Supervised learning of maternal cigarette-smoking signatures from placental gene expression data: A case study

Abstract

This paper aims to conduct supervised learning of the cigarette-smoking signatures from the placental gene expression data sets under the neural network framework and build classifiers to identify the cigarette-smoking moms during pregnancy. First, a unified model for gene selection is proposed to single out a set of informative gene sets (up-or down-regulated genes). The selected signature gene sets are subject to refinement, and then so refined informative gene sets are fed into three supervised statistical learning algorithms, linear discriminant function (LDF), probabilistic neural network (PNN) and support vector machine (SVM) for training and testing. It shows that SVM is the best classifier in predicting the cigarette-smoking moms compared to other methods tested.