AI and film animation tools lead to breakthrough in coronavirus drug discovery

About 30% of all respiratory tract infections are caused by coronaviruses, leading to widespread illnesses and, in some cases, to epidemic and even pandemic outbreaks, as we experienced with the COVID-19 pandemic. Despite the development of groundbreaking technology that enables the design of prophylactic vaccines, access to those vaccines is not equal across the globe, especially in low-resource countries, and also other hesitations prevent their adoption.

In addition, coronavirus variants are emerging that can have higher infectivity and resistance to existing vaccines and antiviral treatments. Therefore, fast-acting antiviral drugs with broad activity against multiple respiratory coronaviruses and the ability to be rapidly distributed as oral treatments are urgently needed.

In 2020, at the beginning of the COVID-19 pandemic, a multidisciplinary team of computational biologists, and infectious disease, medical chemistry and drug development experts formed at the Wyss Institute for Biologically Inspired Engineering at Harvard University. With early support from the Defense Advanced Research Agency (DARPA), the team, led by Wyss Founding Director Donald Ingber, M.D, Ph.D., sought to leverage the Wyss Institute's existing computational and biological modeling capabilities to rapidly repurpose existing FDA-approved drugs for the fight against the disease.

By creating a cohesive AI-enabled and physics-based molecular modeling and drug discovery pipeline built around film industry procedural animation software, they identified the orally available, FDA-approved drug bemcentinib, as a potential antiviral agent. However, to further optimize its activity with follow up support from Open Philanthropy-Good Ventures Foundation, they used this chemical compound as a launch pad for developing a more specific and effective antiviral drug with efficacy against a broad range of coronaviruses. Their findings are published in Frontiers in Molecular Biosciences.

Taking coronaviruses to the movies

During COVID-19, many research groups sought to develop drugs that target the external surfaces of Spike proteins shared by coronaviruses, which bind to receptor molecules on the surface of host cells and mediate virus entry. At the start of the project, the researchers hypothesized that rather than targeting these external sites, which are prone to mutation under the pressure of vaccines and therapies, focusing on hidden regions of the Spike protein could offer unprecedented benefits.

"We thought that constant regions that remain hidden while the virus initially binds to its host cell, but become accessible during a critical time window when it prepares itself for membrane fusion could be ideal sites. Targeting those could be a way to essentially lock the virus in at the pre-fusion stage before it can release its genetic material into the cytoplasm of host cells," said first-author Charles Reilly, Ph.D., a former Wyss Principal Scientist who led the project with Ingber. "The major challenge lay in identifying these moving cogs for which we could design drugs that could work as a universal wrench for multiple coronaviruses."

Previously, Reilly and Ingber had used tools from the film special effects industry to develop computational models of how molecules move and function from the atomic level to that of whole cells, and they used this to depict movements during the origins of human life – the fusion of a sperm with an egg cell – as featured in the short film, "The Beginning".

The team used a similar approach to model the dynamic transformations that the SARS-CoV-2 Spike protein undergoes after binding its receptor and while literally moving towards fusion. To accomplish this complex modelling, they essentially animated a variety of structural snapshots of the Spike protein that other researchers had captured at different stages of virus binding and fusion in X-Ray crystallography and other studies. This "molecular dynamics simulation," based on actual physical features of the Spike protein, allowed them to generate "synthetic data" in which they could glean regions in the so-called S2 subunit of the protein that underwent large-scale mechanical transitions. By applying AI methods to their synthetic data, they were able to pinpoint a specific site within one of those mechanically highly-active regions that revealed itself at a pre-fusion step before facilitating the Spike protein's embedding within the host cell membrane. This key site had potential to function as a binding pocket for drug compounds that would gum up the works to prevent membrane fusion and viral entry and, importantly, it was shared by different members of the coronavirus family.

Drugging the moving target

To identify drugs that could precisely fit themselves into the predicted binding pocket within the Spike protein's S2 subunit, the researchers computationally screened through about 10,000 existing drugs and ranked them according to their predicted binding affinities to the pocket. The highest-ranking drug that was also orally available was bemcentinib, which in their AI-enabled simulations and molecular docking studies remained strongly bound across multiple structural changes the entire region was undergoing, eventually freezing it at the pre-fusion stage. In fact, the team, with drug development expertise of Wyss Senior Principal Scientist Sylvie Bernier, Ph.D., and Wyss' Senior Director, Translational R&D, Ken Carlson, Ph.D., demonstrated that bemcentinib potently inhibited SARS-CoV-2 infection of human lung cells expressing the ACE2 receptor on their surface by preventing its entry into the cells.

Interestingly, the infection-blocking effect of bemcentinib had also been observed by other researchers, and bemcentinib had actually improved clinical outcomes in a clinical trial with hospitalized COVID-19 patients. But it was generally assumed that the drug had to function through its original target, the AXL kinase protein, which helps drive the growth of tumor cells.

To unequivocally prove that bemcentinib achieved its antiviral activity through our proposed mechanism and further improve its efficacy, we needed to create structurally similar compounds (analogs) that lack any affinity to AXL but retain their affinity to the Spike protein's binding pocket." Charles Reilly, Ph.D., a former Wyss Principal Scientist

Joel Moore, Ph.D., a talented medicinal chemist on the team, designed a series of novel compounds, the best of which, dubbed WYS-633, inhibited the entry of SARS-CoV-2 viruses into lung cells just like bemcentinib did and, importantly, without binding to AXL. WYS-633 also prevented SARS-CoV-1, MERS, and multiple SARS-CoV-2 variants from infecting cells, suggesting that the approach could indeed lead to a broad-spectrum antiviral drug against multiple coronaviruses.

To further improve the drug-like features of WYS-633, also with an eye towards oral bioavailability, the team performed another round of medical chemistry, AI-enabled computational analysis, and in vitro testing, which culminated in WYS-694, an antiviral compound 12.5-fold more potent than WYS-633. In a last crucial step, Bernier, Carlson, and other members of the team tested WYS-694 as an oral prophylactic treatment in mice expressing the human ACE2 receptor, which they infected with the SARS-CoV-2 virus. While both bemcentinib and WYS-633 failed to reduce the viral load, WYS-694 reduced the viral load in the animals by more than 4-fold and significantly inhibited SARS-CoV-2 infection.

"While additional computational predictions using Google's AlphaFold 3 machine learning algorithm supported our central hypothesis, we will still have to demonstrate that WYS-694 precisely binds to the target site we identified in the Spike protein's S2 subunit through detailed structural studies, and that the interaction alone and no other potential target enables the antiviral activity," said Reilly. "But our findings are extremely compelling and open up a new path for dealing with viral outbreaks."

"By aiming for an orally available drug that broadly inhibits multiple coronaviruses, even as a prophylactic treatment, we deliberately set the bar at maximum height," said senior author Ingber. "Yet, our integrated approach merging new AI-driven computational and experimental technologies has proven to be incredibly powerful in achieving this goal. Beyond producing a promising new drug that could be useful in future respiratory pandemics, this approach holds great potential for the discovery of drugs against a number of other virus families utilizing membrane fusion proteins, including influenza, HIV, Ebola, Measles, and others." Ingber is also the Judah Folkman Professor of Vascular Biology at Harvard Medical School and Boston Children's Hospital and the Hansjörg Wyss Professor of Biologically Inspired Engineering at the Harvard John A. Paulson School of Engineering and Applied Sciences.

Additional authors on the study were Shanda Lightbown and Austin Paul. The study was supported by the Wyss Institute at Harvard University, a contract from the DARPA (under contract #HR0011-20-2-0040), and additional funding by the Open Philanthropy–Good Ventures Foundation and Alfred P. Sloan Foundation.

Reilly, C. B., et al. (2025). Broad-spectrum coronavirus inhibitors discovered by modeling viral fusion dynamics. Frontiers in Molecular Biosciences. doi.org/10.3389/fmolb.2025.1575747.