Skip to main content
SHARE
Blog

AI for pharmaceuticals

Researcher is standing at his lab's computer with blue gloves and glasses, working on something on the computer

Like materials science, drug development can be an endless guessing game.

We know a lot about chemicals and their properties; what we need to know is how they will interact to address specific health problems. It’s a much more difficult problem, like looking for a needle in an ocean of haystacks.

“One of the challenges for developing a drug to attack COVID, for instance, would be that you need to create a molecule that will attach itself to a particular part of the virus and prevent it from infecting humans,” said ORNL computational scientist John Gounley. “The problem is, there are a very large number of molecules.”

Fortunately, you can skip right to the most promising candidates using an AI foundational model, which is a large deep learning neural network. Instead of creating task-specific AI models that perform a single function, all your information is in one place. Chemists can then use these foundational models to communicate with drug developers to organize information, run simulations and have the best suggestions for testing in a fraction of the time.

This approach played a major role during the COVID-19 pandemic as researchers from across the world began working on drugs to treat the disease. With speed at the highest premium, ORNL scientists used AI.

“We are at a moment where the potential of AI is being recognized for developing pharmaceuticals,” said fellow ORNL computational scientist Jens Glaser. “It’s the obvious choice to use AI here because it’s a field where there’s a lot of empirical knowledge available and data that exists. Depending on what kind of disease you’re targeting, you may have a wealth of existing data you can leverage for finding a new therapeutical approach.”

 

Continue reading ORNL Review: Turning AI into something we can trust