Deep Learning for Ligand-Based Virtual Screening in Drug Discovery

Due to the time and cost problems with traditional drug discovery, new methods must be found to increase the declining efficiency of traditional approaches. Virtual Screening (VS) is one possible solution to solve this problem. VS of databases has become an attractive method for pharmaceutical research. It plays a crucial role in the early stage of the drug discovery and development process. It aims to reduce the enormous search space of chemical compounds. As the number of ligands in the databases is increasing rapidly, this step should be both fast and effective in order to distinguish between active and inactive ligands. Deep learning algorithms can be used for screening big databases of molecules and classifying the ligands as drug-like and non-drug-like against a particular protein target and therefore speed up the VS process. In this paper, we propose a fast compound classification method based on a deep neural network for Virtual Screening called (DNN-VS) using the Spark-H2O platform in order to label small molecules from huge databases. Experimental results have shown that the proposed approach outperforms state-of-the-art machine learning techniques with an overall accuracy more than 99%.