Hi Robin, if you have enough tips in your phylogeny, looking first at the transition matrix obtained from the ARD model can help you to know whether or not transition rate are symmetric or not. You can get an AIC and even an AICc from several evolutionary models and to compare them for selecting one or the other model (see the package geiger, function fitDiscrete). This package can also tell you whether you should prefer something ordered or unordered (model "meristic"). Finally for the fourth structure, just play as before, try to see what is independent or not to see whether you have to atomize or not (atomizing may sometimes increases the number of things you want to estimate and also overparametrize your stuff, especially if you have for assumptions that your characters are ordered).

julien

Le 25/08/2017 11:04, Robin van Velzen a écrit :
Hi Julien,

Many thanks for your suggestion! It is a nice compromise between using all three characters separately or combined in a single multistate character.
But I do see some caveats:

First, given the nature of your suggested multistate character, I would
assume it should be ordered, where 0>B>BC. Is that possible?

Second, if I understand correctly the ARD model would significantly
increase the number of rate parameters so I am not sure if that would be a
good idea (given that I have few observations which quickly results
in overparameterization). I think a SYM model would be good enough for my
purposes.

Third, as I mentioned before, I have more dependent characters. I ignored those in my original query to keep things simple but here we go: There is a substructure within the developmental tissue that can also comprise cells originating from the different layers (A, B, AB, or BC). But obviously if a layer does not contribute to the main structure it can also not contribute
to the substructure.
You can see that this additional level of dependency complicates things a lot and I am not sure if there would be a way to atomize this similarly to
your solution for the original three states.

Therefore, I still think the best solution would be if I can keep the
characters separate and specify which character state combinations are
possible and which are not. Unless this is impossible? Or, alternatively, perhaps I can achieve the same thing by combining characters while setting
some rates to be zero?

Thanks again & with best wishes,

Robin

PS the solution does not necessarily have to be in R. I just figured this
would be the best place to ask for advice.


Date: Wed, 23 Aug 2017 16:52:16 +0200
From: Julien CLAUDE <julien.cla...@univ-montp2.fr>
To: r-sig-phylo@r-project.org
Subject: Re: [R-sig-phylo] Mapping complex characters
Message-ID: <58d4d9b746f44b308189fac26f72a...@univ-montp2.fr>
Content-Type: text/plain; charset=UTF-8; format=flowed

Hi Robin, if I understand  C depends on B.

to atomize your developmental set and possibly avoid
overparametrisation, look at A and the combination of BC independently
(with 0, B and BC as states), then use the model ARD in ape, it will
return actual the rates for these two characters.

julien

Le 23/08/2017 14:16, Robin van Velzen a ?crit?:
> Dear all,
>
> I want to reconstruct ancestral states of morphological characters on a
> phylogenetic tree. However, these are 'complex' and I am not sure how
> best
> to appropriately code these characters and do the mapping.
>
> Specifically, I have developmental data where three adjacent cell
> layers
> (A,B,C) can contribute to a particular developing tissue, either singly
> or
> in combination. However, layer C can in principle not contribute
> singly,
> and combinations can only comprise adjacent layers (so AB, BC, and ABC
> are
> possible but AC is not).
>
> A solution would be to code everything as a single character with a
> total
> of 5 states (A,B,AB,BC,ABC). But this would require a matrix that would
> appropriately set the rates of change between all these states. Given
> that
> I do not have many observations I fear that this would result in
> overparameterization (also I have many more than three dependent
> characters
> that I ignored for the sake of clarity). Moreover, reconstruction as a
> single character would not give much insight in the underlying
> processes at
> the level of cell layers.
>
> Ideally, therefore, I would code the three layers separately so that
> all
> three ancestral states can be reconstructed. But that would require
> specifying character combinations that are not allowed. But I am not
> sure
> if that is possible and if there are packages that would be able to do
> such
> reconstructions?
>
> Any advice would be greatly appreciated!
>
> Robin van Velzen
>
>       [[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-phylo mailing list - R-sig-phylo@r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
> Searchable archive at
> http://www.mail-archive.com/r-sig-phylo@r-project.org/



------------------------------

Subject: Digest Footer

_______________________________________________
R-sig-phylo mailing list
R-sig-phylo@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo

------------------------------

End of R-sig-phylo Digest, Vol 115, Issue 8
*******************************************


        [[alternative HTML version deleted]]

_______________________________________________
R-sig-phylo mailing list - R-sig-phylo@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/

_______________________________________________
R-sig-phylo mailing list - R-sig-phylo@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/

Reply via email to