NAME

       Bio::Tools::SVMLoc - Perl implementation of the SVM based protein subcellular localization
       method.

SYNOPSIS

         use Bio::Tools::SVMLoc;
         use Bio::SeqIO;

         # Load a previously trained model from a file.
         $svmloc = new Bio::Tools::SVMLoc(-model  => 'svmloc.model');

         # Classify a Bio::Seq object.
         $loc = $svmloc->classify($seq);

         # Train SVMLoc on a new dataset.
         $accuracy = $svmloc->train(\@vectors, $localization,
                                    { Kernel  => 'linear',
                                      Optimize => 1 });

         # Save the model to a file.
         $svmloc->save('svmloc.model.new');

         # Load a model from a file.
         $svmloc->load('svmloc.model.new');

DESCRIPTION

       Bio::Tools::SVMloc is an implementation of the SVMLoc protein subcellular localization
       method, which predicts localizations based on amino acid composition.  The method was
       originally outlined in "Support Vector Machine Approach for Protein Subcellular
       Localization Prediction" paper by Sujun Hua and Zhirong Sun.

CONSTRUCTOR

          $svmloc = new Bio::Tools::SVMLoc(-model  => 'svmloc.model');

       The SVMLoc constructor accepts the name of an existing model file.

METHODS

          $loc = $sl->classify($seq, \@frequent_patterns);

       The classify method accepts a Bio::Seq object and an array reference of the known frequent
       patterns as arguments and returns a string containing the predicted localization of the
       protein.  The return value will be one of: Cytoplasmic, CytoplasmicMembrane, Periplasmic,
       OuterMembrane, Secreted, Mitochondrial, Nuclear or Unknown.

         $accuracy = $svmloc->train(\@vectors, $localization,
                                    { Kernel  => 'linear',
                                      Optimize => 1 });

       The train method allows the creation of a new SVM model based on a set of user defined
       sequences.  The first parameter an array reference containing the list of vectors to train
       the SVM on, those in the positive set will be prefixed by 1 while those in the negative
       set will be prefixed by -1.  The second parameter is a string containing the localization
       that is being targeted with this SVM.

       The resulting model is unlikely to provide optimal results without some degree of fine
       tuning to the underlying Support Vector Machine.  The third (optional) parameter to the
       train method is a hashref which allows one to pass options directly to the SVM.  The set
       of valid keys are as follows:

         Kernel     - The Support Vector Machine kernel to use, possible values
                      are linear, polynomial, radial, and sigmoid.

         Optimize   - If true, the module will attempt to search for the best
                      values of the gamma and C parameters passed to the SVM.
                      For any new model, this should be run at least once.

                      By default, the SVMLoc module will try all values for gamma
                      between 1 and 100 (increments of 1) and C between 1 and 50
                     (increments of one) until the optimal values of each are
                      located.

                      Note that this **WILL** take quite some time, but only
                      needs to be done once per model.  The model should be
                      saved with the save method after training so that all
                      optimal parameter values will be retained.

         GammaStart - Allows specification of the start value when searching for
                      the optimal gamma value.  (Default 1)

         GammaEnd   - Allows specification of the end value when searching for
                      the optimal gamma value.  (Default 100)

         GammaInc   - Allows specification of the size of the increment when
                      searching for the optimal gamma value.  (Default 1)

         CStart      - Allows specification of the start value when searching for
                      the optimal C value.  (Default 1)

         CEnd        - Allows specification of the end value when searching for
                       the optimal C value.  (Default 50)

         CInc        - Allows specification of the size of the increment when
                       searching for the optimal C value.  (Default 1)

         $svmloc->save('svmloc.model.new');

       Saves the currently loaded model to the specified file.  Returns true on success, false on
       failure.

         $svmloc->load('svmloc.model.new');

       Loads an existing model from file.  Returns true on success, false on failure.

AUTHOR

       Fiona Brinkman, Cory Spencer, Brinkman Lab, Simon Fraser University <psort-mail@sfu.ca>

MAINTAINER

       Matthew Laird <lairdm@sfu.ca>

SEE ALSO

       Algorithm::SVM and the original SVMLoc paper titled "Support Vector Machine Approach for
       Protein Subcellular Localization Prediction" by Sujun Hua and Zhirong Sun

ACKNOWLEGEMENTS

       Chih-Jen Lin, one of the authors of the libsvm package upon which Algorithm::SVM was
       based, was invaluable in creating this version of SVMLoc Thanks also to Sujun Hua and
       Zhirong Sun, the authors of the original SVMLoc.  Last but not least, thanks to Fiona
       Brinkman, Jennifer Gardy and the other members of the Simon Fraser University Brinkman
       laboratory.