Gist contains software tools for support vector machine classification and for kernel principal components analysis. The SVM portion of Gist is available via an interactive web server at http://svm.sdsc.edu. The Gist package contains the following programs:

• gist-train-svm trains a support vector machine based upon a given set of labeled training examples, permitting feature selection and leave-one-out cross-validation,
• gist-classify applies a trained support vector machine to unlabeled data to produce predicted binary classifications.
• gist-kpca performs kernel principal components analysis on a given data set, and
• gist-project projects a data set onto the components discovered by gist-kpca.

Note that with version 2.1, the names of the programs in Gist have changed. If you need to use executables with the old names, version 2.0.9 is also available.

If you have problems with or questions about Gist, please first read the FAQ page.

In addition to the primary programs, the following auxiliary programs are included:

• gist-fselect performs linear feature selection on a given data set, using binary classification labels,
• gist-matrix performs basic manipulations of matrices,
• gist-score-svm computes performance statistics from the outputs of gist-train-svm and gist-classify,
• gist-rfe performs SVM recursive feature elimination on a given data set,
• gist-sigmoid converts the discriminant values produced by gist-train-svm into probabilities, and
• gist2html converts an output file from one of the Gist programs into HTML format.

The current version of Gist is 2.1

Gist is written in ANSI C. Source code, as well as pre-compiled versions for several popular platforms (Linux, Solaris, Cygwin, Mac OSX) can be downloaded here. Here are some installation instructions and release notes. Once you have installed the software, you can try this example.

A list of publications that used Gist is available.

Please read the Copyright notice/License agreement.

The Gist software was written by William Stafford Noble in the Department of Computer Science at Columbia University and by Paul Pavlidis in the Columbia Genome Center.