A New Strategy to Boost the Pace of Antiviral Drug Discovery

So the runny nose you’ve been enduring all week is “just a virus.” You know the drill: drink plenty of liquids and wait it out. Why is it that there’s an abundance of antibiotics while virus-fighting drugs are so much harder to come by?

Unlike antibiotics, in which one strain of penicillin can treat multiple bacterial infections including pneumonia, scarlet fever, and strep throat, antiviral drugs are typically designed to target just a single virus. This narrow scope makes it unprofitable for drug companies to invest in developing new antivirals. The obvious solution is to develop antiviral drugs that can treat a broad spectrum of diseases, but that’s proven to be easier said than done.

Now, a research team including Dr. Noa Katz of Professor Roee Amit’s laboratory in the Technion Faculty of Biotechnology and Food Engineering, in collaboration with a group of electrical and computer engineers at Ben-Gurion University of the Negev, seems to have found a new way to streamline the drug discovery process for antivirals by combining the power of synthetic biology and machine learning.

Dr. Noa Katz

The typical lifecycle of a virus comprises a series of steps that include attaching to a cell’s outer layer, penetrating the cell, replicating the viral genetic material, and then sending newly formed viruses back out into the host’s bloodstream. Antivirals work by disrupting this process at any point along the way. But traditional methods of antiviral discovery involve lots of trial and error and are slow, labor-intensive, and costly.

Professor Roee Amit

In contrast, the new method uses synthetic biology to identify molecules with properties that have a relatively good chance of disrupting a virus. This is done by first generating a large, high-quality experimental dataset of molecules that are suspected of being able to bind to the virus. These data are then fed into a neural network, or machine learning algorithm, which creates a narrower profile of molecules with a high probability of inhibiting the viral life cycle. Once those molecules are identified, the neural network looks for others with a related structure that might also prove effective, thereby potentially speeding up the process of antiviral drug development.

In the groundbreaking study published in Nature Communications, the researchers demonstrated that this combined synthetic biology/machine learning approach resulted in the discovery of molecules that could bind to proteins from two distinct viruses. They hope this achievement provides the blueprint for an approach that could eventually be used to develop broad-spectrum antiviral drugs.