Hybrid modeling for in silico optimization of a dynamic perfusion cell culture process

Abstract

The bio-pharmaceutical industry heavily relies on mammalian cells for the production of bio-therapeutic proteins. The complexity of implementing and high cost-of-goods of these processes are currently limiting more widespread patient access. This is driving efforts to enhance cell culture productivity and cost reduction. Upstream process intensification (PI), using perfusion approaches in the seed train and/or the main bioreactor, has shown substantial promise to enhance productivity. However, developing optimal process conditions for perfusion-based processes remain challenging due to resource and time constraints. Model-based optimization offers a solution by systematically screening process parameters like temperature, pH, and culture media to find the optimum conditions in silico. To our knowledge, this is the first experimentally validated model to explain the perfusion dynamics under different operating conditions and scales for process optimization. The hybrid model accurately describes Chinese hamster ovary (CHO) cell culture growth dynamics and a neural network model explains the production of mAb, allowing for optimization of media exchange rates. Results from six perfusion runs in Ambr® 250 demonstrated high accuracy, confirming the model's utility. Further, the implementation of dynamic media exchange rate schedule determined through model-based optimization resulted in 50% increase in volumetric productivity. Additionally, two 5 L-scale experiments validated the model's reliable extrapolation capabilities to large bioreactors. This approach could reduce the number of wet lab experiments needed for culture process optimization, offering a promising avenue for improving productivity, cost-of-goods in bio-pharmaceutical manufacturing, in turn improving patient access to pivotal medicine.

Keywords: CHO cells; digital twin; hybrid models; modeling; optimization; perfusion.

FIGURE 1

Schematic of a perfusion bioreactor.

FIGURE 2

Multi‐layer perceptron neural network.

FIGURE 3

Three representative runs are shown with high, low, and in‐between media exchange rates (~2, ~1, and 1–2 vessel volumes per day [VVD]).

FIGURE 4

Bleed rates for perfusion runs with high, low, and in‐between media exchange rates (~2, ~1, and 1–2 vessel volumes per day [VVD]).

FIGURE 5

Experimental measurements (colored markers) and predicted/simulated data (dashed lines) for training batches with high, low, and in‐between media exchange rates (~2, ~1, and 1–2 vessel volumes per day [VVD]). Predicted and measured (a) viable cell density (VCD), (b) cell viability, (c) simulated lysed cell density, and (d) bio‐material as a function of cell culture time.

FIGURE 6

Experimental measurements (colored markers) and predicted data (dashed lines) for validation batches with high, low, and in‐between media exchange rates (~2, ~1, and 1–2 vessel volumes per day [VVD]). Fitted and measured (a) viable cell density (VCD), (b) cell viability, (c) simulated lysed cell density, and (d) bio‐material as a function of cell culture time.

FIGURE 7

Experimental measurements (colored markers) and predicted/simulated data (dashed lines) for batches V01 and V02 media exchange rates.

FIGURE 8

Plots of observed specific growth rate with Ammonia and Lactate, demonstrating the lack of correlation between these metabolites and growth rate.

FIGURE 9

Plots of observed versus predicted specific growth rate predicted from Equation (7) and the time trajectory of the observed and predicted specific growth rate.

FIGURE 10

Plots of observed specific growth rate versus predicted biomaterial (${\mathrm{\Phi }}_b$) and the time trajectory of the observed specific growth rate and biomaterial.

FIGURE 11

Plots of (a) observed specific death rate versus ammonia, (b) observed specific death rate versus lactate, (c) observed specific death rate versus ammonia for a later manufacturing campaign, and (d) observed specific death rate versus lactate for a later manufacturing campaign.

FIGURE 12

Plots of (a) observed specific death rate with predicted lysed cells and, (b) observed specific death rate with predicted lysed cells for a later manufacturing campaign.

FIGURE 13

Plots of calculated and predicted specific‐productivity for training and validation batches.

FIGURE 14

Normalized experimental measurements (colored markers) and predicted (dashed lines) titer values using a trained NN model for (a) training batches and (b) validation batches with different media exchange rates (between 1 and 2 vessel volumes per day [VVD]).

FIGURE 15

Normalized experimental data (markers) and prediction of titer (dashed lines) using the OPLS model for (a) training batches and (b) validation batches with different media exchange rates (between 1 and 2 vessel volumes per day [VVD]).

FIGURE 16

Correlation between lysed cell density and measured cell diameter.

FIGURE 17

Variable importance plot (VIP) estimated using SHAP values.

FIGURE 18

Volumetric media exchange rates and product concentration for reference and optimized batch trajectories.

FIGURE 19

Trajectories for viable cell density and bleed flow rate for optimized and reference batches.

FIGURE 20

Trajectories for lysed cell density and diameter for optimized and reference batches.

FIGURE 21

Volumetric productivity for optimized and reference batches.

FIGURE 22

Plots of (a) media exchange rate for the training set (B01‐B03) and a later manufacturing campaign matching the optimized feeding (V01) and, (b) observed volumetric production rate.

FIGURE 23

Feed rate and bleed rate for Run 1 and Run 2.

FIGURE 24

Experimental measurements (colored markers) and predicted/simulated data (dashed lines) for 5 L scale perfusion run (Run 1). Predicted and measured (a) viable cell density (VCD), (b) cell viability, as a function of cell culture time.

FIGURE 25

Experimental measurements (colored markers) and predicted/simulated data (dashed lines) for 5 L scale perfusion run (Run 2). Predicted and measured (a) viable cell density (VCD), (b) cell viability, as a function of cell culture time.

FIGURE 26

Experimental measurements (colored markers) and predicted/simulated data (dashed lines) for both 5 L scale perfusion runs (Run 1 and Run 2). Predicted and measured (a) Run 1: viable cell density (VCD), (b) Run 1: cell viability, (c) Run 2: VCD, (d) Run 2: cell viability, as a function of cell culture time.