In this work, we propose a deep learning approach to improve docking-basedvirtual screening. The introduced deep neural network, DeepVS, uses the outputof a docking program and learns how to extract relevant features from basicdata such as atom and residues types obtained from protein-ligand complexes.Our approach introduces the use of atom and amino acid embeddings andimplements an effective way of creating distributed vector representations ofprotein-ligand complexes by modeling the compound as a set of atom contextsthat is further processed by a convolutional layer. One of the main advantagesof the proposed method is that it does not require feature engineering. Weevaluate DeepVS on the Directory of Useful Decoys (DUD), using the output oftwo docking programs: AutodockVina1.1.2 and Dock6.6. Using a strict evaluationwith leave-one-out cross-validation, DeepVS outperforms the docking programs inboth AUC ROC and enrichment factor. Moreover, using the output ofAutodockVina1.1.2, DeepVS achieves an AUC ROC of 0.81, which, to the best ofour knowledge, is the best AUC reported so far for virtual screening using the40 receptors from DUD.
展开▼
