Machine Learning Models Identify Inhibitors of SARS-CoV-2.

Area of Science:

• Virology
• Computational Chemistry
• Drug Discovery

Background:

• The emergence of SARS-CoV-2 variants necessitates new COVID-19 treatments.
• Drug repurposing offers a rapid strategy for identifying antiviral therapies.
• Existing research has generated databases of molecules with in vitro antiviral activity.

Purpose of the Study:

• To develop machine learning models for predicting SARS-CoV-2 antiviral compounds.
• To prioritize FDA-approved drugs from an in-house library for in vitro testing.
• To identify novel therapeutic candidates against SARS-CoV-2 and its variants.

Main Methods:

• Implemented multiple machine learning algorithms using SARS-CoV-2 in vitro inhibition data.
• Prioritized FDA-approved compounds based on model predictions.
• Selected lumefantrine for in vitro testing and characterized its binding to the spike protein.
• Utilized microscale thermophoresis to determine binding affinity (Kd).

Main Results:

• A Bayesian machine learning model predicted lumefantrine, an antimalarial, for testing.
• Lumefantrine exhibited limited antiviral activity in cell-based assays.
• Lumefantrine demonstrated binding to the SARS-CoV-2 spike protein with a Kd of 259 nM.
• Other prioritized compounds showed in vitro activity in subsequent studies.

Conclusions:

• The combined machine learning and in vitro testing approach effectively prioritizes potential SARS-CoV-2 antiviral drugs.
• This strategy can be expanded for virtual screening against current and future SARS-CoV-2 strains.
• Iterative machine learning models are valuable tools for accelerating antiviral drug discovery programs.
• The latest SARS-CoV-2 machine learning model is available at www.assaycentral.org.