Deep neural network aided multi-omics drug response prediction for breast cancer

Cancer is caused due to alterations in DNA sequences that act as a template in the production of proteins. Proteins control the structure and functions of cells, and hence these mutations lead to abnormal growth of the tissues, which can be fatal. The most common form of cancer in women is breast cancer, amongst whom 10 percent are caused due to germline mutations. People react differently when administered with the same treatment, influenced by factors like the difference in genetic interactions, exposure to a specific environment, and cancer stage. Precision medicine solves this problem by personalizing treatment, taking into consideration all such factors. In this paper, we have implemented a system that can personalize treatment using a deep neural network on genetically classified breast cancer and patient data. This neural network framework uses omics features extracted via an auto-encoder to extract an IC50 score for a cell line-drug pair. This IC50 score is then classified as responder or non-responder based on a threshold value obtained from K-means clustering of IC50 values. Our model performed with an accuracy of 0.80 and outperforms its predecessors.