The integration of artificial intelligence (AI) and machine learning (ML) in biopharmaceutical manufacturing holds the potential to revolutionize the industry by optimizing processes, enhancing product quality, and reducing costs. Recent FDA initiatives underscore the importance of AI-ML in advancing pharmaceutical quality and efficiency, promoting the adoption of these technologies while addressing regulatory challenges.
This presentation explores the latest advancements in AI-ML applications, particularly focusing on their roles in process design, monitoring, and control. A key highlight is the exploration of hybrid models, which combine mechanistic models with data-driven ML approaches. These models leverage the strengths of both paradigms, offering robust and flexible solutions for complex biopharmaceutical processes. By integrating fundamental process knowledge with real-time data analytics, hybrid models enhance process understanding, control, and optimization, paving the way for more reliable and scalable biopharmaceutical manufacturing.
In addition to hybrid models, the presentation will discuss the application of physics-informed neural networks (PINNs). PINNs integrate physical laws directly into neural network training, allowing for the incorporation of multivariate external process parameter variations such as temperature, seeding cell density, and feeding rates into kinetic growth equations. The use of Taylor series expansions within PINNs enables high prediction accuracy for key process variables such as cell densities and concentrations, providing continuous and differentiable functions for process outcomes.
To illustrate these advancements, two case studies will be presented:
Application of Hybrid Models in Downstream Processes: This case study focuses on the commercial production of biopharmaceuticals, aiming to predict critical quality attributes such as yield, host cell protein, DNA, and aggregates.
Application of Physics-Informed Neural Networks in Upstream Processes: This case study examines the combination of PINNs with kinetic growth equations to optimize upstream processes for commercial biopharmaceutical production. The approach demonstrates significant accuracy in predicting viable and total cell densities, product concentration, and nutrient levels under varying process conditions, thereby enabling model-driven optimization and significant experimental savings.

Shengchun Guo

Appearances