Open crsl4 opened 2 years ago
what happens if you re-build the data frame dat
from scratch, before running the second test (with the SNaQ tree)? ticr!
modified the data frame, so it would be best to use the same starting point for dat
in both cases. I doubt that it will change anything, but who knows.
Did you make sure that both trees don't have any polytomy (with branch lengths of 0 if need be)?
To help the diagnostic, it would help to compare the column created by the test: it contains the outlier p-values corresponding to each four-taxon set. The largest differences between the astral tree and snaq tree could help find the tree differences that might contributed to the outlier p-value differences.
I get the same results if I re-read dat
before running the test for the SNaQ tree. I also checked that there are no polytomies on the trees. I will check next the columns created by the test.
Also: did you check the results using the R implementation in phylolm? Do the result vary a lot from one tree to the other using the R implementation? The polytomies are treated a little differently between the 2 implementations, and you would need to choose the test statistic in julia that is implemented in R.
I am adding commands to run TICR in julia in the PhyloNetworks wiki. I get two very different p-values for the ASTRAL tree and the SNaQ h=0 tree, but the tree topologies are the same and the difference in branch lengths (at least visually on the plot) look very similar.
I followed this code inside the
data_results/n15.gamma0.30.20.2_n300
folder from the PhyloNetworks wiki:P-value for ASTRAL tree:
Then, same code for SNaQ h=0 tree:
P-value for SNaQ h=0 tree:
Even if I optimize branch lengths on the SNaQ h=0 tree: