Process Analytical Technology (PAT) and Artificial Intelligence: The Expanding Impact of ML in Drug Manufacturing

Process Analytical Technology (PAT) is an analytical technique that measures critical quality and performance indicators in raw materials, in-process materials, and processes during manufacturing. PAT solutions are now embracing improvements with recent advancements in artificial intelligence.

Process analytical technology has been around for decades, but the recent advancements in artificial intelligence have helped make it more potent than ever before. Process data generated from PAT solutions can now be analyzed by custom machine learning algorithms that use deep learning to predict what will happen next. These technologies help quality by design initiatives by continuously monitoring key performance indicators during production. Machine learning algorithms can learn from experience and form patterns based on what they’ve seen before, making them better at predicting manufacturing outcomes.

Custom AI-based PAT solutions aid Quality by Design and manufacturing process management to ensure the safety of drug products for patients worldwide. This article will explore how novel machine learning techniques can keep biopharmaceutical manufacturing at the forefront of standardized quality, satisfying capacity, and highest efficiency.

Manufacturing Consistent Products in a Highly Variable Ecosystem

Pharmaceutical manufacturing processes include a variety of steps, and the materials used in these operations can differ significantly in terms of characteristics such as moisture content, viscosity, or crystalline structure. Biopharmaceutical manufacturing is tricky since it involves a wide range of bioprocessing methods and working with live organisms. Simultaneously, because the equipment and its components have inherent tolerances, the devices used do not always function in the same manner throughout the production life cycle. Because the raw materials are variable, and the batch procedure has its own variability in equipment behavior, the output products vary even though the operators follow a predetermined formula with specified set-points each time.

Manufacturers of biopharma products can’t count on standard process control procedures to produce repeatable results. Unpredictable variabilities in raw materials, equipment setup, and processing circumstances are unavoidable, resulting in product quality variation. Developing strategies with measurement and control capabilities that compensate for process variation and promoting a culture of continuous learning is a more practical approach.

The goal of applying process analytical technology is to develop a dynamic manufacturing process that adjusts for variances in raw materials and equipment to produce a consistent product. The PAT aims to provide a basic, scientific-based understanding of those variables critical to a stable procedure’s success and resulting product quality. PAT tools and overall approach provide a framework for understanding process performance. It promotes process knowledge and a continuous learning approach by connecting the impact of variables to system effectiveness and process control.

The advantages of adopting PAT tools are reduced process costs, improved product quality, safety and consistency, elimination or reduction of product rework, optimized energy, and material usage, quicker processing cycle times, facilitation of regulatory acceptance and compliance. These sum up to creating a robust and stable process that results in the production of goods that are produced correctly every time.

Quality by Design (QbD) and Process Analytical Technology

PAT positively influences organizational efficiencies; it leads to safe and sustainable processes and enables data-driven decision-making. PAT is means to achieve Quality by Design (QbD). QbD is the notion that quality should be built into a product with an understanding of the product itself and the manufacturing process through which it is developed and manufactured, together with an awareness of any potential risks.

Quality by Design’s first step is to determine Critical Quality Attributes (CQAs) that guarantee a high-performing pharmaceutical product. The next step is to implement Process Analytical Technology tools to control Critical Process Parameters (CPPs) and Key Performance Indicators (KPIs) to improve the quality and yield of manufacturing processes. When used in tandem with a scientific design of experiments strategy, these instruments are necessary for determining the ideal design space for product creation and optimization.

Types of PAT Tools for Comprehensive Process and Reaction Understanding and Control

Many different Process Analytical Technology tools are now available for comprehensive process understanding and control. A combination of three essential PAT tools is required to run a successful project:

• Multivariate data acquisition and data analysis tools - These are often sophisticated program suites that aid in designing experiments, gathering raw data, and statistically analyzing this data to identify what parameters are Critical Process Parameters.
• Process Analytical Chemistry (PAC) techniques - These are analytical instruments used to measure the characteristics of Critical Process Parameters. Near-infrared spectroscopy (NIRS) is one of the most common; however, fiber optics, biosensors, Raman spectroscopy, and other spectroscopic and not spectroscopic methods are also used.
• Knowledge management and continuous improvement tools - Software that collects Quality Control data from multiple processes over time to identify process shortcomings and implement and evaluate process improvement initiatives.

How Do Novel Artificial Intelligence Solutions Take Pat Systems to the Next Level?

The utilization of PAT in complex biopharmaceutical processes is critical for maintaining product quality across a wide range of manufacturing scales despite living cells’ heterogeneous nature. However, process analytical technology can be challenging to implement due to contamination concerns with off-line measurements, limited samples, poor sensitivity, time constraints, etc. On-line monitoring of Critical Quality Attributes (CQAs), Critical Process Parameters (CPPs), and Key Performance Indicators (KPIs) in real-time resolve these issues by facilitating improvements in process control, root cause analysis, and process characterization.

Ongoing trends currently faced by this industry emphasize the importance of these measurements:

• focus on continuous processing
• a shift towards single-use technologies
• the emphasis put on improved process productivity and safety

PAT implementation covers everything from mid and late-stage process development for process design, process scale-up and transfer, and continuous improvement of the commercial manufacturing efforts. At each stage, process analytical technologies can be improved by utilizing emerging tech and artificial intelligence solutions.

Model-Based Design of Experiments (DoE)

Establishing a meaningful relationship between processing conditions and product quality is critical for the successful implementation of PAT. The information from the standard bioreactor control system (e.g., pressure, pH, T) and additional PAT tools must be combined and tallied in a single database to process it further.

Process variables are continuously changed throughout the development process via Design of Experiments (DoE) to see how changing process parameters affect the CQAs. This exploratory research generates a vast amount of multivariate process data. Advanced data modeling techniques are required to manage these quantities of data and extract the information necessary for process control. Traditionally they rely on various multivariate data analysis models. It is not uncommon to move towards more advanced deep learning techniques to process the data and achieve needed outcomes.

For clinical development and production, the MVDA or deep learning model may be utilized to monitor the procedure and ensure the required quality of goods is produced. These models should govern the process when abnormalities or disturbances occur in the second stage. They usually utilize historical data to compare a running process’s output in real-time.

The capabilities of these devices can be utilized to check if the procedure is operating following specifications ensuring final product quality, and even start a corrective action on the process. In this manner, the PAT systems aid in maintaining the process on track toward its intended conclusion.

Related case study: APIs Production Process Predictive MonitoringTo improve the current batch production processes, a producer of active pharmaceutical ingredients approached us to implement AI models to analyze spectral data.Our challenge? Building a model that analyzes real-time data streams from the chemometric techniques and identifies potential outliers that may lead to deterioration of quality, based on historical data. The target of creating data pipelines and the model is the translation of the optical or spectral data into meaningful information that helps in maintaining an innovative pharmaceutical development process. The benefits are improved efficiency, predictability, and quality assurance of manufacturing operations and yields. Read more about this case study.

The integration of all available data and process control modeling is now possible by applying custom-made models for specific product lines or using available ready-to-use solutions. The benefit of applying advanced AI models rather than standard MVDA is the possibility of leveraging massive amounts of data into improved process understanding and moving towards a prescriptive, innovative pharmaceutical development process. Several AI solutions can do all of this, and they’re designed to adapt or learn automatically to new data sets and create new causal links between process parameters and process variables, in which the underlying algorithms will be utilized.

Intensifying Continuous Processing with Deep Learning

New Process Analytical Technology tools in Artificial Intelligence are providing manufacturers with a comprehensive process understanding and control.

Nowadays, with innovative concepts in pharmaceutical manufacturing processes and particularly process intensification, there are many creative ideas for improvements in equipment, production processes, and analytical methods. This leads to a significant boost in productivity and sustainability.

Continuous processing has become increasingly prevalent within the pharmaceutical industry over the past decade due to its ability to enhance productivity while reducing costs associated with manual labor, equipment maintenance, etc.

The main objective of current research is to intensify continuous processes. It allows for more efficient continuous production, improved product uniformity, and significant energy savings compared to conventional batch techniques. Furthermore, continuous manufacturing frequently necessitates stricter safety standards because it only produces hazardous chemicals when needed and does not have to be stored in massive amounts.

Because of the inability to accomplish such advantages from continuous plant activity without a considerable degree of automation, the implementation of intensified continuous flow processes requires automated instruments and tight product quality control.

PAT tools are not only about installing intelligent sensors in the field but mainly about a robust IT solution that enables tracking, providing visibility, and control strategy for continuous processing. Advanced process control (APC) enables the detection, identification, or analysis of anomalies through real-time monitoring processes. APC is usually coupled with various AI solutions like anomaly detection, classification, or predictive analytics.

Complex non-linear process models can be analyzed and interpreted automatically through advanced artificial neural networks or deep learning algorithms that identify exceptions and provide alerts and feedback. Pat probes can be used to detect critical quality attributes in real-time. The slurry can be sent to waste as soon as the deep learning model finds an out-of-specification material or uniformity anomaly, suggesting an abnormal content mix to ensure final product quality and save production time for the quality production process.

The improved data quality and decreased lead time provided by real-time PAT monitoring and data analysis in conjunction with an integrated system control support a real-time release and the QbD strategy. This method reduces the need for post-process quality testing, meets the regulatory framework and process safety, and considerably lowers inventory levels and lead times.

Advanced data-driven modeling based on AI solutions can be applied to comprehensively monitor and predict the whole biopharmaceutical manufacturing process (and not just single unit operations). This is especially vital in continuous manufacturing. The emerging trend in moving towards continuous manufacturing drives the development of more robust artificial intelligence-based systems that automate process management.

Knowledge Management With Big Data Analytics

The correlation between CQAs and CPPs is complex, and it is only possible to describe them using multi-factorial relationships. Applying chemometrics, i.e., mathematical and empirical statistical methods, to physicochemical data is necessary. Therefore, businesses interested in adopting PAT need to move towards novel strategies to deal with the tremendous complexity found in some processes. These strategies bring big data analytics and machine learning.

Big data analytics software enables knowledge-based risk assessment of methods and assures process execution in the appropriate process design space. Risk reduction is possible at all phases of development thanks to a deep understanding and controllability of all critical system parameters that affect process stability and safety (thermodynamics, accumulation, reactor setup parameters, etc.).

Process Control With Digital Twins

Digital twins are representations of the manufacturing process used to help regulations and assure high visibility into production processes. The term “digital twin” refers to a digital replica of a real-world manufacturing plant in operation to efficiently predict and manage maintenance and lifecycle management. A digital twin is a data-driven imitation of a real-world manufacturing facility to optimize production in real-time and provide life cycle support.

Suppose you use an AI-based tool like a neuronal network, which is an advanced statistical model developed using process operation data sets, as a digital twin definition. In that case, it implies that the process design stage has been completed and that the plant has already been run. As a result, there’s no point in using digital twins as a tool for process design. Instead, a digital twin aims to apply already running manufacturing processes for deeper insights and introduce process alterations.

To ensure quality throughout every stage of production processes, some hurdles still need to be overcome with machine learning models, such as parameter tuning, which require expert intervention to validate accuracy at each step along the way without hampering time efficiency gains brought about through AI automation capabilities.

Adaptive and Predictive Control

Adaptive Control involves using process understanding of the effects of adjusting critical control parameters at any given time on future product quality to optimize process development. There is an inseparable link between adaptive control and process modeling with digital twins and the design of experiments, as this is where the models developed by chemometrics are applied to improve process efficiency. Adaptive Control involves monitoring the process to become familiar with its behavior. This method can be applied by process experts (through Model Predictive Control (MPC) or other predictive modeling algorithms with visualization dashboards) or automated using prescriptive machine learning techniques.

The Benefits and Challenges of Process Analysis Technology

The benefits to implementing PAT incorporate better product quality and uniformity, process cost reduction, enhanced process, and product safety. Several issues are involved in implementing PAT that covers technical and company cultural matters. The challenges of applying PAT include:

• high entry costs,
• dedicated analyzers required,
• complex chemometric models,
• systems integration,
• PAT data management,
• PAT systems development,
• regulatory approval needed.

PAT implementation costs are included in the acquisition of new technology, modifications to existing infrastructure, long-term maintenance, and training staff. Corporate culture is crucial to achieving smooth implementation without operation barriers. PAT positively impacts organizational efficiency and leads to the development of safe and sustainable processes, which accelerates the development pipeline.

The benefits of applying AI-based process analytical technology in the pharmaceutical industry include:

• account for process variability,
• reduced production cycle time,
• prevent the risk of lost batches,
• enable real-time release,
• increase automation to free up operators time,
• improve energy and material consumption,
• facilitate continuous processing.

Future Trends of AI in Pharmaceutical Industry Applications of PAT

Process Analytical Technology (PAT) has been around for decades, but recent advancements in artificial intelligence enhance its capabilities. PAT involves the measurement of key quality and performance indicators in raw materials, in-process materials, and processes in real-time. Artificial intelligence solutions are being used to improve quality by design and manufacturing processes. The benefits of PAT include better product quality and uniformity, process cost reduction, enhanced process, and product safety, and increased organizational efficiency. The challenges of PAT implementation include the cost of acquisition and modifications to existing infrastructure, long-term maintenance, and training staff. However, the future trends of AI in industrial applications of PAT look very promising.

The future PAT tools application in the pharmaceutical industry is to provide autonomous control and automated feedback and correction mechanisms applied to the pharmaceutical manufacturing process.

References

A Review of Machine Learning Algorithms for estimating Critical Quality Attributes from Multi-Sensor Data - O’ Mahony, et al.

Process analytical technology (PAT) tools for the cultivation step in biopharmaceutical production - Streefland, et al.; Engineering in Life Sciences

Current and future requirements to industrial analytical infrastructure—part 2: smart sensors - Eifert, et al. - Analytical and Bioanalytical Chemistry