Public source repository for FEAST.
- Precompiled binaries: feast-bin.zip
- Mac 1.05.
- Linux 64/32 1.05.
- Windows 32 1.06 [1].
- Parameter files: feast-parameters.zip
- Example input: feast-example-input.zip
[1] Version 1.06 contains a small fence-post bug fix. Mac and Linux binaries to be updated later.
- A recent version of g++.
- Cmake 2.6+
- Boost 1.34+
- Blitz++
Running the build.pl script will produce a static binary. The build is tested on Linux 64/32, Mac 10.6, and Windows 7.
To align two sequences you must supply a parameter file such as 2-R8. Input is expected to be in fasta format. Repeats should be lowercase masked. There can be any number of query sequences in the query file, but only the first sequence from the target file will be used. The output format is MAF. If no output file is specified, standard output is used.
$ feast -t target.fa -q query.mfa -o output.maf -p 2-R8
To train parameters, you may either use an existing parameter file, or you can supply an init file that initializes default parameters. See the supplied sample init file for more information on the format. The default training method is Baum-Welch. However, this method does not work for more than one sub-model at this time. Thus we recommend switching to Viterbi. In training mode the output is a parameter set. Status information is sent to standard error. If you break the procedure before it completes, the parameters for the last training iteration will be in in the output file. We do not support the standard output stream in this case, you must specify a file.
$ feast -t target.fa -q query.fa -o trained_parameters -i init_file --train --cut-threshold 50
IMPORTANT: feast assumes lowercase masked input files. Thus, input that is completely lowercase will return no results. Use the --ignore-mask flag to disable masking of repeats and treat all files as uppercase.
Please contact ([email protected]) if you have further questions.