Gene Shaving is a method for clustering groups of similarly behaving genes whose changes in expression are most tightly linked to observed biological changes. The basic method is similar to observed principal components (singular value decomposition, maximum eigenvalue, etc.) with a sequential twist: a canonical “gene vector” is identified based on the eigenvectors, and the genes are ranked according to their agreement with this vector. The worst fitting are then “shaved off” and a new canonical vector is identified and fit.
The GeneClust distribution consists of a Java front-end for an S-plus back-end which calls various C functions implementing the statistical method. A pseudoterminal application qua agent accepts commands from the Java front-end and returns output from S-plus.
The Java application can process raw data or generate its own. The information input by the user is checked for validity. Invalid input will cause the offending field to be displayed in reverse video. ToolTips with numeric range information are provided.