Cell therapy developers looking to use predictive modeling in manufacturing should start small, according to industrial AI firm, Kosten Digital, which says taking the time to establish the right infrastructure is vital.
Artificial intelligence, at its most basic level, is the practice of automating the analysis and detection of patterns in data. This information can be used to model the process from which the data was derived and predict how any changes will impact its operation.
For complex, data-rich processes like the manufacture of a cell therapy, AI has obvious application. Process optimization is a good example. AI systems can analyze huge amounts of data in real-time to fine-tune manufacturing processes, which would help reduce variability and improve product consistency, something that is critical in cell therapy production.
AI can also predict when equipment is likely to fail and suggest maintenance before failure occurs, cutting down on unexpected downtime.
However, while the cell and gene therapy (CGT) industry has recognized the potential, few companies to date have embraced AI, according to Kosten co-founder Claudia Zylberberg, PhD.
“There’s a lot of buzz around AI in biopharma and CGT manufacturing, but actual use has been fairly limited so far. Companies see the potential for AI to improve things like process control, efficiency, and quality, but we’re still in the early stages of adoption.”
Upfront investment
One of the major adoption hurdles is the need to set up systems able to monitor production processes and share information.
“To implement AI in manufacturing, a CGT or drug firm would need a solid data infrastructure as the foundation,” points out Zylberberg. “This means investing in sensors for real-time data collection, cloud computing or powerful in-house servers for processing, and a robust data management system to organize it all.”
Managing the analytics tools and AI software required is another potential challenge.
“Firms would need to build or hire AI expertise to develop and maintain these AI models,” she continues. “They would also need to ensure compliance with regulatory requirements, which might mean additional resources for validation and testing to prove that the AI systems are reliable and meet industry standards.
“Re-skilling the workforce, quality of data acquisition, and automation are critical aspects for implementing ML/AI. Lastly, companies would need IT security and cybersecurity infrastructure to protect sensitive data and ensure the integrity of AI systems. In short, it’s a combination of tech, talent, and regulatory frameworks.”
Small steps
The key to implementing the tech and developing the talent needed to use AI in manufacturing is to build slowly.
“If a company is still developing its digital capabilities or doesn’t yet have a strong data infrastructure, it might make more sense to focus on building those basics first,” notes Zylberberg. “AI works best when it has a lot of high-quality data to learn from, so if those systems aren’t in place, the ROI on AI could be lower.”
“Starting small with ML/AI, and even digital twins like in pilot projects or specific areas like predictive maintenance or process optimization, can be very valuable. It allows a company to experiment with AI, learn how it fits into their processes, and build toward full integration when they’re more digitally mature. So, while it’s a significant investment, ML/AI is a tool that can grow with the company over time.”
