[Obtaining ML tree from partitioned data in PAUP
1. Find appropriate type of substitution model for the data (e.g. using Modeltest).
2. Define character sets that you want to partition your data by (e.g. by gene or codon position).
3. Using the combined data, find a set of trees, hoping that the ML tree will be among them.
Length of search and number of trees in the set will depend on time constraints and size of data set (100 saved trees at minimum, more is better).
This could be done for example doing a heuristic search under parsimony (or using a credibility set of trees from Mr.Bayes etc).
4. Sort trees (e.g. by length for parsimony).
5. Switch to likelihood and for each partition separately
- estimate model parameters from the data (i.e. the trees),using the type of substitution model determined under 1.
- write resulting lscores to file ("scorefile= *.txt ")
6. Open scorefiles in Excel (or other spreadsheet program), perform 'text to columns', and cut and paste columns of likelihood scores from all partitions together.
7. For each tree, add up ML scores from different partitions and rank trees by their total score. Best tree will be the ML tree.
For example, the command block for an analysis partitioned by codon position could look like this:]
BEGIN PAUP;
charset 1st = 1-.\3;
charset 2nd = 2-.\3;
charset 3rd = 3-.\3;
Set criterion=parsimony maxtrees=100 increase=auto;
hs addseq=simple nreps=10000 nbest=100;
sorttrees;
savetrees brlens=yes file=100best.tre replace;
Set criterion=likelihood;
include 1st/only;
lscores 1-100/ nst=2 base=est tratio=est rates=gamma shape=est pinv=est
scorefile=1st.MLscores.txt replace;
include 2nd/only;
lscores 1-100/ nst=2 base=est tratio=est rates=gamma shape=est pinv=est
scorefile=2nd.MLscores.txt replace;
include 3rd/only;
lscores 1-100/ nst=2 base=est tratio=est rates=gamma shape=est pinv=est
scorefile=3rd.MLscores.txt replace;
End;