Ebola Virus Bayesian Machine Learning Models Enable New in Vitro Leads.

Area of Science:

• Computational chemistry and virology
• Drug discovery and development

Background:

• Previous work established Bayesian machine learning models for identifying Ebola virus (EBOV)-active compounds from FDA-approved drug screens.
• These models identified tilorone, pyronaridine, and quinacrine, with tilorone showing 100% in vivo efficacy in mouse models.

Purpose of the Study:

• To leverage established machine learning models and medicinal chemistry insights to select novel compounds for EBOV inhibition testing.
• To validate the strategy of using machine learning to prioritize compounds for in vitro screening before expensive in vivo studies.

Main Methods:

• Utilized previously developed Bayesian machine learning models for EBOV inhibition.
• Selected 12 compounds absent from the training set, combining model predictions with chemical expertise.
• Evaluated selected compounds for in vitro EBOV inhibition and cytotoxicity.

Main Results:

• Nine compounds selected by the model showed promising in vitro activity (EC50 < 15 μM).
• Arterolane (IC50 = 4.53 μM) and lucanthone (IC50 = 3.27 μM) were identified as novel EBOV inhibitors with low cytotoxicity.
• The strategy successfully identified active compounds, validating the predictive power of the machine learning models.

Conclusions:

• Machine learning combined with medicinal chemistry expertise is effective for prioritizing compounds for EBOV drug discovery.
• This approach offers a validated, cost-effective method for identifying potential antiviral agents.
• The strategy shows promise for application to discovering treatments for other viral pathogens.