Chromatography Process Simulation With ML-Predicted Binding and Diffusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current chromatography process simulation methods are complex, requiring extensive experimental time and expertise due to detailed physicochemical modeling, and are not well-suited for modern chromatography columns with complex properties, limiting their practical application and scalability.
Innovation Solution
A hybrid model combining a simple physics-based mass balance model with a machine learning model to predict binding and diffusion parameters, allowing for efficient simulation of chromatography processes across different geometries and conditions, using ordinary differential equations and discrete volume elements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If detailed first-principles mechanistic models are used to describe chromatography processes, then prediction accuracy under varying conditions is improved, but model complexity and experimental time requirements increase significantly
Solution Approach 1:
The patent introduces machine learning models as intermediary components that bridge the gap between simple mass balance equations and complex mechanistic models. The ML models predict the kinetic parameters (k_a, k_d, K_D) that capture binding and diffusion behavior without requiring explicit physical chemistry formulations, thus achieving accurate predictions while reducing model complexity
Solution Approach 2:
The patent replaces complex physics-based mechanistic modeling with a data-driven machine learning approach. Instead of deriving and solving detailed partial differential equations describing mass transfer and binding mechanisms, the system uses ML models trained on experimental data to directly predict chromatography behavior, substituting mechanical/physical modeling with computational learning
2Reliability
If first-principles models with multiple physical effects are used, then predictive capability is improved, but the time required for coefficient identification through experimentation increases
Solution Approach 1:
The patent performs preliminary action by training machine learning models on experimental chromatography data before actual process simulation. The ML models are pre-trained to capture binding and diffusion characteristics, so that when predicting new processes, the system only needs to input process conditions rather than performing time-consuming coefficient identification experiments
Solution Approach 2:
The patent creates a computational copy of experimental chromatography behavior through machine learning models. Instead of physically conducting extensive experiments to identify model coefficients, the system trains ML models on existing experimental data to create a virtual representation of chromatography behavior, which can then be used for prediction without additional experimental time
3Measurement precision
If complex mechanistic models describing all physical interactions are used, then accuracy is improved, but the requirement for prior expertise and characterization knowledge increases
Solution Approach 1:
The patent replaces expensive, time-consuming expert modeling efforts with more economical machine learning approaches. The ML models are trained on experimental data and can be applied to multiple different chromatography systems without requiring expert recalibration or re-characterization for each new application, reducing the barrier to accurate modeling
Solution Approach 2:
The patent changes the approach from modeling all physical interactions explicitly to predicting key behavioral parameters (binding rate k_a, dissociation rate k_d, equilibrium constant K_D) through machine learning. This parameter transformation simplifies the model construction process while maintaining predictive accuracy, as the ML models automatically capture the net effect of complex physical interactions
4Productivity
If simple mass balance models are used, then computational efficiency is improved, but the ability to capture binding and diffusion phenomena is reduced
Solution Approach 1:
The patent merges the simplicity of mass balance equations with the predictive power of machine learning models. The system combines ordinary differential equations describing bulk flow with ML-based parameter predictions for binding and diffusion, creating a hybrid model that maintains computational efficiency while capturing essential chromatography phenomena through the ML-predicted parameters
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The hybrid model provides a stable, computationally efficient simulation that is independent of column geometry and stationary phase type, enabling rapid and accurate predictions of chromatography performance, facilitating process optimization, monitoring, and scale-up without the need for detailed physicochemical characterization.
Implementation Method 1
a binding term captures the binding of a product to the stationary phase
Implementation Method 2
a diffusion term captures the diffusion of a bound product
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
The present disclosure provides methods of simulating a chromatography process, methods of monitoring a chromatography process, methods of controlling a chromatography process, and associated systems and products, which use a bulk flow mass balance model solved over one or more discrete volume elements along the chromatography unit by numerical integration, the model being an ordinary differential equations model comprising binding and diffusion terms each parameterised by a single respective parameter that is predicted by a machine learning model.