29 Colorectal Cancer Prognosis in Gene Expression Data

We have designed an automated process comprised of three basic components: coarse feature reduction and classification, and ∕ or fine feature measurement and selection, and then classification. Specifically, these three steps are: (1) coarse feature reduction using a combination of variance pruning and the Student's t test, where the resulting features are classified by a combination of logistics regression, the Wald test in combination with standard statistical testing using p values as a guide, (2) fine feature selection using the reduced feature set as input to a wrapper method consisting of a modified GA process, configured for this specific application, where the fitness function results are developed by support vector machines, and (3) the classification (and ∕ or diagnostic) process, which consists of training and validating specified support vector machines and a kernelized partial least squares (K-PLS) process as well as use of the evolutionary programming support vector machines (EP-SVM). This paper focuses only on the first component of course feature reduction process only using colorectal microarray gene expression data.