(377b) Optimizing Biopharmaceutical Production: Integrating Data Pipelines, Kinetic Models, Full Cell Culture, and LLM-Powered Data Analysis in Virtual Bioreactor

Bioreactors are central to the production of biopharmaceuticals, where optimizing the culture environment is key to ensuring product quality, consistency, and scalability. To support this, Amgen, in collaboration with M-Star, has developed a virtual bioreactor model that combines advanced computational fluid dynamics (CFD) with reduced-order biochemical kinetics. This in silico platform predicts critical parameters such as mass transfer, pH, and titer while offering spatial insights into hydrodynamic behaviors across various bioreactor scales.

Building upon the initial model, recent developments now enable simulation of the full duration of a cell culture process. A refined kinetic framework tracks the dynamic interactions and transport of glucose, glutamine, lactate, ammonium, viable cell density (VCD), and titer throughout the culture period. These enhancements are embedded in a Python-enabled automated workflow, improving model preprocessing, ensuring reproducibility, and significantly reducing manual intervention.

Additionally, a newly integrated data analysis and visualization interface—powered by large language models (LLMs) and embedded within Genie Databricks—enables natural language querying of our virtual bioreactor simulation data on Enterprise Data Fabric (EDF). This capability makes exploration of complex bioprocess datasets more intuitive, democratizing access to insights and accelerating decision-making.

This holistic virtual bioreactor framework is already being applied across multiple scales, from early development to commercial manufacturing. It enhances tech transfer, supports scale-up activities, and reduces experimental burden, ultimately contributing to a faster, more reliable development pipeline for biopharmaceutical products.