Interoperable AI Biomanufacturing Systems
MIE Assistant Professor Wei Xie, in collaboration with Ohio University and IMPSystems, is leading a $244,235 National Institute of Standards and Technology grant, “Biomanufacturing Modeling and Machine Learning Ontology Platform.” The research will address manufacturing systems integration, interoperability, and standardization development.
Machine learning (ML), including data-driven decision-making and predictive modeling, is playing a more critical role in production process development and automation in the biopharmaceutical manufacturing marketing, which is currently challenged by a number of issues, including inherent complexities, limited data, and needed improvements in time to market, especially for personalized gene/cell therapy.
Ontologizing ML development and deployment in the biomanufacturing industry can facilitate manufacturing systems integration, interoperability, and standardization development. Considering new bio-tech products and advanced sensors, which can provide real-time monitoring at single molecular and cellular scale) come to the market more frequently, this project will pave the way for ML lifecycle ontology which models the digital thread of model development and deployment along with the representation of associated data and its evolution through the ML pipeline.
The multi-scale knowledge graph hybrid modeling with a modular design, developed by Xie’s research team, enables systematically configurable digital implementations. This can represent physical processes with models and simulations, and facilitate informational transformations through standard interfaces. Representing biomanufacturing process knowledge, control strategies, actual occurrences, and observations in coherent ontologies can aid humans and computers working together on both real and digital systems to handle the complexity of manufacturing systems and the challenges induced by biotech with fast evolution. Then the ontologically encoded probabilistic knowledge graph and machine learning are informed and updated with heterogeneous online and offline data collected from distributed real and digital systems. With domain-specific ontologies as the backbone, it can provide logical rigor, create a standard queryable representation, and serve as a reasoning engine for root-cause analysis of both real and digital systems.
The proposed modeling and ML ontology platform has bottom-up and top-down designs for the integration and interoperability of distributed production systems. This platform allows us to integrate production processes with different inputs, control strategies, and online/offline measurements collected at molecular, cellular, and macroscopic levels. It also enables the development of standards to support consistent data representation in biomanufacturing systems modeling and simulation, and provides validation and verification for data integration and information exchange.