Improving Prediction Accuracy of Tumor Classification by Re-using the Discarded Genes during Gene Selection

Background: Since the high dimensionality of gene expression microarray data set hurts generalization performance of classifiers, feature selection has been widely used in the bioinformatics field, which selects relevant features and discards irrelevant and redundant features. While redundant features contain useful information, so multi-task learning is a novel technique to improve prediction accuracy of tumor classification by using these discarded redundant features, but which features should be discarded or used as input, output? The previous works used heuristic methods to search features, the number of features as input, output and discarded are arbitrarily determined. Results: We demonstrate a framework for automatically selecting features as input, output and discarded by using genetic algorithm. GA-MTL (Genetic algorithm based multi-task leaning) and e-GA-MTL (Enhanced version of GA-MTL) are proposed. Experimental results illustrate that our proposed framework is effective to select features for multi-task learning, GA-MTL and e-GA-MTL perform better than other heuristic methods. Conclusions: Genetic algorithm is an powerful technique to select features for multi-task learning automatically, the proposed GA-MTL and e-GA-MTL are two alternative algorithms to improve generalization performance of classifiers for analysis of microarray data sets.