Analysis of Microarray Gene Expression Data
                    Using Support Vector Machines
        
                        William Grundy
                 Department of Computer Science
                      Columbia University

                      February 7, 2000
  
We have developed a method of functionally classifying genes by using
gene expression data from DNA microarray hybridization experiments.
We train a support vector machine (SVM) to distinguish expression
patterns from genes in one class of co-regulated genes from expression
patterns of genes in other classes.  We have applied the method with
some success to expression data collected for yeast genes by the Pat
Brown laboratory.  After training on expression patterns for two
thirds of the ribosomal yeast genes, the method identifies the
remaining known ribosomal genes from among the approximately 6000
yeast genes with few errors, and identifies one incorrect annotation
in the MYGD database.  Experiments with other classes of co-regulated
genes also give useful results, but accuracy depends on how strongly
the class of co-regulated genes exhibits a common and distinctive
expression pattern in the set of DNA microarray hybridization
experiments that is performed.