Seminar by Wei Xie

Multi-scale Bioprocess Hybrid Modeling, Mechanism Learning, and Optimization

Driven by the critical challenges faced by biopharmaceutical manufacturing, including high complexity, high uncertainty, and very limited process data, we create multi-scale bioprocess knowledge graph (KG) hybrid (“mechanistic + machine learning”) model with modular design characterizing causal interdependencies from molecular to cellular to macro-kinetics. It can facilitate the integration of heterogeneous online and offline measurements collected from different production processes and speed up manufacturing process development. Then, to facilitate bioprocess mechanism learning and optimization, we introduce a model-based reinforcement learning (RL) scheme on the Bayesian KG, accounting for inherent stochasticity and model estimation uncertainty, that can provide an insightful prediction on how the effect of inputs propagates through bioprocess mechanism pathways and impacts on the outputs. It can selectively reuse the most related historical observations and find optimal control policies that are interpretable and robust against model risk and overcome the key challenges of biopharmaceutical manufacturing.