Artificial intelligence (AI) is now an essential tool in biopharma manufacturing, yet industry still needs guidelines on how LLMs, ML, and other technologies should be used in drug making, say the authors of a new study.

Data has always been critical in biopharmaceutical manufacturing. The ability to control drug production relies on being able to monitor process parameters, detect anomalies, and make corrections—all of which depend on data.

In recent years, drug makers have started to track a wider range of parameters per process. In part, this is because the monitoring technologies have advanced, but the increase also reflects regulatory demands for more detailed information about how drugs are made.

The resulting increase in process data has fueled the search for more sophisticated analytical tools, says Shyam Panjwani, PhD, principal biotech data scientist at Bayer, who is co-author of a recent study looking at evolving biopharma AI use.

“Advanced AI/ML is essential for analyzing the high-dimensional data produced in new biologics manufacturing. For instance, in cell therapy production, flow cytometry methods and single-cell RNA sequencing generate high-dimensional cell characterization data that is used to quantify quality attributes such as cell viability and purity. AI/ML techniques can be leveraged for the precise estimation of these quality attributes,” he tells GEN.

And flow cell data analysis is just one way AI can make cell therapy manufacturing more efficient, according to Panjwani, who cites process automation as another example.

“It is true that cell and gene therapy manufacturing is currently less automated than traditional drug manufacturing. AI-based robot-arms may be able to minimize manual handling in the long term.

“In the current realm, AI/ML models can certainly automate some of the manual tasks, such as cell imaging using microscopy for confluency estimation during cell expansion and differentiation. To achieve this automation, high-throughput image capture devices need to be installed in the production line to collect the data for AI model training and real-time inference,” he says.

Regulatory needs

In their study, Panjwani and Bayer colleagues, Hao Wei and John Mason, suggest that clearer AI guidelines would help more biopharma companies adopt the technologies for commercial-scale production.

Panjwani adds, “A variety of applications of AI/ML models for pharmaceutical manufacturing, such as process monitoring, predictive maintenance, and real-time testing, are widely reported in the literature and implemented at a pilot-scale.

“However, the potential of such models for commercial manufacturing has not been fully realized due to the additional burden of validation requirements and ambiguity regarding their acceptability by regulatory agencies.”

There are some signs regulators are moving to clarify their requirements on AI, according to Panjwani, who points to draft guidelines the FDA released last January as an example.

“With the new regulatory draft guidance on AI/ML use for biological products, drug manufacturers have become aware of the regulatory expectations regarding the stringency of AI/ML models used in life-cycle management,” he adds.

However, there is much more to do. Panjwani cites large language models (LLMs), a type of AI program that can recognize and generate text, as an area still in need of clarification.

“Clear guidance on the use of LLMs, the latest class of AI, is missing. The dos and don’ts of LLM applications for pharmaceutical manufacturing are needed to not only increase the manufacturer’s efficiency, but also ensure product quality for patients.”