Understanding PartNH results

[nvgiap]

Dear Dr Julien Yann Dutheil


1. I get confusing in choosing what is the best partion from the following information. Could you please help?

 Threshold    NbPartitions        LogL        DF    AIC          BIC
0                             1            -44887.8    42    89859.6    90083.2
0                             2            -44837.5    43    89760.9    89989.9
3.374e-05                 3            -44837.3    44    89762.7    89997
0.0151876                4            -44832.7    45    89755.3    89995
0.0664231                5            -44832.2    46    89756.3    90001.3
0.29818                   6            -44826.2    47    89746.4    89996.7
0.371422                 7            -44825.5    48    89747       90002.6
0.516112                 8            -44823        49    89744.1    90005
0.657527                 9            -44822.9    50    89745.8    90012
0.75047                 10            -44819.7    51    89741.4    90012.9
0.76034                 11            -44817.3    52    89738.5    90015.4
0.761954               12            -44817.1    53    89740.1    90022.4

2.1. As I understand, performing robustness testing of substitution mapping is optional, am I right?
2.2. Also, I do not know how to interpret the result from robustness testing and how to compare the results between homogeneous model and non-homogeneous model.

Thank you very much!
Best regards
Giap Nguyen

[Julien Yann Dutheil]

Dear Nguyen,

> 1. I get confusing in choosing what is the best partion from the following
> information. Could you please help?
>
> Threshold NbPartitions LogL DF AIC BIC
> 0 1 -44887.8 42 89859.6
> 90083.2
> 0 2 -44837.5 43 89760.9
> 89989.9
> 3.374e-05 3 -44837.3 44 89762.7 89997
> 0.0151876 4 -44832.7 45 89755.3 89995
> 0.0664231 5 -44832.2 46 89756.3 90001.3
> 0.29818 6 -44826.2 47 89746.4 89996.7
> 0.371422 7 -44825.5 48 89747 90002.6
> 0.516112 8 -44823 49 89744.1 90005
> 0.657527 9 -44822.9 50 89745.8 90012
> 0.75047 10 -44819.7 51 89741.4 90012.9
> 0.76034 11 -44817.3 52 89738.5 90015.4
> 0.761954 12 -44817.1 53 89740.1 90022.4

You can select your models based on AIC or BIC. BIC suggests a 2
partitions model (lowest value for n=2), whereas AIC proposes n=11. I
recommend BIC over AIC (actually, we should also implement cAIC, as it
is known to improve over AIC...).
ParNH also output the best model in separate files, according to the
options you have specified in the option file.

> 2.1. As I understand, performing robustness testing of substitution mapping
> is optional, am I right?

Indeed.

> 2.2. Also, I do not know how to interpret the result from robustness testing
> and how to compare the results between homogeneous model and non-homogeneous
> model.

For the first point, the robustness test is simply to perform a new
substitution mapping with the selected NH model, and rerun PartNH on
this new mapping. Ideally, if substitution mapping is completely
robust, then the new partitions should be identical to the previous
ones. In the MBE paper, we report that it is indeed the case to a
large extent.

For the second part of your question, I need more information... as I
do not know the question you are addressing, I do not know which
results you would like to compare :s

All the best,

Julien.

>
> Thank you very much!
> Best regards
> Giap Nguyen
>

--
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