A New Approach to Gene Prediction Using the Self-Organizing Map

In this poster we present a gene prediction approach based on the Self-Organizing Map that has the ability to automatically identify all the major patterns of content variation within a genome. The genome may then be scanned for regions displaying the same properties as one of these automatically identified models. Even using a relatively simple coding measure (codon usage), this method can predict the location of protein-coding sequences with a reasonably high accuracy. We also show other advantages of the approach, such as the ability to indicate genes that contain frame-shifts. We believe that this method has the potential to become a useful addition to the genome annotation process.