# Multi-fidelity Gaussian Process for Biomanufacturing Process Modeling with Small Data

**Authors:** Yuan Sun, Winton Nathan-Roberts, Tien Dung Pham, Ellen Otte, Uwe Aickelin  
**arXiv:** [2211.14493](https://arxiv.org/abs/2211.14493)  
**Date:** 2022-11-26  
**Summary page:** https://www.summarizepaper.com/en/paper/2211.14493v1/

## Key points

- Developing an accurate model to simulate bioprocess dynamics is crucial in biomanufacturing
- Challenges include uncertainty, high data acquisition costs, and limited availability of data
- The authors propose using a statistical machine learning approach called multi-fidelity Gaussian process for process modeling
- Multi-fidelity Gaussian process combines advantages of Gaussian process regression and multi-fidelity techniques
- It can effectively model bioprocesses with limited data by leveraging multiple sources of information with different levels of fidelity
- Applied to solve two significant problems: bioreactor scale-up and knowledge transfer across cell lines
- Accurate simulation models are essential for achieving a digital twin and improving efficiency in bioprocesses
- Traditional methods for bioprocess modeling include mechanistic, data-driven, and hybrid approaches
- Mechanistic methods require extensive knowledge but may not capture all complexities accurately
- Data-driven methods struggle with limited data availability and high uncertainty
- Hybrid methods combine strengths but require deep understanding and parameter tuning
- Multi-fidelity Gaussian process offers a promising alternative by incorporating multiple sources of information with different levels of fidelity
- Effective in solving bioreactor scale-up and knowledge transfer problems in biomanufacturing
- Presents a valuable tool for improving accuracy, efficiency, and cost-effectiveness in bioprocess modeling
- Has the potential to advance bio manufacturing by enabling better process control and optimization strategies based on accurate simulation models.

## Summary

In biomanufacturing, developing an accurate model to simulate the complex dynamics of bioprocesses is a crucial task. However, this task is challenging due to the uncertainty associated with bioprocesses, high data acquisition costs, and limited availability of data to learn complex relations in bioprocesses. To address these challenges, the authors propose using a statistical machine learning approach called multi-fidelity Gaussian process for process modeling in biomanufacturing. The multi-fidelity Gaussian process combines the advantages of Gaussian process regression and multi-fidelity techniques. Gaussian process regression is a well-established technique that considers uncertainty in datasets through Gaussian noise. On the other hand, multi-fidelity techniques leverage multiple sources of information with different levels of fidelity, making them suitable for modeling bioprocesses with limited data. The authors apply the multi-fidelity Gaussian process to solve two significant problems in biomanufacturing: bioreactor scale-up and knowledge transfer across cell lines. Bioreactor scale-up involves predicting the behavior of a larger-scale bioreactor based on data from a smaller-scale one. Knowledge transfer across cell lines aims to generalize knowledge learned from one cell line to another. The authors demonstrate the efficacy of their approach on real-world datasets. To achieve a digital twin - an accurate simulation model mimicking the dynamics of a biosystem - accurate simulation models are essential for improving the efficiency of bioprocesses and meeting increasing demands for bioproducts. These models can help identify optimal operating conditions for process control and design effective feed strategies for fed-batch cell culture. Traditional methods for bioprocess modeling can be categorized into mechanistic, data-driven, and hybrid approaches. Mechanistic methods rely on physics-based models that use differential equations based on time-dependent mass balances of components in the biosystem; however they require extensive knowledge about underlying biological processes and may not capture all complexities accurately. Data driven methods learn models directly from data without explicit knowledge of underlying mechanisms but may struggle with limited data availability and high uncertainty; hybrid methods combine mechanistic and data driven approaches to leverage strengths from both but require deep understanding of system as well as extensive parameter tuning . In contrast, the proposed multi fidelity Gaussian process offers a promising alternative for bioprocess modeling with small data by incorporating multiple sources of information with different levels of fidelity which can effectively capture complex dynamics while considering uncertainty . The authors demonstrate its effectiveness in solving two important problems in biomanufacturing:bioreactor scale up and knowledge transfer across cell lines . Overall ,the multi fidelity Gaussian process presents a valuable tool for improving bioprocess modeling in terms accuracy ,efficiency ,and cost effectiveness . It has potential to advance bio manufacturing by enabling better process control and optimization strategies based on accurate simulation models .