Deep Learning for Accurate Cancer Type Classification Using Gene Expression Data

An accurate diagnosis and personalized treatment rely on the type of cancer classification through gene expression data. Here, we address this issue by proposing a deep learning framework based on Convolutional Neural Networks that classifies the same type of cancers with high accuracy from a gene expression profile. We compare our result with traditional methods including SVM and Decision Trees models, in which it automatically extracts features and leads to better classification performance. We tested this model on the problem with a simulated gene expression dataset that captures some biological variability as well as with a few real-world datasets. For real-world datasets, we obtained a test set accuracy of 80% for the proposed model. We tested proposed classification models along with traditional classifiers to show improvements of great robustness in classification and improved accuracy. This framework, once further validated, could possibly enter the clinical workflows and provide more accurate and timely cancer diagnoses to make it useful for real-world applications.