Evolutionary Algorithm for Feature Subset Selection in Predicting Tumor Outcomes Using Microarray Data

Feature subset selection for outcome prediction is a critical issue in large scale microarray experiments in cancer research. This paper introduces an integrative approach that combines significant gene expression analysis, the genetic algorithm and machine learning for selecting informative gene markers and for predicting tumor outcomes including survival outcomes. In case of survival data, full use of individual’s survival information (both censored and uncensored) is made in selecting informative genes for survival outcome prediction. Applications of our method to published microarray data on epithelial ovarian cancer survival and breast cancer metastasis have identified prognostic genes that predict individual survival and metastatic outcomes with improved power while basing on considerably shorter gene lists.