BEAST skyline plots population structure

[Vanessa Bieker]

Hi,

I am new to BEAST and would like to use the skyline plots analysis for my dataset. The tutorial says that for that, the population needs to be well mixed as the analysis. I however have some relatively old population structure in my dataset (3 main clades that group based on geography). The tutorial says that in the case of population structure, one should use a structured model instead. I tried to find some more information about how to set up such a structured model but could not find anything. 

Basically, I would like to do the skyline plots for each of the 3 clades separately to account for the structure. Is that possible to do in a single analysis or would I need to run BEAST on each of the 3 clades separately? 

Thanks a lot in advance!

Vanessa

[Omar Mejía]

Dear Vanessa,

There is a paper of Heller with Buffaloes that deals with the lumping and splitting approach. If you have three well-supported clade, either by geography or fbaps or a similar approach you can perform the BSP for each one of the genetic clusters, just remember that coalescent-based approaches are too sensitive for sample size (thumb rule at least 30 individuals), but also, you could make the BSP of all individuals without regarding the geographic location, i have reported both and also see several similar papers.

Best regards

Omar

Libre de virus.www.avast.com

--
You received this message because you are subscribed to the Google Groups "beast-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beast-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/beast-users/18870d35-761d-4e64-978a-704cbb37b32cn%40googlegroups.com.

--

Omar Mejía G

[Artem B]

Hi Vanessa

I think Omar means this article: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0062992

But approaches or subsampling strategies were called as local, pooled, scattered. And they are very demanding about sample size for each population.
In our study of tick-borne encephalitis virus, the analysis of different highly isolated population together demonstrated some kind of a "chimeric" effect (top plot), when skylines of each population fused into one skyline (https://www.mdpi.com/1422-0067/24/3/2921):
 

I also got this effect in my simulation analysis, but the data is not published yet.

The good news is that Nicola Muller et al presented the skyline approach for structured coalescent, based on their MASCOT package for BEAST2: https://www.biorxiv.org/content/10.1101/2024.03.06.583734v2.full

I didn't try this yet, but as far as I know about MASCOT and other structure coalescent models, they are very sensitive to priors. So you should run analysis with different priors to check if there is a difference in results (maybe Nicola correct me here if he notice your discussion).

Best,
Artem

среда, 18 сентября 2024 г. в 02:37:25 UTC+8, Vanessa Bieker:

[Vanessa Bieker]

Thanks a lot for the very helpful and quick replies! 

Best,

Vanessa