Farrell Lab @ Harvard University

Determining error bars on ages of nodes

For determining error bars on nodes, two methods can be used, both implemented in Sanderson's r8s program. The first method involves looking at the shape of the likelihood surface. The second method uses bootstrapped datasets and calculates error bars from the distribution of branchlengths on the bootstrapped dataset. We have done the latter method using PAUP and Excel, due to a lack of Linux x86 computers to run r8s. First, Phylip's SeqBoot is used to create bootstrapped datasets. These are edited in PAUP to merge them into one large dataset (for example, if the original dataset is 3000 characters long, and 500 bootstrap replicate datasets are created, they are edited to make a single 1,500,000 character dataset. A PAUP batch file is then constructed in Excel which divides the combined datasets into character sets and then has PAUP evaluate the branchlengths of the given topology (the clock tree being evaluated) for each replicate dataset (corresponding to a charset). A sample batch file is below for a dataset 2432 characters long. Some of the repetitive commands have been deleted, but the pattern should be clear. Before executing this batch file, load your long datafile, set the likelihood parameters, and load your clock tree.

#nexus
begin paup;
charset rep1 = 1 - 2432 ;
charset rep2 = 2433 - 4864 ;
charset rep3 = 4865 - 7296 ;
[.... repetitive commands not included here. Just use Excel to follow the pattern....]
charset rep500 = 1213569 - 1216000 ;
exclude all;
include rep1 ; [! rep 1 ]
savetrees root=yes brlen=yes maxdecimals=6 replace=no append=yes file=brlenwbootstrapdata.tre;
exclude all;
include rep2 ; [! rep 2 ]
savetrees root=yes brlen=yes maxdecimals=6 replace=no append=yes file=brlenwbootstrapdata.tre;
exclude all;
include rep3 ; [! rep 3 ]
savetrees root=yes brlen=yes maxdecimals=6 replace=no append=yes file=brlenwbootstrapdata.tre;
[.... repetitive commands not included here. Just use Excel to follow the pattern....]
exclude all;
include rep500 ; [! rep 500 ]
savetrees root=yes brlen=yes maxdecimals=6 replace=no append=yes file=brlenwbootstrapdata.tre;
end;

The trees are all saved in one file with branchlengths by PAUP. This tree file is then opened within Excel, with the length of each branch put in a column. The total length of branches from the root to the node of interest is computed for each tree and divided by the distance from the root to the tips. Then the percentile function is used in Microsoft Excel to determine the 95% confidence interval on the distance from the root to the node of interest, with the distance expressed as proportion of total tree length.