Machine intelligence design of 2019-nCoV drugs

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

ChEMBL, 29 an open database which brings chemical, bioactivity, and genomic data together to translate genomic information into effective new drugs is employed to construct our 2019-nCoV training set. Considering the high sequence identity between viral proteases of 2019-nCoV and SARS-CoV, we take the protease of SARS-CoV as the input target in ChEMBL and a total 115 ChEMBL IDs of the target can be found. Therefore, our 2019-nCoV training set is built up by 115 SARS-CoV protease inhibitors. Figure 4 describes the distribution of experimental ∆G values of 2019-nCoV training set. It can be seen that the experimental ∆G ranges from −10.0 kcal/mol to 7.5 kcal/mol, and most training samples have the experimental ∆G located in the range of [−10, −5] kcal/mol. Followed by the second law of thermodynamics, the more negative ∆G can result in a more spontaneous binding process. Figure 3 depicts the top five anti-SARS CoV compounds and their banding affinities. 29 In this work, 115 SARS-nCoV protease inhibitors from ChEMBL are used as the 2019-nCoV training set. A collection of these 115 compounds is given in the Supplementary material.