Construction of a deep learning model for predicting cancer prognosis using multi-omics data

Abstract

Cancer is one of the significant causes of death globally. Therefore, early and precise detection and proper treatment decision of cancer can considerably reduce the risk of the disease in clinic practice. As many genomic data getting digitalized and become more easily accessible, deep learning is developed into a useful tool for predictive and prognostic models. Through this research, multi-omics data of liver cancer patients including seven molecular features such as expression data is systematically collected and organized as a database. Using this dataset, the prognostic model with convolutional neural network (CNN) is constructed and trained. Compared to previous prognostic models, such as linear cox, random survival forest and glmnet, the CNN prognostic model shows outstanding and robust prediction accuracy. Multi-omics data can preserve various causes of malfunction in genes. By computing feature importance of each gene’s molecular features, certain change in certain gene that affects patients’ prognosis can be discovered. These prognostic markers help to propose appropriate approach of treatment to cancer patients.