Digital Twin Calibration for Biological System-of-Systems: Cell Culture Manufacturing Process (2405.03913v2)

Abstract: Biomanufacturing innovation relies on an efficient Design of Experiments (DoEs) to optimize processes and product quality. Traditional DoE methods, ignoring the underlying bioprocessing mechanisms, often suffer from a lack of interpretability and sample efficiency. This limitation motivates us to create a new optimal learning approach for digital twin model calibration. In this study, we consider the cell culture process multi-scale mechanistic model, also known as Biological System-of-Systems (Bio-SoS). This model with a modular design, composed of sub-models, allows us to integrate data across various production processes. To calibrate the Bio-SoS digital twin, we evaluate the mean squared error of model prediction and develop a computational approach to quantify the impact of parameter estimation error of individual sub-models on the prediction accuracy of digital twin, which can guide sample-efficient and interpretable DoEs.

• The paper introduces a novel digital twin calibration method integrating multi-scale mechanistic models with optimized sequential DoE to reduce prediction errors.
• The calibration approach employs maximum likelihood estimation and bootstrap techniques alongside gradient-based optimization to refine model fidelity.
• Empirical results using a CHO cell culture model demonstrate significant improvements in mAb yield predictions compared to traditional random design methods.

Digital Twin Calibration for Biological System-of-Systems: Cell Culture Manufacturing Process

Introduction

The paper presents a sophisticated approach to digital twin calibration within the context of Biological System-of-Systems (Bio-SoS) applied to cell culture manufacturing processes. The main focus is on enhancing the fidelity of digital twins through optimized sequential designs of experiments (DoE). The work highlights the integration of multi-scale mechanistic models into digital twins, aiming to bridge the gap between the theoretical modeling and practical experimental data in biomanufacturing.

Bio-SoS Mechanistic Model

The Bio-SoS mechanistic model is fundamental to this research, providing a framework for capturing the complexities of cell culture processes. It comprises various interconnected modules, each representing a different biological scale. The model primarily includes:

1. Single Cell Mechanistic Model: This module focuses on the behavior and interaction of individual cells with their environment.
2. Metabolic Shift Model: Captures transitions in cellular metabolic phases responding to environmental changes and cellular aging.
3. Macro-Kinetic Model: Describes the behavior of a bioreactor system, incorporating the dynamics of various cell populations under different conditions.

These modular representations facilitate the flexible adaptation of the model to different production systems and experimental datasets.

Figure 1: An illustration of the multi-scale mechanistic model for cell culture process and Bio-SoS.

Digital Twin Calibration Approach

The paper introduces a novel approach to calibrating digital twins, utilizing the Maximum Likelihood Estimation (MLE) coupled with bootstrap techniques to assess parameter uncertainty. The objective is to match digital twin predictions closely with physical outcomes by minimizing prediction errors using mean squared error (MSE) as the calibration criterion.

The calibration process involves:

• Sequential DoE dynamically guiding experiments to refine model fidelity.
• Utilizing a gradient-based optimization to update policy parameters driving the experimental designs.
• Implementing Linear Noise Approximation (LNA) for uncertainty propagation, providing a surrogate model for MSE estimation.

  Figure 2: The procedure illustration of the proposed digital twin calibration approach.

Empirical Study and Model Validation

An empirical paper using a CHO cell culture model illustrates the effectiveness of the proposed method. By incrementally improving parameter estimates through intelligently designed experiments, the digital twin's prediction accuracy for mAb production significantly outperforms traditional random design approaches.

Figure 3: Yield Prediction Error showcasing reduced error in mAb predictions with the proposed calibration method.

Conclusion

This research contributes to both the theoretical and practical aspects of digital twin calibration in biomanufacturing. By introducing a robust calibration strategy, it enhances the accuracy and interpretability of digital twins, promoting efficient bioprocess control. The methodology is extendable to other biological systems, positioning it as a valuable tool for advancing biomanufacturing technologies. The paper underscores the significance of integrating complex mechanistic models with intelligent experimental designs to achieve reliable and actionable predictions in biological systems.