Integrated process model for the prediction of biopharmaceutical manufacturing chromatography and adjustment steps

Image‐based hybrid model incorporating initial spatial distribution for mesenchymal stem cell cultivation process design

“…Compared to the conventional kinetic models such as the Monod equation, which assumes a homogeneous culture environment, 43 the developed model quantifies the initial distribution and incorporates the spatial heterogeneity into the kinetic model. In addition, the process variation propagated from seeding to cultivation can be evaluated by mechanistic model integration 44,45 and multi-process hybrid modeling 46 showing good agreement with the experiment (see Figure 5). One alternative way of incorporating seeding heterogeneity is calculating ε using numerical simulation results with a fully mechanistic model.…”

“…Compared to the conventional kinetic models such as the Monod equation, which assumes a homogeneous culture environment, 43 the developed model quantifies the initial distribution and incorporates the spatial heterogeneity into the kinetic model. In addition, the process variation propagated from seeding to cultivation can be evaluated by mechanistic model integration 44,45 and multi‐process hybrid modeling 46 based on the novel seeding heterogeneity indicator. Finally, the image‐based hybrid model used a very simple statistical model with a set of ODEs, which means no huge experimental data set was needed for model development, and process simulation could be conducted in a computationally‐friendly way.…”

Realization of digital twin for dynamic control toward sample variation of ion exchange chromatography in antibody separation

“…This discrepancy might be attributed to the occurrence of multilayer adsorption at high loading density, and variants with protein–protein interaction might tend to elute earlier as a “preshoulder” (Seelinger et al, 2023). While the current model can only describe the case of single‐layer adsorption (Equation 2), other models like multistate adsorption model (Diedrich et al, 2017), multicomponent model (Seelinger et al, 2022), asymmetric activity coefficient (Huuk et al, 2017), hybrid model (Narayanan et al, 2021), and colloidal particle adsorption model (Briskot et al, 2021; Rischawy et al, 2022) can be utilized to address high loading density issues and further enhance the prediction accuracy. Although there was a discrepancy in the model prediction in the first half of elution profile, the collection started at the second half (retention time >40 min) with accurate model prediction (RMSE = 4.6%), resulting in a relatively minor impact on the collection window, yield and acidic variant content prediction.…”

“…This discrepancy might be attributed to the occurrence of multilayer adsorption at high loading density, and variants with protein-protein interaction might tend to elute earlier as a "preshoulder" (Seelinger et al, 2023). While the current model can only describe the case of single-layer adsorption (Equation 2), other models like multistate adsorption model (Diedrich et al, 2017), multicomponent model (Seelinger et al, 2022), asymmetric activity coefficient (Huuk et al, 2017), hybrid model (Narayanan et al, 2021), and colloidal particle adsorption model (Briskot et al, 2021;Rischawy et al, 2022) The analysis results indicated that when extending the well-fitted model parameters to higher loading density, a re-evaluation is necessary. The reassessment should not only encompass the model accuracy but, more importantly, the impact of the model's accuracy on the products attributes.…”

“…In recent years, with the maturation of the first principle model for IEC, process development based on the mechanistic model has become increasingly common (Babi et al, 2022; Chen, Yao, et al, 2022; Hahn et al, 2015; Osberghaus et al, 2012; Pirrung et al, 2017), which can not only save a significant amount of wet experiments but also yield higher purity and recovery compared to traditional optimization methods such as the design of experiments. Furthermore, mechanistic models have found applications in real‐time process control (Gomis‐Fons et al, 2020; Rischawy et al, 2022, 2023; Tiwari et al, 2023). By connecting the mechanistic models to the input and output of different steps (such as protein capture, polishing, pH adjustment, etc.…”