After cloning the git repository, it is necessary to run the following commands:

git submodule init git submodule update

This loads parts of this project that come from other git repositories.

An experiment is defined by a numerically named .bash file in REFINEMENT_SETTINGS, for example REFINEMENT_SETTINGS/1.bash

A refinement method is defined by an executable file in the METHODS subdirectory.

See METHODS/EXAMPLE for a bash script that parses the command-line parameters that are expected from a refinement method, you can put anything in METHODS and as long as it uses the correct command-line parameters and environment variables, producing the right output files, it should work.

The intention is that you will wrap calls to GEMSTAT with a bash script here.

A scoring method is defined by an executable in SCORING whose STDOUT will be just one floating-point score with no newline.

See SCORING/SSE for an example, there are several command-line parameters that a scoring method should be sensitive to. I should explain them

• --data [DIRECTORY] A data directory containing the training data as it was input when the method was trained
• --parfile [FILENAME] The parfile whence refinement began (to be symmetric to the refinement method command-line parameters)
• --parout [FILENAME] The parfile after refinement. (Could be used if you want to score something about the properties of the par files, such as distance from known true parameters...)
• --out [FILENAME] The output from cross-validation. Currently that output might contain the ground-truth, but ideally, you should get the ground-truth from the --data option, since in the future we might give no data at cross-validation time, to prevent methods from somehow cheating.

TODO: Mention the subdirectory structure for storing a dataset

A dataset directory must contain the following:

• template.par (optional, but used if you want to randomly generate starting points on-the-fly.)
• base/ (required by the system)
  • seqs.fa (assumed by gemstat)
  • whatever_else
• ORTHO/
  • whatever_your_training_ortholog_name_was
    • whatever files you want to have overwrite the base dataset
  • another_ortholog
    • whatever files are specific to this ortholog

Ensembles of starting-points are stored in the ENSEMBLES directory.

An ensemble definition make take one of three forms:

• A randomly generated ensemble from a template.par file in the dataset. This will be expected if your REFINEMMENT_SETTING file does not specify an "ENSEMBLE_NAME" environment variable.
• A fixed ensemble from an ASCII format table of values, substituted into a template.par file. If the file ENSEMBLES/${ENSEMBLE_NAME} is a regular file, that file is assumed to be a table of values which will be used this way.
• A directory of .par files. If the file ENSEMBLES/${ENSEMBLE_NAME} is a directory, it is assumed to contain files named 1.par 2.par ... etc.