Protein A Chromatography Multi-scale Data Driven/Mechanistic Modeling (Phase 2)

Mechanistic modeling functions on the premise that assumptions made about the transport and equilibrium processes of the system describe the system well and can be used to represent the system in most process conditions under a range of industrial scales and resin lifetimes. Such models can be very effective for describing and predicting behavior in response to many changes in structural (e.g., column length or diameter) or operating (flow velocity, pH) conditions, other known but often subtler changes are less easy to characterize or predict. In order to integrate such, potentially important, effects into our models, they will be augmented into hybrid models. One alternative approach is to use an artificial neural network that learns to depict and update the isotherm and transport parameters based on measured column data. This can provide additional degrees of freedom to the isotherm and mass transfer relationships, untied from mechanistic assumptions and enable ongoing model learning from diverse settings and processes in real time. It can also be important in describing the behavior of incompletely characterized impurities. In this project software is developed that will employ both mechanistic and hybrid modeling to predict protein A chromatographic behavior in a variety of settings