Identification of African Swine Fever Virus Inhibitors through High Performance Virtual Screening Using Machine Learning

Abstract

African swine fever virus (ASFV) is a highly contagious virus that causes severe hemorrhagic viral disease resulting in high mortality in domestic and wild pigs, until few antiviral agents can inhibit ASFV infections. Thus, new anti-ASFV drugs need to be urgently identified. Recently, we identified pentagastrin as a potential antiviral drug against ASFVs using molecular docking and machine learning models. However, the scoring functions are easily influenced by properties of protein pockets, resulting in a scoring bias. Here, we employed the 5′-P binding pocket of AsfvPolX as a potential binding site to identify antiviral drugs and classified 13 AsfvPolX structures into three classes based on pocket parameters calculated by the SiteMap module. We then applied principal component analysis to eliminate this scoring bias, which was effective in making the SP Glide score more balanced between 13 AsfvPolX structures in the dataset. As a result, we identified cangrelor and fostamatinib as potential antiviral drugs against ASFVs. Furthermore, the classification of the pocket properties of AsfvPolX protein can provide an alternative approach to identify novel antiviral drugs by optimizing the scoring function of the docking programs. Here, we report a machine learning-based novel approach to generate high binding affinity compounds that are individually matched to the available classification of the pocket properties of AsfvPolX protein.

Keywords

• African swine fever virus
• cangrelor
• fostamatinib
• molecular docking
• <i>Asfv</i>PolX
• machine learning