Data Infrastructure for Biomanufacturing Process Control

Summary

According to the National Science and Technology Council (NSTC) report on Strategy for American Leadership in Advanced Manufacturing¹ and recent industrial roadmaps² , Continuous Manufacturing (CM) of biopharmaceuticals is a priority objective to ensure a more responsive domestic manufacturing capability of medical and bio-industrial products.

Biopharmaceutical products such as vaccines are typically manufactured in large batches. In batch manufacturing, qualification of the product is performed in a large quantity against each batch; therefore, any problems in raw materials or processing can scrap the entire batches or result in costly recalls. While CM promises to transform biopharmaceutical manufacturing into small, more uniform, more efficient, and more adjustable drug production, it requires integration of manufacturing process elements and employment of dynamic and holistic control strategies that are able to continuously regulate process parameters to account for variation is raw materials properties as they flow through the process. Moreover, the integration of manufacturing process elements and the uninterrupted flow of material between unit operations mandate that the unit operations be holistically optimized together. Without an innovation, such a requirement significantly increases the experimental burden in the process development phase.

Digital twins have the potential to reduce the experimental burden and ensure a more optimized process design. Moreover, digital twins can be a basis for a holistic control architecture that would encompass the entire process. Building a digital twin requires access to information from various sources, such as historical data, clinical data, and process data.

Also, dynamic control strategies are reliant on real-time data from a variety of manufacturing equipment and sensors to predict the critical quality attributes and control the process parameters in the order of milliseconds. In other words, to advance process control and design of CM, there is a need for ubiquitous access to research and experimental data across software systems (e.g., Lab Information Management Systems - LIMS, Electronic Lab Notebook - ELN) and to data collected from equipment and sensors during the manufacturing run. The data must also be connected to form a digital thread that scientists and engineers can easily explore and analyze. This demand for reliable and efficient access to data across the manufacturing enterprise is heightened when elements of control or digital twins are based on artificial intelligence (AI).

These needs are also evident in the 2019 National Institute for Innovation in Manufacturing Biopharmaceuticals (NIIMBL) workshop report, which indicates that the data standardization and contextualization around raw materials, equipment, and process are among the critical areas needed to achieve the CM for biopharmaceuticals³.

---

 ¹https://www.manufacturing.gov/sites/default/files/2021-06/Advanced-Manufacturing-Strategic-Plan-2018.pdf  

²Sheng Lin-Gibson and Vijay Srinivasan, Recent Industrial Roadmaps to Enable Smart Manufacturing of Biopharmaceuticals, IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 18, NO. 1, JANUARY 2021.

³https://www.niimbl.org/Downloads/Roadmap_Report_November_2019.pdf

Description

Objective
The objective of this project is to develop measurement science that includes ontology models, ontology expression (axiomatic) patterns, guidelines for qualifying those ontologies, and distributed ontology development software that will allow U.S.  manufacturers to have better access to experimental, process, and analytical model data necessary for enabling continuous manufacturing of bioproducts.

Technical Idea
NIST has been a leading organization in the Industrial Ontology Foundry (IOF) community to develop reference ontologies for digital manufacturing. To develop a high-quality and consistent manufacturing ontology for the biopharmaceutical industry and beyond, the technical idea is to use the Basic Formal Ontology (BFO), which has been successfully used as the top-level ontology for the biomedical domain, as a basis. From there, a hub-and-spoke architecture to derive increasingly domain-specific ontologies will be used. In such an architecture, outer ontologies are formally and logically defined based on inner ontologies. Several ontology development principles are also available from prior large ontology efforts such as the Open Biomedical Ontology (OBO). Other technical ideas for developing domain ontology includes ontologizing based on existing industry standards such as the ISA-88 and ISA-95 and defining ontology quality metrics.
Finally, to develop and use a hybrid or data-driven analytical model better and faster, another technical idea is to develop a Machine Learning Lifecycle Ontology (MLLO). MLLO will provide a standard way to share ML model lineage from dataset development to model development, and use phases. This can allow better understanding, comparison, and reuse of ML pipelines. An integration of a MLLO knowledge base with a domain knowledge base may also speed up the model development.

Research Areas
Smart Manufacturing and the Infrastructure for AI are identified as two of the Critical and Emerging Technologies (CETs) in the most recent NSTC report released in February 2022. The NSTC’s report on Strategy for American Leadership in Advanced Manufacturing identifies three actions necessary to realize these two CETs including 1) facilitate a digital transformation in the manufacturing sector; 2) develop standards that enable seamless integration between smart manufacturing components; and 3) develop best practices, new approaches, and standards to provide consistent and reliable access to manufacturing data within and across industries. The research areas in this project that directly address these three actions include:  

• Developing a refined hub-and-spoke ontology architecture for the entire manufacturing domain and validating it with the biomanufacturing subsectors including biopharmaceutical and bio-industrial subsectors;
• Developing and refactoring industrial ontologies in biomanufacturing into the hub-and-spoke architecture such that they can provide a basis for other manufacturing sectors as well;
• Developing ontology quality metrics to increase the reliability of the ontologies;
• Developing processes, guidelines, and software infrastructure to provide efficient and repeatable collaborative, open-source ontology development; and
• Developing a Machine Learning Lifecycle Ontology, associated framework, process knowledge base, and software tool for computer-aided ML model development, analysis, and reuse.

Highlights

• Draft biopharmaceutical ontology available at the IOF GitHub Repo. Select the ‘biopharma’ branch or similar.
• IOF Ontology releases.
• Publication about the state of the art in continuous biopharmaceutical manufacturing.
• Publication about the IOF Core Ontology.
• Publication about an example utilization of the IOF Core ontology for a biomanufacturing application.