Raphael ThysGenetic Engineering & Biotechnology News
Bill Whitford Life Sciences Strategic Solutions Leader, DPS Group
There is no longer any doubt that artificial intelligence (AI) is advancing biological discovery and biomanufacturing operations. In biological discovery, AI systems such as AlphaFold and the Atomic Rotationally Equivariant Scorer are celebrated for their uncanny ability to predict tertiary structures for proteins and RNA molecules. In biomanufacturing, AI systems usually enjoy less fanfare. Yet they can provide valuable functions such as pattern recognition, real-time assessment of batch quality, multivariable control for continuous manufacturing, prediction/optimization of critical process parameters, and anomaly detection. Such functions are critical to the success of gene and cell therapy operations.
Toni Manzano, PhDCo-Founder and CSO, Aizon
AI-driven deep learning algorithms are being applied to monitoring and inspection activities that used to be too nuanced for true automation. Machine learning applications can update biomanufacturing processes by drawing on diverse data sources, and multiple AI models can actively interact with equipment or processes to realize the “digital twin” approach, that is, the digital emulation of physical systems. These systems provide a starting point for a true adaptive strategy.
AI in biomanufacturing
Systems supported by AI can mimic human cognitive functions, use data from many sources (including historical and remote sources), learn from data of various types (including unstructured and multidimensional data), and accommodate new information (including empirical data) as it becomes available. Efficiency in data handling is increased by observability analyzers that distinguish between data that is valuable and data that is redundant, irrelevant, or corrupt. AI can support in silico modeling (including digital twins) and will soon be applied to varied tasks (Table 1), enabling full automation of adaptive model–based experimental design in product and process development, as well as closed-loop control of robotic activities in biomanufacturing.
We are already familiar with self-contained devices that perform dedicated tasks with living cells. For example, we’ve grown to trust microbioreactors that mimic the characteristics of large bioreactors and enable cost-effective experimentation. They are supporting savings in facility space, capital, labor, media, and consumables. Now we are becoming familiar with integrated systems.
Besides replicating human activity, fully automated and integrated systems will deploy machine learning–equipped robots that can iterate processes endlessly without fatigue or distraction. Many of these technologies are being applied in the pharmaceutical industry.
AI in cell and gene therapies
Gene and cell therapies reflect a range of technical approaches and production practices. For example, there are in vivo and ex vivo approaches, as well as autologous and allogeneic approaches. Nonetheless, there are commonalities that can be exploited in biomanufacturing operations. Shared opportunities include access to many of the same vectors (including viral vectors). Shared challenges include the inability to terminally sterilize products.
AI-enabled “smart manufacturing” is now beginning to support gene and cell therapy processing with an advanced digitalized integration of the manufacturing process—from supply chain management to operations control to final product track and trace. It is also empowering Industry 4.0 technologies such as the internet of things; real-time, integrated big data analytics; automated and cyber-physical systems; and advanced sensing technologies (including “soft” sensing technologies). Finally, it is supporting the widespread application of digital twins in modeling both equipment and operations.
AI can enhance systems that are common to many gene and cell therapies and that have previously presented challenges in traditional process design and manufacturing operations (Table 1). These challenges include the coordination of track-and-trace operations in patient-distal cell processing, as well as the decontamination of incoming process materials. Yet other challenges are encountered in establishing and maintaining standards for patient-related data with respect to security, privacy, curation, storage, and distribution.
Sensitive personal information such as economic status, location, habits, gender, or race can be deduced from the bio-psycho-social context used while reviewing systems to determine the best conditions for a therapy’s preparation. When the therapy is a gene or cell therapy, the requirements for tools that would ensure patient privacy and data security are especially (even uniquely) stringent. We are seeing the need for concurrent implementation of AI and machine learning–centric data governance, risk, and compliance protocols, as well as for the expertise and guidance of experienced, security-focused AI experts.
With gene and cell therapies, products and practices are so new, critical process parameters are often poorly understood. Consequently, production processes can evolve even after technology transfer.
Typically, gene and cell therapies must work after just one attempt, that is, after one course of treatment. To ensure that the first attempt is as effective as it can possibly be, producers are under pressure to discover process deviations—or better yet, prevent them.
Individual patient’s cell samples are diverse with respect to their condition, viability, and drug exposure. This can greatly influence a sample’s characteristics and performance, requiring greater process monitoring and advanced dynamic control. The expected efficacy, quality, and safety of the theraputic under the gene and cell therapy framework is even harder to achieve, batch after batch, than with small- or large-molecule entities.
Furthermore, similar manufacturing processes applied by operators and biopharma engineers present critical differences depending on the specific vector operation to be applied in each batch or patient. Continued supervision is required to double-check the specificity of the particularities associated with such manufacturing. Therefore, each patient must be linked to a single batch which is manufactured and controlled by means of both singularities (specific targets) and commonalities (universal specifications).
AI-empowered process control can aid in sensing or predicting processing anomalies, correlating current performance to past experiences, and determining what measures might remediate deviating parameters. Orchestration of distributed control with centralized processing of an individual’s cells or tissues will impose added processing and logistical burdens. Modern data connectivity and AI algorithms have a unique capability to maintain a real-time and dynamic picture of bio-based events. Because they can model nonlinear functional relationships, they excel at reducing model-process mismatches. Such systems can perform sensor validation, detect faults, and incorporate expertise and results from both bioprocess and control engineering.
Conclusion
The potential of AI in realizing gene and cell therapies is remarkable, and expectations for imminent improvements in biomanufacturing operations are high. The modernization of existing gene and cell therapy equipment and investment in digitalization are the initial steps that our industry must take if it is to follow—like other industries have already followed—the path to “smart” control of manufacturing.
Bill Whitford (william.whitford@dpsgroupglobal.com) is the Life Sciences Strategic Solutions Leader at DPS Group, and Toni Manzano, PhD (toni.manzano@aizon.ai), is a co-founder and the CSO of Aizon.