|
Legofit
infers population history from nucleotide site patterns.
|
Re-express free variables in terms of principal components.
First run legofit on a series of bootstrap or simulation replicates to produce a series of .legofiles. All these legofit runs should use the same .lgo file. pclgo reads this .lgo file and also all the .legofit files. It uses the .legofit files to do a principal components analysis of the free variables, and then rewrites a portion of the .lgo file, re-expressing all the free variables as linear functions of the principle components. In the rewrite, all the free variables are principal component scores and are therefore uncorrelated. The output of pclgo is not a complete .lgo file. It must be integrated into a new .lgo file.
You can do this with a text editor. But here is a trick that uses the unix shell:
(grep ^# a.lgo; pclgo a.lgo a2.legofit a2boot*.legofit;\ grep -v ^# a.lgo | egrep -v "\<free\>") > b.lgo
This assumes that you have used a.lgo to analyze a real data set and several bootstrap replicates. The legofit output files are named a2.legofit (for the real data), and a2boot*.legofit (for the bootstrap replicates). The parentheses group the pipeline within it so that we can redirect the output with a single >. Within the parentheses, the first command pulls the comments out of a.lgo and writes then to standard output. They will end up at the top of b.lgo. Next, the pclgo command writes lines in .lgo format, which define variables called "pc1", "pc2", etc., which refer to principal components. Following these, pclgo writes lines that re-express all the free variables of a.lgo as linear functions of the principal components. Following the pclgo command is a grep -v command that excludes comments. Its output is piped to an egrep command that excludes the free variables of a.lgo. All of this gets written into b.lgo. The result is a .lgo file that re-expresses all the free variables of a.lgo as functions of principal components.
By default, pclgo uses all principal components and therefore does not reduce the dimension of the search space. To reduce dimension, try pclgo --tol 0.001. This would exclude principal components that account for less than a fraction 0.001 of the variance.
1.9.4