Deep Learning Models for Cancer Classification from Microarray Gene Expression Profiles

Abstract

Gene expression profiles measured by microarray technology enables accurate identification of disease genes, prediction of cancers, and distinguishing tumor subtypes at the molecular level. However, these profiles are characterized by a small sample size and high dimensionality, which would inevitably degrade the performance of analysis models. In this study, we proposed a deep learning-based model to improve the prediction accuracy. Specifically, we first use the minimum redundancy maximum relevancy feature selector to discard irrelevant and noisy features. Then, we utilize a deep autoencoder to learn complex and nonlinear relationships among data. Finally, a predictor is trained on the latent representation to classify cancer. We conduct extensive experiments on four publicly available microarray datasets and compare the proposed model with six commonly used feature selectors using naïve bayes and decision tree in terms of accuracy and F1. Results demonstrate the superiority of the proposed model over its competitors.