NIIMBL Request for Proposals (RFP): Data Driven Model for Predicting CADET Model Parameters

Hello Everyone,

I’m the workstream lead for NIIMBL’s BD5 (Big Data) program for downstream process modeling. For FY2027 we’re drafting a request for proposal (RFP) to implement hybrid parameter fitting and process optimization into CADET. Here’s a short summary of the proposal:

This request for proposal entails the development of a hybrid machine learning–mechanistic modeling framework for automated parameterization and optimization of chromatographic unit operations within the CADET simulation environment. The framework integrates physics-informed neural network architectures to provide rapid initial estimates of mechanistic model parameters — including adsorption equilibrium constants, mass transfer coefficients, and dispersion terms — from either experimental chromatographic data or molecular and resin descriptors. These data-driven parameter estimates serve as initializations for a hybrid refinement loop in which CADET’s forward mechanistic simulator is leveraged iteratively to minimize residuals between predicted and observed chromatograms via gradient-based or gradient-free optimization.

Upon model validation, a weighted multi-objective cost function is employed to optimize controllable process variables. The framework is designed to be modality-agnostic and is intended to close the existing gap between parameter estimation and process optimization in a single, accessible, open-source platform.

The RFP will be open globally in the next week or so and close by July 10th. Don’t hesitate to reach out if you’re interested.

-M. Rosario Cervellere,

rosario.cervellere@milliporesigma.com

Hi Rosario,

thanks for sharing this exciting initiative. The proposed hybrid machine learning/mechanistic modeling framework represents a significant step forward for downstream process modeling.

To better understand the scope, could you clarify how you envision the interaction between this new framework and established tools such as CADET-Process? Specifically, will the hybrid parameter fitting and optimization be implemented as a standalone module, or is direct integration into CADET-Process intended to ensure compatibility with existing workflows? I am also aware of an ongoing PhD project at RWTH working on this topic.

If you are interested, we would welcome the opportunity to discuss this further, e.g., during a call.

Best regards,

Johannes