AstraZeneca's Focus On Predictive Control, Semi-Autonomous Processing
By Tyler Menichiello, Chief Editor, Bioprocess Online
Most conferences these days have dedicated sessions and tracks for AI, and if you’ve attended any in the past few years, you’ll know that most of these conversations have been full of hype and promise but have generally offered little in the way of practical advice or results. However, that seems to be changing. Enough time has passed for Big Pharma to conduct pilots and reflect on the lessons it’s learned. And while the hype will stick around for the foreseeable future, I left this year’s BioProcess International conference in Boston feeling encouraged by where the industry is headed with regard to AI implementation.
The sessions I attended were less centered on the hype and more focused on the tangible progress being made, as well as the comprehensive benchmarks that lie ahead to fully realize the power of these technologies. One such session was a keynote delivered by Preeya Boppana, director of MSAT process analytics and laboratories at AstraZeneca, titled “Transforming Biologics Manufacturing: Leveraging Predictive Analytics and Advanced Process Control for Next-Generation Operations.”
As the title suggests, her keynote painted a picture of what the next generation of biomanufacturing may look like. The goal is to build processes that run on a framework of real-time analytics, digital twins, and AI systems capable of not only catching anomalies sooner and reducing the occurrence of deviations, but that are self-learning and can offer predictive process control through automation.
She shared a bit about the work AstraZeneca is doing on this front and where she believes the industry needs to invest to achieve this future state of biomanufacturing.
Picture Next-Gen Bioprocessing As A Self-Driving Car
Throughout the keynote, Boppana used the apt analogy of a self-driving car to explain what she sees as the future of bioprocessing.
“How do we take our manufacturing process from being driver-assisted to semi-autonomous?” she asked the audience.
What the industry is working towards, she explained, is a system that maintains process control within set boundaries (picture a vehicle with lane assist), one capable of making automatic corrections and shifting the process back to center as needed.
“We’re not asking, within our regulatory space, to widen the lane,” she said. “We just want to stay in [the lane] a bit more semi-autonomously.”
The ultimate goal is to build autonomous bioreactors capable of maintaining their own environments and parameters. This futuristic bioreactor is built on improved process awareness, prediction, and adaptation.
To achieve this next-generation bioprocessing, said Boppana, there are five dimensions that need to come together:
- Sensing (i.e., PATs and real-time process monitoring)
- Understanding (i.e., hybrid models, multi-grade analytics, digital twins)
- AI-Driven Decision-Making (i.e., using AI to do things like flag process drift, review deviations, and accelerate hypothesis generation, ultimately moving toward predictive control)
- Robotics (i.e., implementing robotics for routine process tasks to free up operators, such as in stock solution preparation or column-packing)
- Orchestration (i.e., end-to-end digitalization from development through tech transfer, into operations, quality, and through the entire supply chain)
Getting to the next generation of biomanufacturing operations will require investment in these key aspects over the coming years, said Boppana.
AstraZeneca’s Focus On Improving Sensors And Real-Time Monitoring
According to Boppana, AstraZeneca is currently focused on investing in the sensing component. “For us, one of the critical aspects to move away from operator-driven decisions is investing in process analytical technology in the cell-culture space,” she said.
Broadly speaking, the company is intent on moving from offline analytical testing to inline monitoring. The specific example she pointed to is using inline Raman spectroscopy to visualize cell growth and live cell density. Not only is inline testing better from a data-gathering perspective, but it reduces the chance of contamination, said Boppana.
Furthermore, offline testing is reactive, not predictive.
“Every time you take [an offline] sample in your manufacturing process, you’re that many hours behind what the process is actually telling you,” she explained.
Aside from these lagging indicators, offline analytics are generally univariate, which can miss the bigger picture of complex process interactions.
Tying back to the car analogy, Boppana described inline testing and continuous process monitoring as the sensors that “see” where the “car” is going. These inline sensors, in theory, will feed into intelligent systems that, when necessary, trigger automatic responses, thus keeping the process in optimal conditions and shifting the paradigm away from reactive process control to predictive process control.
“That enables consistent batch processing, adaptive feed strategies, and early detection of nonconformances,” said Boppana.
Controlling Raw Materials
“One of the key components in process control is to limit the variability of your raw materials prior to hitting your product stream,” Boppana said, emphasizing the value of these technologies in validating raw materials.
She outlined the basic steps necessary to control raw material variability:
- Breaking down raw material data silos (e.g., vendor data, critical functional attributes, QC testing data) and building a single raw-material database
- Data engineering: piecing together which raw material went into which media batch for which production batch
- Discover and forecast: looking for variability in the raw material and correlating that variability in your process
- Testing and validating these correlations in the lab, at a small scale
- Prospective control, which for AstraZeneca means implementing free-screening techniques to avoid using compromised or variable material in processing
The Decisive Decade Of Biomanufacturing
Boppana called these next ten years the “decisive decade” in terms of achieving next-generation bioprocessing. Her presentation slides outlined three key horizons the industry needs to cross to get there:
- Horizon 1: Connect and Improve (2026 – 2028)
- Building the data foundation
- PAT expansion
- Predictive monitoring pilots
- AI-assisted systems to investigate and reduce deviations
- Initial robotics pilots
- Horizon 3: Embed and Scale (2028 – 2031)
- Increased use of hybrid models
- Predictive control in select unit operations (with humans retaining ultimate decision-making power)
- Improved digital twins
- Human-machine workflows
- Data-rich review
- Horizon 3: Orchestrate and Adapt (2031 – 2035)
- Achieve semi-autonomous bioprocessing
- Network-level learning
- Broader robotics execution
- Near real-time release
“We have a digital revolution on our hands, and we get to choose how we get there,” Boppana said to conclude her keynote. She believes the future of biologics manufacturing will be shaped by the organizations that combine process science, digital infrastructure, AI, and automation into a coherent operating model.
And you better believe Bioprocess Online will be covering the industry’s evolution through this decade, every step of the way.