Artificial intelligence can help advanced therapy manufacturers bring their products to market by enabling them to cut costs and understand their processes better. That’s according to Michael Sokolov, PhD, the co-founder and COO of DataHow, a spinoff company from ETH Zurich specializing in bioprocess modeling and operations support.

As Sokolov explains, around five percent of batches fail—with impacts both on patient health and potentially thousands of dollars lost in manufacturing revenues.

Decision support in process development is even more important for the latest advanced therapies, notes Sokolov, where there are fewer established processes and product development can be highly complex.

“There’s less background knowledge about how to robustly develop these drugs compared to established processes for producing therapeutic proteins,” he explains. “In these new cases, AI can enable robust manufacturing and eventually help these new drugs reach the market.”

Intense pressure to optimize processed and reduce costs

Gene and cell therapies are often expensive, putting manufacturers under intense pressure to optimize their processes and reduce costs,” points out Sokolov. With autologous CAR-T therapies, for example, each treatment is personalized to an individual patient, who may die if they don’t receive a dose on time.

“Pharmaceutical companies need to learn faster than ever before,” says Sokolov. “Now drugs are more personalized to patients, they can only harshly reduce their costs by gaining all the value from their data and running through fewer iterations in the lab.”

Companies need to use AI to transfer learning between multiple batches of similar but not identical therapies. He believes the pharmaceutical industry also must learn to share data inside and, in the future, outside of their organizations.

Going forward, he maintains that companies will move to Industry 4.0 by using machine learning to provide well-documented results that support human decision-making.

“We expect a big boost in digital twin technology in the next five years, not as a buzzword, but as software to run alongside everything that is happening [in process development and manufacturing] to provide real-time predictions and support.”

Previous articlePrediction Stability for Optimizing Machine Learning Output
Next articleMachine Learning Tool, A-Plus, Makes the Grade for Liquid Biopsies