Accelerating Drug Discovery Using Convolution Neural Network Based Active Learning

Abstract

Drug discovery is an expensive and time consuming process, especially in the era of new technology, such as personalized medicine where tremendous experiments and analysis are needed before bringing new drugs to the market. While In vivo and In vitro experiments are expensive, In silico methods become important and they can reduce the cost in drug discovery by prioritizing the experiments in more efficient ways. In this paper, we propose a new convolution neural network based active learning model which helps to reduce the number of experiments needed in drug discovery. Using the drugs performance on other cell lines as assisting information, our model can precisely select the most promising drug from those candidates for a new cell line. Our model uses a deep neural network structure where there are two CNN channels for drugs and cell lines respectively, which are followed by a fulled connected network. The experimental results show that our model can achieve significant better performance than the existing methods.