Feedforward Neural Network in Cancer Treatment Response Prediction

Abstract

AbstractRecently, research efforts are exerted on cancer treatment prediction based on the biomarkers related to the tumor. In order to save time and human effort, computational power is used to analyze huge sequences such as human DNA (DeoxyriboNucleic Acid). The use of machine learning and deep learning algorithms becomes a must. Using machine learning (ML) and deep learning (DL) algorithms to predict cancer treatment or drug response is considered a recent approach. There is no specific approach that proves its efficiency against all other approaches, but the indicators show the performance of deep learning-based approaches overcomes others. In this paper, nine different feedforward network architectures are introduced to predict drug responses. The proposed architectures are different in the number of layers and the number of nodes in each layer. Principal Component Analysis (PCA) is used as a dimension reduction method with different degrees of reduction to find the best degree of reduction. The proposed architectures are used with 19702 genes as input to predict the response to 265 different anti-cancer drugs. The proposed Feedforward architectures achieve evolution accuracy over other feedforward model architectures. The enhancement is between 45% and 52% reduction in the Mean Squared Error (MSE).KeywordsArtificial intelligenceArtificial neural networksBiomedicalFeedforward neural networksPersonalized medicineDrug response prediction