Using the output of your best model M1a or M2a you can check how many sites belong significantly to the first class of sites (w<1), this may be a solution.

However if you want an answer for the full gene, you may do a branch test with:
H0: wanted branched fixed (w=1)
H1: wanted branch free

If H1 has a better fit and an estimated omega <1 than dN/dS is significantly lower than 1.

note that you can either use one estimte of omega for all the background, let all background branches independent (perhaps a bad idea if you have big tree)

EDIT:
tipically at the end of your main output file you will find a summary of the proportion and estimated values of the class of sites (M1a model has 2 class of sites the first: omega<1, and the second: omega=1):

```
dN/dS (w) for site classes (K=2)
p: 0.94075 0.05925
w: 0.07025 1.00000
```

thus here we would have 94% of the sites in our alignment belonging to a class of site with omega=0.07
and 6% of the sites belonging to a second class of sites with omega =1

these values are means for the whole alignment. In order to know the values of significance to belonging to conserved class of site, for each site, you may look at an other output file, the rst.

sample of rst file:

```
dN/dS (w) for site classes (K=2)
p: 0.94075 0.05925
w: 0.07025 1.00000
Naive Empirical Bayes (NEB) probabilities for 2 classes & postmean_w
(amino acids refer to 1st sequence: Gorilla_gorilla)
1 M 0.97088 0.02912 ( 1) 0.097
2 A 0.95138 0.04862 ( 1) 0.115
3 R 0.97919 0.02081 ( 1) 0.090
4 Y 0.98049 0.01951 ( 1) 0.088
5 R 0.97381 0.02619 ( 1) 0.095
....
```

here for example first site has a probability of belonging to the first class of site of 97%, and its estimated value of omega is 0.097

this output is for model M1a, where only NEB probabilities are computed, in M2a model BEB probabilities are also computed and these values are though to be more accurate (see PAML doc.)

•

link
modified 7.7 years ago
•
written
7.7 years ago by
fransua • **390**