Depending on whom you ask, we are either in the early stages of, somewhat immersed in, or already fully immersed in Industry 4.0, the fourth industrial revolution. Industry 4.0 incorporates artificial intelligence (AI), machine learning (ML), and big data to enable integrated and autonomous manufacturing systems to operate independently of humans. Like 5G, the metaverse, and genetic engineering, Industry 4.0 is assumed to be a revolution of gargantuan scale. Your opinion as to whether we are at the beginning or in the midst of this transformation is likely to be based on your industry and what part of that industry you work in.
In pharmaceutical development and manufacturing, AI and ML have certainly arrived. Together, they already represent an important aspect of how modern contract development and manufacturing organizations operate. The steady increase in the complexity of manufacturing new medicines and the desire to reduce time to market drive the need for faster development and manufacturing. This pressure is propelling the implementation of AI solutions into many activities related to pharmaceutical development and manufacturing.
AI is not the only answer to these new challenges, but it is often a pivotal part of the solution. In practice, established methods still deserve their place in development and manufacturing, and they can be used to reinforce AI’s support of human intelligence, ingenuity, and innovation. Reaching the ultimate goal of lights-out manufacturing—closer and closer to the vision of AI—will still require tremendous effort and a substantial boost in IT infrastructure.
Nevertheless, we believe the industry is at a critical stage in its digital transformation journey. Today, AI has the potential to shorten operational cycle times while simultaneously increasing quality and/or reducing overall costs and raw material consumption. It provides a solid basis for greater automation and knowledge expansion as an enabler of better, faster decision-making.
Key distinctions, wide-ranging opportunities
At Lonza, we have successfully optimized product quality using computer vision technologies in quality assurance. We are also taking our first steps toward AI-enhanced genetic engineering based on bioinformatics methods. Most recently, we have been developing hybrid approaches leveraging AI, mechanistic models, and traditional statistics for scaling up processes.
It is important to understand the difference between AI and ML. AI is made up of ready-to-use products that leverage human-like pattern recognition or decision-making capabilities to solve individual tasks and perform various activities. ML is a set of mathematical algorithms solving individual tasks by making predictions based on assumptions derived from historical data. ML is a subset of AI, meaning that all ML is AI, but not all AI is based on ML.
At Lonza and across the biotechnology and pharmaceutical spaces, AI, ML, and big data are used routinely in areas such as research, computer-aided drug design, protein profile assessment, the engineering of mammalian expression systems by means of DNA element design, and the prediction of side effects for novel therapy forms. In small-molecule development and manufacturing, ML is used for synthetic route optimization, retrosynthesis, toxicological assessment of new chemical entities, and formulation design. On the other side of the drug development and manufacturing spectrum, ML is employed in developing controlled-release tablets—to assess the hardness, particle size, moisture, and other factors to predict a tablet’s in vitro behavior.
AI and ML are also increasingly being employed in pharmaceutical manufacturing. In an area such as process analytical technology (PAT), spectroscopical methods like Raman are used in combination with an ML algorithm to monitor critical process parameters. When used with a Raman in-line probe, the PAT and ML combination can monitor metabolites and raw material concentrations, which cannot be measured directly through Raman linear regression. Today, we can even find research describing the indirect measurement of pH values using Raman-ML methods. Similar results have been reported using a combination of either Fourier transform infrared spectroscopy or ultraviolet-visible spectroscopy with ML. Even monitoring of Escherichia coli contamination with Raman and ultraviolet-visible spectroscopy was recently published.
These methods allow production process performance to be monitored without taking a manual sample if an in-line spectrometer probe is installed. This has significant advantages, including less variability in analytical test results, fewer verification activities (in the case of a validated system), and more process knowledge, given that we are measuring process performance continuously and can reduce the risk of contamination through the manual sampling process.
Another significant impact that ML is having on pharmaceutical manufacturing is its ability to make predictions based on historical data. This can have a very important impact in an area like predictive maintenance. By combining an ML algorithm with high-frequency sensors and assessing assets for factors such as sound, vibration, or electricity consumption, it is possible to predict the latest possible time for maintenance or repair of an asset. This can reduce costly production equipment maintenance time and increase the asset’s availability. This approach is applicable not only for manufacturing equipment but also for laboratory equipment (such as high-performance liquid chromatography equipment for quality contol) and utility systems (such as heating, ventilation, and air conditioning systems for clean air).
In the commercial manufacturing area, we are focusing on bio operations and processes. We are working on ML algorithms in combination with Raman spectroscopy to give us the ability to monitor glucose levels (or the levels of different metabolites) in bioreactors without taking manual samples. This ability also allows us to accrue valuable process knowledge at the same time.
We are also investigating the potential of AI and ML applications in product technology transfer. We encounter different scales and different equipment setups during technology transfers. The number of process variables and critical quality attributes involved in technology transfers adds another dimension of complexity. AI and ML applications are predestined to predict process performance or critical process steps in such technology transfers, helping to address these complex challenges.
Another area that we are investigating is the use of AI and ML in deviation management and change control applications. Such applications can add significant value as transactional intelligence systems.
AI and ML systems can also be used to assess behavioral challenges and improve training. For example, AI is being used with computer vision technologies to study how people behave in clean rooms. And we are using virtual reality to train unit operations. When virtual reality is in place, neither trainers nor trainees need to be on site. Also, equipment that has yet to be installed can be the subject of training. Finally, routine operations can be handled by workers who need to be off-site.
Data science, which represents the next revolution in pharmaceutical manufacturing, will become a fundamental technology in all pharmaceutical manufacturing areas. A crucial aspect driving this AI revolution is the collection, management, and utilization of specific process data. In addition, general data in areas such as warehouse conditions or raw materials will play a critical role. These data sets will be the basis for advancing our use of AI applications.
As the manufacturing process itself becomes more and more automated, like a self-driving car, it will be able to react to unpredictable events during operations. There will be significantly more data available, and edge computing will help to process this data, in real time, at the source to steer the process toward a “golden batch.” More process knowledge will be gained through these technologies, and the decade-old vision of a parametric release of the product may well become a reality.
There are currently a wide variety of uses for AI, with many associated positive impacts. The range of positive impacts driven by AI and ML is helping us increase safety, quality, and sustainability while lowering costs. While a continuing increase in AI and ML integration will put new demands on IT infrastructure and employees, the indications are that AI and ML may yield exponential results if employed in a thoughtful way.
Stephan Rosenberger, PhD, is head of digital transformation at Lonza.