|
Legofit
infers population history from nucleotide site patterns.
|
Bootstrap estimate of prediction error.
This program, like it's sibling clic, provides a tool for selecting among models that differ in complexity. It implements the "bootstrap estimate of predictive error", which is described in section 17.6 of An introduction to the bootstrap, (Efron and Tibshirani, 1993). This method provides a solution to the problem of overfitting.
Bepe does this by using bootstrap or simulation replicates as a proxy for samples from the underlying (and usually unknown) statistical distribution. Its value estimates the mean squared error between predicted site pattern frequencies and those of unobserved samples from the same statistical distribution. The best model is the one for which bepe reports the smallest value.
usage: bepe <realdat> <bdat1> <bdat2> ... -L <real.legofit> <b1.legofit> <b2.legofit> ... where realdat is the real data, each "bdat" file is the data for one bootstrap or simulation replicate, and each "b#.legofit" file is the legofit output from the corresponding bootstrap replicate Must include realdat file and at least 2 bootstrap replicates. Options: -h or --help : print this message
In typical usage, one would type something like
bepe realdat.txt boot*.txt -L real.legofit boot*.legofit
This usage assumes that your computer's shell or command interpreter sorts the files globbed by boot*.txt and boot*.legofit in a consistent order, so that the i'th .legofit file is the output produced from the i'th bootstrap data file.
The lines of data in bepe's output are arranged according to the order in which data files are listed on the command line. It is important that this order be consistent, if multiple .bepe files are to be compared using booma. (Otherwise, booma will abort with an error.) Consistency problems can arise because of differences in locale settings on different machines. The example command in the preceding paragraph may generate file names in different orders on different machines. To ensure a consistent order on Unix-like operating systems (linux and osx), set the LC_ALL environment parameter first. Using the bash shell:
export LC_ALL=C bepe realdat.txt boot*.txt -L real.legofit boot*.legofit
This ensures that file names will be listed in the same order, regardless of the locale setting of the local machine.
The first few lines of bepe's output look like this:
################################################ # bepe: bootstrap estimate of predictive error # # version 1.80 # ################################################ # Program was compiled: Feb 8 2019 15:26:39 # Program was run: Fri Feb 8 15:31:50 2019 # bepe DataFile LegofitFile 1.908861378e-07 sim0.opf a2-0.legofit 2.495970126e-07 sim1.opf a2-1.legofit 2.627267214e-07 sim10.opf a2-10.legofit 1.71297571e-07 sim11.opf a2-11.legofit 1.917256551e-07 sim12.opf a2-12.legofit
Many of the lines begin with the comment character (#) so that they can be removed easily in downstream processing. Uncommented lines refer to particular data sets, whose names are given in columns 2-3. The order of these lines reflects the order in which files are listed on the command line. If you are analyzing real data plus bootstrap replicates, it is useful to list the real data file first on the command line, so that its bebe value will appear first in the output. Column 1 (bepe) gives the mean squared difference between the site pattern frequencies in this data set, and the predictions obtained from all the other data sets. Bepe also includes a bias correction, which is described by Efron and Tibshirani. The predicted values are taken from the various .legofit files.
1.9.4