Deep Learning of Gene Expression Data for Breast Cancer Classification

Abstract

Currently, breast cancer is a type of cancer where most of the cancer patients suffer. Identifying the type of breast cancer is an essential step to be done in drug discovery for breast cancers. Thus, the treatments could address the key features of cancer and successfully the breast cancer can be cured.Majority of the related studies on cancer classifications are based on clinical diagnosis, hence affected with restricted classifications. Thus, gene expression data which has obtained via transcription profiling on microarrays have been used as the input for the classification of cancer in this study. The key goal of this study was to research the existing methods of breast cancer classifications and implement an efficient breast cancer classification method based on deep learning using gene expression data which overcomes the defects of existing methodologies of breast cancer classification. Two different deep learning architectures were implemented by this study which is Convolutional Neural Network(CNN) and Deep Belief Network(DBN) using Tensorflow framework to classify breast cancers under the classification based on gene and protein status. Finally, we compared the performance of those two architectures with the deep learning architecture, autoencoder which was implemented before in another study in classifying breast cancer using gene expression data. The two proposed architectures perform better than autoencoder with respect to precision, recall, F1 score and accuracy. In conclusion, CNN is the best supervised deep learning architecture which yielded an accuracy of 63.4395% and DBN is the best unsupervised deep learning architecture which yielded an accuracy of 63.3545%in classifying breast cancers using gene expression data based on gene and protein status.