A Deep-Learning-Aided Drug Screening Based on Visualization of a Hidden Layer as Chemical Space

Drug discovery using deep learning has attracted much attention. However, deep learning models remain unpolished and do not always provide drug leads. Herein, we propose a drug screening method based on the visualization of a hidden layer in the output process of a graph-convolutional-network-based deep learning. Unlike conventional deep learning methods, our approach allows us to prioritize compounds for experimental testing from numerous candidate compounds predicted to be active by the deep learning model. Additionally, it provides information on the relationships between compound structures and their activity. Using this approach, we efficiently identified new leads for histone deacetylase inhibitors. These findings support the usefulness of our deep-learning-aided screening method.