From lab to computer
To bring your experiment into the computer with ChromX, a suitable mechanistic model must be selected and calibrated. Once calibrated, the mechanistic model is ready to be used in a wide range of development activities including process optimization, process characterization and troubleshooting. To check the model’s predictive capabilities, ChromX allows the model quality to be assessed by computing the parameters’ sensitivities. In addition, an experimental validation can be performed e.g. at the edges of failure.
ChromX is based on mechanistic models describing the physical and biochemical principles in the chromatography column. There is a large variety of models for different separation problems and different levels of complexity. Depending on the application and the set-up, a suitable model must be selected that adequately describes the fluid dynamic and thermodynamic effects inside the chromatography column.
Model selection for chromatography simulation consists of three consecutive steps:
- Select a column model describing the fluid dynamic effects inside the chromatography column. ChromX offers different models for axial flow and radial flow columns with different geometrical complexities for membranes, monoliths, fibers and bead-based resins.
- Select a pore model describing the diffusion effects in the pores of a particle-based adsorber or in the hydrogel of a membrane capsule.
- Select an adsorption isotherm to describe the interaction of the molecules and ions with the surface of your adsorber material.
To simplify model selection, ChromX offers a Model Selection Wizard that guides the user through the decision-making process.
The chosen models describe the underlying fundamental dynamics of a chromatography system by using partial differential equations. Several unknown model parameters are inherent in these equations. The model calibration step describes the adaptation of these equations’ parameters to a specific chromatography process.
To calibrate the model, a small set of lab experiments is needed, including runs with non-interacting tracer substances to characterize the chromatography skid and column, and experiments to investigate protein-ligand interactions. The number of experiments is dependent on the desired model complexity. Usually about four to eight experiments are needed to investigate protein-ligand interactions. Additionally, up to three experiments can be performed to characterize the column to increase the resulting model quality. Depending on your application, we also recommended to perform additional offline analytics for the different proteins of interest.
Via inverse peak fitting, ChromX estimates all unknown column, adsorber and isotherm parameters based on the experimental data. The isotherm parameters describing the protein-ligand interaction and the competition for binding sites are estimated simultaneously for all protein species based on the recorded UV signals. Purified protein samples are not required for this approach.
Quality assessment and quantitative measurement of the models’ predictive capabilities are key aspects for applying models in a regulated environment. But even in unregulated settings, high model quality is imperative to any model application.
ChromX offers different complementary approaches for quantitatively evaluating the model quality:
- Visual fit. During model calibration you can already see improvement in how the model fits to the experimental data. This visualization gives the user a direct and intuitive measure of the model quality.
- Confidence intervals. Once calibrated, confidence intervals of the model parameters can be used to compute the remaining uncertainty of the simulation and to identify an inappropriate model selection.
- Parameter correlations. After the calibration, parameter correlations of the different model parameters can also be used to indicate a poor model calibration.
All three aspects combined – visual fit, confidence intervals and parameter correlations – can be used to assess model quality, quantify the predictive capabilities of the model and identify potential for improving the model. Where necessary, improvements can be achieved by reducing or increasing the model complexity or extending the set of calibration experiments.
A fourth measure which is frequently used to check the model quality is to execute experimental model validation. In this, the in silico optimized process setpoint and several conditions at the edges of failure are reproduced in the lab and compared to the computed predictions.