Improving Reliability of Response Prediction to Platinum-Based Therapy by AdaBoost and Multiple Classifiers

It is a challenge to construct a reliable classifier based on microarray gene expression data for prediction of chemotherapy response, because usually only a small number of samples are available and each sample has thousands of gene expressions. This paper uses boosting and bootstrap approaches to improve the reliability of prediction. Specifically, AdaBoost and multiple classifiers based methods are used, in which support vector machines (SVMs) are utilized as the classifiers due to their good generalization ability. We compare the performance of proposed methods with a single SVM classifier system using MAS gene expression dataset in prediction of the response to platinum-based therapy for advanced-stage ovarian cancers. Statistical tests show both of the proposed methods achieve better prediction performance and have good reliability in terms of mean and standard deviation of the prediction performance for different number of selected features.