Artificial intelligence approach fighting COVID-19 with repurposing drugs

Abstract

Background: The ongoing COVID-19 pandemic has caused more than 193,825 deaths during the past few months. A quick-to-be-identified cure for the disease will be a therapeutic medicine that has prior use experiences in patients in order to resolve the current pandemic situation before it could become worsening. Artificial intelligence (AI) technology is hereby applied to identify the marketed drugs with potential for treating COVID-19.

Methods: An AI platform was established to identify potential old drugs with anti-coronavirus activities by using two different learning databases; one consisted of the compounds reported or proven active against SARS-CoV, SARS-CoV-2, human immunodeficiency virus, influenza virus, and the other one containing the known 3C-like protease inhibitors. All AI predicted drugs were then tested for activities against a feline coronavirus in in vitro cell-based assay. These assay results were feedbacks to the AI system for relearning and thus to generate a modified AI model to search for old drugs again.

Results: After a few runs of AI learning and prediction processes, the AI system identified 80 marketed drugs with potential. Among them, 8 drugs (bedaquiline, brequinar, celecoxib, clofazimine, conivaptan, gemcitabine, tolcapone, and vismodegib) showed in vitro activities against the proliferation of a feline infectious peritonitis (FIP) virus in Fcwf-4 cells. In addition, 5 other drugs (boceprevir, chloroquine, homoharringtonine, tilorone, and salinomycin) were also found active during the exercises of AI approaches.

Conclusion: Having taken advantages of AI, we identified old drugs with activities against FIP coronavirus. Further studies are underway to demonstrate their activities against SARS-CoV-2 in vitro and in vivo at clinically achievable concentrations and doses. With prior use experiences in patients, these old drugs if proven active against SARS-CoV-2 can readily be applied for fighting COVID-19 pandemic.

Keywords: AI; COVID-19; DNN; Drug repurposing; Feline coronavirus; SARS-CoV-2.

Fig. 1

Flow scheme of AI approaches along with antiviral activity verification assays. Two independent datasets were compiled as the learning inputs to generate two AI prediction models of different approaches. Firstly, Model 1 was generated form the known drugs with antiviral activities. Model 2 was generated form the 3C-like protease inhibitors. A database for the market-approved drugs were adopted in which the AI system screens for potential drugs with antiviral activities. The AI-predicted drugs were verified with antiviral activities by a cell-based FIP virus replication assay. Finally, these assay results served as feedbacks for the AI relearning and evolution progress. The Modified AI model was established to screen further and again verified by the FIP virus replication assay. The processing flows are indicated with the arrows.

Fig. 2

Fcwf-4 cells infected with FIP virus showed cytopathic effects, crystal violet staining. FIP virus (NTU156) infected cells treated with brequinar of various concentrations (0.4, 2, 10, 50 μM) in duplicate wells (Row A and B) and cytotoxicity of brequinar at 0.4, 2, 10, 50 μM was also investigated (Row C). FIPV-infected cells treated with a positive control, anti-FIP virus compound GC-376 (Row D) and vehicle only (Row E). Uninfected cells without drugs in parallel treatments were shown (Row F).