non stationarity for codon models

[Laurent Guéguen]

Hello,

using mapnh,

the option "stationarity" in line:

map.type = DnDs(stationarity=no)

does not have any importance, since mapping is apparently done with
"stationarity=yes" in this case.
But if the model is not stationary, the root freq given in the .bpp
file has an influence on the result.

How much is is false to use non stationary model while mapping DnDs,
since I suppose that "stationarity" option in map.type has an
influence somewhere?

Why is the "stationary" option inactivated for DnDs mapping?

Thanks,
Laurent

[Julien Yann Dutheil]

Hi Laurent,

This options performs an "ad hoc" correction for non-stationarity. So
far we only used it with GC as it was the only non-stationary model.
It does not influence the mapping itself, but rather adds sthg on top
of it (namely, it devides the counts by the ancestral frequencies of
states, as I remember). This turned to be important for the clustering
procedure and then the model selection. In theory it should work for
DnDs models, if some non-stationary ones become implemented.

Julien.

--
Julien Y. Dutheil, Ph-D
0 (+49) 6421 178 530

§ Max Planck Institute for Terrestrial Microbiology
Department of Organismic Interactions
Marburg -- GERMANY

§ Intitute of Evolutionary Sciences - Montpellier
University of Montpellier 2 -- FRANCE

[Laurent Guéguen]

Ok, so I have to program it for codon models if I look for
clustering,
if I get it well?

By the way, I test partNH, but the "Ancient" ignored parameter is
refused,
why is it so?

Thanks,
Laurent

[Laurent Guéguen]

Hi again,

by the way, how can DnDsSR be a categorySR ? Because
if I understand well categorySR are to counts substitutions towards
specific (categorized) states (such as in GC). But in the case of
DnDs,
this does not apply.

L

[Julien Yann Dutheil]

Indeed, but where do you see that DnDs is a CategorySR? As far as I
know it is not :s

J.

[Laurent Guéguen]

Indeed it is not, but you wrote just before that taking input account
non-stationarity
was important for clustering procedure, but it is handled only with
CategorySR.
So I wonder how it can be done with DnDs since it is not a categorySR.

L

[Julien Yann Dutheil]

Ah, sorry, I did not get that...
I am not sure about which type of non-stationarity you are thinking of
there, I guess it really depends on your model. Which type of
substitution are you counting? something like ATsyn=>GCsyn,
ATnon=>GCnon and so on? If you can design a SR that falls back to a
CategorySR, then it makes things easier. If not, we have indeed to
think of a way to generalize the stationarity correction for
non-CategorySR, where it is applicable...

Cheers,

J.

[Laurent Guéguen]

Even in this case it is not so easy to make categories,
because a codon is not only more of GCs or ATs.
For example, in which category is TTG, where both CTG
and TTA code also for leucine? Ok, leucine is the only example,
and a good proxy is to consider only the third position in the
substitutions.
This is what I will do.

But my concern is a more general one, to better understand how
you handle non-stationarity in your process. How much is clustering
accurate using DnDs, if the process is non-stationary, without the
"stationarity=no" option?

Cheers,
L