Potentially highly potent drugs for 2019-nCoV

Duc Duy Nguyen; Kaifu Gao; Jiahui Chen; Rui Wang; Guo-Wei Wei

The current pneumonia outbreak caused by a new coronavirus (CoV), called 2019-nCoV in China, has evolved into a global health emergency declared by the World Health Organization. Although there is no effective anti-viral medicine for the 2019-nCoV, the 3CL proteases of 2019-nCoV and SARS-CoV have a sequence identity of 96%, which provides a foundation for us to hypothesize that all potential anti-SARS-CoV chemotherapies are also effective anti-2019-CoV molecules. We build a three-dimensional (3D) 2019-nCoV 3CL protease structure using a SARS-CoV 3CL protease crystal structure as a template and collect a set of 84 SARS-CoV inhibition experimental data. The molecules of this set are docked to the 3D 2019-nCoV 3CL protease structure to form a machine learning training set. Additionally, the PDBbind 2018 general set of 15,843 protein-ligand complexes is also included as an additional machine learning training set. Using these training sets, we develop two deep learning models based on low-dimensional algebraic topology representations of macromolecular complexes. A total of 1465 FDA-approved drugs is evaluated by their binding affinities predicted by the consensus of two models built with 1) a combination of algebraic topology and deep convolutional neural networks (CNNs), and 2) a combination of algebraic topology and deep multitask CNNs. According to the predicted binding affinities, we recommend many FDA-approved drugs as potentially highly potent medications to 2019-nCoV, which serve as a crucial step for the development of anti-2019-nCoV drugs.