Machine intelligence design of 2019-nCoV drugs

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

Lopinavir is an antiretroviral medication used to inhibit HIV/AIDS viral protease. It is often used as a fixed-dose combination with another protease inhibitor, ritonavir, sold under the name Kaletra or Aluvia. Ritonavir, sold under the trade name Norvir, is another antiretroviral medication. Its combination with Lopinavir is known as highly active antiretroviral therapy (HAART). Although there is no tractable clinical evidence, Kaletra or Aluvia has been proposed as a potential anticoronavirus drug for 2019-nCoV. The possibility of repurposing some HIV drugs for SARS-CoV treatment has also studied in the literature. 16 It is important to evaluate their binding affinities, which are obtained with two ligand-based methods (i.e., LS-BP and 2DFP) and two 3D models (3DALL and 3DMT). To carry out 3D model predictions, we dock them to the 2019-nCoV protease inhibition site. The resulting complexes are optimized with molecular dynamics and then evaluated by 3DALL and 3DMT. Table 1 shows the low sequence identity between HIV viral protease and 2019-nCoV protease, which might suggest the limited potential for repurposing Aluvia and Norvir for 2019-nCoV treatment. For Lopinavir, our LS-BP and 2DFP predicted the binding affinities of -5.66 kcal/mol and -5.54 kcal/mol, respectively. For Ritonavir, similar low binding affinities of -5.14 kcal/mol and -4.96 kcal/mol were predicted by our LS-BP and 2DFP, respectively. However, our 3D model 3DALL predicted better binding affinities, i.e., -7.78 kcal/mol and -8.44 kcal/mol for Lopinavir and Ritonavir, respectively. The other 3D model, 3DMT, also predicted moderately high binding affinities of -8.13 kcal/mol and -8.07 kcal/mol for Lopinavir and Ritonavir, respectively. Considering the fact that the small training set for LS-BP and 2DFP models is very small, the results predicted by 3D models are more reliable. Figures 20 and 21 indicate that these drugs have reasonable dock poses with 2019-nCoV protease. Therefore, HIV drugs Kaletra (or Aluvia) and Norvir might indeed have a moderate effect in the treatment of 2019-nCoV. However, Many new compounds generated by our GNC appear to have better druggable properties than these HIV inhibitors do. epidemic. Although we know quite a little about 2019-nCoV, it is fortunate that the sequence identity of the 2019-nCoV protease and that of severe acute respiratory syndrome virus (SARS-CoV) is as high as 96.1%. In this work, we show that the protease inhibitor binding sites of 2019-nCoV and SARS-CoV are almost identical, which provides a foundation for us to hypothesize that all potential anti-SARS-CoV chemotherapies are also effective anti-2019-CoV molecules. Additionally, we employed a recently developed generative network complex (GNC) to seek potential protease inhibitors for effective treatment of pneumonia caused by 2019-nCoV. Two datasets 13 . CC-BY-NC 4.0 International license author/funder. It is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/2020.01.30.927889 doi: bioRxiv preprint are utilized in this work. One is a SARS-CoV protease inhibitor dataset, which is constructed by collecting 115 SRAS-CoV inhibitors from open database ChEMBL. The other dataset is a binding affinity training set mainly containing the PDBbind refined set. Our GNC model predicts over 8000 potential anti-2019-nCoV drugs which are evaluated by a latent space binding predictor (LS-BP) and a 2D fingerprint pre