Austrian researchers are teaching artificial intelligence to describe the 3D structure of therapeutic proteins. A team, from the University of Natural Resources and Life Sciences in Vienna, hope their software can one day help companies design processes for purifying advanced therapies during manufacturing.

According to Johannes Buyel, PhD, a professor at the Institute of Bioprocess Science and Engineering, their research aims to link the 3D appearance of proteins to their molecular structure.

“If we can come up with descriptors linking the physical properties of a protein to its molecular structure, we can predict–based on that structure–how it will behave,” he explains.

His hope is that one day manufacturers of cell and gene therapies can use the AI to predict how protein products will behave when purified in a chromatography column—“that is the ultimate goal.”

For the moment, the team is training AI in geometry to understand in numerical terms how to describe different shapes of protein.

“As a human, you can describe a door knob but, if you train an AI to understand a doorknob, it might not be able to identify all [types] of them, depending on which training dataset is used,” says Buyel.

Relevant physical parameters

The AI is being trained to identify relevant physical parameters of the protein in numbers, in the same way you might use diameter or height to quantify the size of the opening in the top of a coffee mug. [“The key point is] which numbers are representative of the object,” he notes.

The work to train the AI is, according to Prof. Buyel, an extension of previous research– but using the latest developments in machine learning.

“Using molecular descriptors to correlate with experimental data is nothing new, we did it about ten years ago,” he says. “The problem was that, back then, it was a merely descriptive approach and now we have a mechanistic model underlying the predictions.”

The team also needs to understand how close the predictions need to be to experimental data to help manufacturers improve their process development. In the future, manufacturers may also be able to understand the biology of proteins better by using AI.

Previous articleLeveraging Chromatogram Shapes
Next articleInnovative Commercial NK-EV Production Process