New Ebola Treatment Using Artificial Intelligence

March 24, 2015
Case StudyFeatured

Ebola virus replicating in a human host cell

Overview & Collaboration

The West African Ebola virus disease epidemic resulted in at least 11,310 deaths according to the World Health Organization. Ebola virus infections were reported across 6 West African countries, as well as in 3 European countries and in the United States, underscoring the potential for a global pandemic.

During the West African Ebola virus disease epidemic, Atomwise partnered with the University of Toronto and IBM to rapidly develop a treatment for Ebola virus infections. Atomwise provided the core artificial intelligence technology to perform the drug research. Researchers from the University of Toronto contributed biological insights about the Ebola virus, and recommended targeting Ebola virus glycoprotein 2, which is important for viral entry and fusion. This protein has three outer helices that fold in around a central trimeric core, and the hypothesis was that blocking this glycoprotein’s conformational shift could block viral entry. IBM supplied the supercomputer, a 64,000-CPU Blue Gene/Q.

The Technology

Atomwise defined a region to investigate for potential small molecule inhibitors by drawing a box around the bound outer helices in a closed x-ray crystal structure, and then deleted the helices, leaving the central core (see video). This region was then screened for compounds that bind to it with high affinity. Due to the on-going epidemic, we limited the screen to the almost 7,000 molecules that had at that time been evaluated in Phase 2 or later clinical trials.

Using our AtomNet system, we analyzed each of the molecules for its potential to bind the glycoprotein and rank ordered them.

These compounds already had safety data for use in patients and therefore could be rapidly brought forward for clinical trials to prevent and/or treat Ebola virus infections. Using our AtomNet system, we analyzed each of the molecules for its potential to bind the glycoprotein and rank ordered them.

Successful drug repurposing

Our research partners at the University of Toronto tested the top compounds that Atomwise predicted would bind to the glycoprotein and prevent Ebola virus entry into cell in an in vivo assay. The assay used modified Zaire and Sudan Ebola virus that contained the gene for green fluorescent protein, which caused cells to fluoresce green when the virus entered HEK293T cells. Counting the proportion of green fluorescent cells yielded a measure of infectivity of the virus. Researchers found that the best compound we identified gave a dose-response IC50 of approx. 12 uM. This binding affinity was remarkably high given the molecules Atomwise screened were only those drugs that had been tested for safety and already pursued as treatments for other diseases. Further, this compound appeared to act specifically on glycoprotein 2 because it did act on other viruses that entered cells by other mechanisms.

Importantly, the compound we identified had no previously described anti-viral activity, demonstrating that AtomNet is capable of identifying novel and non-obvious drug uses.  This approach may be applicable for rapid response to other pandemics, or highly-accelerated drug repurposing projects.


AtomNet predicting candidate treatments for Ebola infection – this prediction resulted in the lead molecule now awaiting animal trials.