Classification of tumor samples is a promising application of gene expression data from microarray experiments. Classification includes the identification of new classes within a population (class discovery) and the assignment of new cases to known classes (class prediction). Most studies involving microarrays have focussed on class discovery, though methods for class prediction are now gaining popularity. We present here a method for class prediction based on gene expression data and show results obtained by this method for various tumor class prediction problems. Validation of classification results are discussed as well. Finally, in an attempt to aid researchers in microarray study design, we examine the effect of the size of the training set (i.e., the number of microarray experiments used to build the predictor) on the misclassification error rate.
Copyright. All Rights Reserved.