coalescent, do you mean a HWE population? Would multiple peaks in a histogram be indicative of periodic subdivision or bottlenecks? I would be interested to see any papers on non-neutral coalescence if anyone can suggest something. I will look up the human and neanderthal example.

Many thanks
Mark

[Peter Beerli]

Mark,

coalescence-based programs usually assume HWE, because they are allele based and assume interactions among individuals in populations follow some random mating scheme (Wright-Fisher, Canning, or Moran). If there is selection at a particular locus then that locus may deviate in the timings of the coalescence events (for example they could be much longer than expected because of balancing selection or much shorter because of positive selection. If such signatures are shown on all loci we would talk about growth or shrinkage, temporary reduction of intervals are the signature of bottlenecks (again using multiple loci). To my knowledge there is no inference program for such things (Lamarc can do growth); in the human literature there is considerable talk about this, you may look at applications of dadi (Gutenkunst) and population approaches by Reich and Patterson, this (http://biorxiv.org/content/biorxiv/early/2015/10/09/028753.full.pdf) may give a critical overview of these, but I have not read it yet, 

but looks promising. 

Peter