ON THE CONVERGENCE OF BIOTECH AND TECHNOLOGY REVOLUTIONS
Since the discovery and introduction of first biopharmaceuticals in late 70s to early 80s, the biopharmaceutical industry has come a long way with many life-changing medicines available to patients. From insulin to recombinant proteins, then to monoclonal antibodies as dominant modalities, we are seeing more cell and gene therapies and other novel modalities coming to the mainstream. While the various modalities are being advanced, in parallel we also see advances in different and more intensified with reduced manufacturing footprint biopharmaceutical manufacturing/processing formats such as single-use technologies and continuous manufacturing. Of these developments, we have been experiencing the proliferation of data generation during the technical development and manufacturing of medicines. In the world of technology, we have witnessed the Industry 4.0 revolution (with 5th revolution is around the corner). This involved connected cyber and physical systems, a boom in industrial internet of things and sensory networks towards enabling Smart Manufacturing concepts harnessing the power of artificial intelligence in the form of machine learning/deep learning, process analytical technologies and process control/automation (1, 2). It was then natural to see the convergence of these two revolutions in the biopharmaceutical development and manufacturing (and for wider Operations, hence, sometimes it is also called as Pharma 4.0). We are adapting these powerful technologies such as data and digital, automation, IoT, computational modeling and advanced analytics in advancing molecules to make them into medicines. This convergence is very well suited into achieving the efficiency, fast-to-patients and robust process design objectives across the technical development value chain.
DATA AND DIGITAL STRATEGY ENABLING THE VISION
We are working on a three-tiered digital strategy to achieve our aspirations. These are (i) Seamless Flow of Data and Information, (ii), Insights from Advanced Analytics and Modeling, (iii), Establishing Digital Mindset and Upskilling of our workforce. One of the main focus areas in seamless data and information flow includes capturing the key data across the development activities following FAIR data principles, i.e., Findable, Accessible, Interoperable and Reusable (3). While capturing the scientific data the right way is critical with the right ontologies, and taxonomies in place, the advent of large language models (LLMs) also offer very enabling possibilities to make the data FAIR, which is a trend we will continue to explore how it will evolve in the near future.
PROMISE AND POTENTIAL OF COMPUTATIONAL MODELING
Advances in computational modeling across the technical development phases is well aligned and even expected. For instance, the ICH Q11 Guidance states “Design and conduct studies (e.g., mechanistic and/or kinetic evaluations, multivariate design of experiments, simulations, modeling) to identify and confirm the links and relationships of material attributes and process parameters to drug substance CQAs” (4). Much of the current modeling has its roots in engineering, biology, chemistry, physics and fluid mechanics (mass transfer, kinetics, etc.). In some cases, current scientific understanding and/or analytical resolution is insufficient to utilize mechanistic modeling, therefore we might need hybrid approaches such as data driven and mechanistic modeling with machine learning. In turn, we are also observing the convergence of various disciplines (such as life sciences, physical sciences, engineering and I would add computational/data sciences) to make the most out of our data and systems via advanced computational modeling and techniques (5).
Another emerging and powerful trend is low code platforms and with the power of scalable cloud computing and data storage solutions, these technologies are more accessible than ever to the end users.
Case Studies on How to Get Insights and Value from Data
Accordingly, at Roche, we are making these computational models accessible to scientists and engineers towards achieving right first time, such as scale up for clinical manufacturing and commercial transfer, conduct more targeted experiments, and support regulatory filings.
Case 1: In silico modeling for synthetic molecules drug product development and scale up support
As an example, our teams in synthetic molecules drug product development used an in silico model for roller compaction across scales and equipment. In this case, the model was first calibrated on laboratory scale data, and is then used for scale up to pilot scale. Once calibrated and validated on the scale up data, the model was then used for risk assessment and identification of operating ranges for critical process parameters. Then the model is applied to manufacturing scale applications. The team has standardized this workflow and made it accessible via an app for studying parameters at different scales. This is one of the many examples of using first principles, in silico modeling in technical development.
Case 2: Automation of particle image classification using artificial neural networks
In this case, we are looking at many images generated from a high throughput instrument to classify visible particles in drug products, which are a critical quality attribute of formulations, based on their morphological features. Doing this manually is a very time-consuming task if not infeasible for the whole data set. Our team has trained a convolutional neural network on the particle images to build a classifier for unknown images along with a digital application to assist the scientists with the data management, labeling and re-training procedures. The trained model was capable of distinguishing different particle classes with very high accuracy. This application then translated into significant productivity gains due to automating this complex task.
Case 3: Hybrid machine learning and first principles-based model predictive control (MPC) for improving biopharmaceutical manufacturing performance
In the last example, we have demonstrated an advanced application by combining a first principles modeling approach, machine learning and advanced control techniques to predict and improve cell culture processes. The main goal was to maximize the productivity in production bioreactor of a therapeutic protein production while staying within the designed operating ranges and without adversely impacting quality attributes. Various MPC designs for modulating glucose and feeds were tested and a Gaussian process-based MPC resulted in about 15% increase versus control which was fixed bolus feeds (6). This is a promising application in the spirit of self-adapting processes as part of Smart Biomanufacturing concept in Industry 4.0 paradigm.
CONCLUSION
We have discussed how biotech and tech industry revolutions converged along with the data and digital advances. Our strategy highlighted the seamless flow of data and information, establishing digital mindset and generating insights from the data via prescriptive and predictive modeling. We have emphasized the criticality of the FAIR data, appropriately capturing the technical development data across the development labs and into the manufacturing to enable efficient and faster technical development via more advanced applications such as computational modeling. The three case studies have spanned various aspects of technical development across the value chain showing the significant potential of data and digital to achieve desired business outcomes in advancing molecules into medicines, hence, doing what the patients need next.
ACKNOWLEDGEMENTS
Author would like to thank his Roche colleagues Patrick Piccione and Emmanuela Gavi for Case 1, Frank Zettl and Steffen Sass for Case 2, and also previously published Case 3 for ex-Amgen colleagues as mentioned on the authorship in that reference.
Figure. From data to insights/decisions applying FAIR principles (3)
REFERENCES AND NOTES
- Undey, C, Low, D, Menezes, J, Koch, M, Eds. PAT Applied in Biopharmaceutical Process Development and Manufacturing, 2011, CRC Press, Boca Raton, FL.
- Undey, C, Digital Innovation in Doing What Patients Need Next, The 15th Annual Bioprocessing Summit, Boston, MA, Cambridge Healthtech Institute’s 4th Annual Smart Biomanufacturing & Digitalization: Empowering Smarter Bioprocesses, 16 Aug 2023.
- Wilkinson, M, et al. The FAIR Guiding Principles for scientific data management and stewardship. Sci. Data 2016; 3, 160018.
- US FDA, Guidance for Industry, ICH guideline Q11 on development and manufacture of drug substances. November 2012. (https://www.fda.gov/media/80909/download, Accessed on 10/22/2023)
- Sharp, PA et al., The Third Revolution: The Convergence of the Life Sciences, Physical Sciences, and Engineering, 2011, MIT Whitepaper, (https://www.immagic.com/eLibrary/ARCHIVES/GENERAL/MIT_US/M110103S.pdf, Accessed on 10/22/2023).
- Rashedi, M, et al., Machine learning-based model predictive controller design for cell culture processes, Biotechnology and Bioengineering, 2023; 120(8), 2144-2159.
