Hi!
Liam, Tony, Sam, thank you for the answers! They forced me to think on type of
model I've chosen.
I forgot to mention previously that I have used only one input tree. I choose
the model ARD, because it was the best model (among SYM and ER) according to
fitDiscrete (geiger).
Installation of the new phytools did not change the results. As well increasing
nsim (from 1000 to 5000) did not change rates qualitevly. I still have two
peaks of Q-rates.
What is interesting these happened only for transition state1 - state 2 and
backward (state 2 - state 1). So I decided to check SYM model in make.simmap.
The result is more convenient – one peak rates' distributions, although
distribution of Q-rates were asymmetrical with long right tail. Make.simmap
Q-rates with ER model looked much more gaussian. All of the analyzes were
simulated with nsim=1000, Q=”mcmc”,pi=”estimated” with one input tree.
The other question is which model to choose – ARD or SYM, or ER? I decided to
average the log likelihood estimates of the output simulations. Is it correct
to choose the model which have the maximum mean logL? Mean.logL.ARD = -84;
Mean.logL.ER = -82; Mean.logL.SYM = -77
Thank you!
Maria
--- Original message ---
From: Tony Gamble UMN gambl...@umn.edu
Date: 27 January 2015, 18:13:53
Hi Maria,
In addition to Liam and Sam’s comments you may want to look at some of your
trees individually. It is possible you have conflicting topologies that fit
quite different transition rate matrices. Examining the transition rates from
individual trees will let you determine if they fit one or the other local
optima.
Tony
-
Tony Gamble, Ph.D.
Postdoctoral Researcher
Department of Genetics, Cell Biology and Development
University of Minnesota
6-160 Jackson Hall
321 Church St SE
Minneapolis MN 55455
USA
Phone: 612-626-4259
E-mail: gambl...@umn.edu
Web: www.tc.umn.edu/~gambl007
Twitter: @tony_gamble1
On Jan 27, 2015, at 9:26 AM, Liam J. Revell liam.rev...@umb.edu wrote:
Hi Maria.
To make sure this is not an error of the software, I highly recommend you
install the latest version of phytools (or at least the latest version on
CRAN). The current CRAN version is 0.4-31. I know in some earlier version
of phytools there was an issue with make.simmap if your data had more than
two states, but this should be fixed in the current CRAN version.
Having updated phytools (and if the result persists) you might also compare
your result to one obtained using Q=empirical (the default). If you do
this then the function will use the single most likelihood value for the
transition matrix. This could tell you if the result you have found is
being influenced by the prior (although the mean of the prior in
make.simmap is 'hyper-parameterized' by the ML value of Q, so this seems
somewhat unlikely).
All the best, Liam
Liam J. Revell, Assistant Professor of Biology
University of Massachusetts Boston
web: http://faculty.umb.edu/liam.revell/
email: liam.rev...@umb.edu
blog: http://blog.phytools.org
On 1/27/2015 9:28 AM, Maria Ghazali wrote:
Hi!
I am using phytools 0-3.10. I am trying to do ancestral reconstruction
with make.simmap for 3-state discrete character with Q=mcmc, nsim=1000,
and default priors (alpha=beta=1). In the result I have transition rates,
Q-matrix, for each of 1000 trees. The function proposes mean Q-matrix. But
I've looked at the histograms of the between-state transition rates (i.e.
state1 to state2, or state2 to state3, etc.). Some of the transitions have
bimodal distribution - with 2 maximums. Some are unimodal.
Is that wrong? Is that a sign of bad model?
Сould I use medians to present the results? To show which transitions are
more frequent.
Thank you,
Maria
___
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/
[[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/