# Interpretable Biomanufacturing Process Risk and Sensitivity Analyses for Quality-by-Design and Stability Control

**Authors:** Wei Xie, Bo Wang, Cheng Li, Dongming Xie, Jared Auclair  
**arXiv:** [1909.04261](https://arxiv.org/abs/1909.04261)  
**Date:** 2019-09-10  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/1909.04261v4/

## Key points

- Biomanufacturing plays a crucial role in supporting the economy and ensuring public health
- Challenges faced in biomanufacturing include complexity, high variability, lengthy lead time, and limited process data
- The authors propose an interpretable semantic bioprocess probabilistic knowledge graph to address these challenges
- A game theory-based risk and sensitivity analysis is developed for production processes to facilitate quality-by-design and stability control
- Causal relationships and interactions of critical process parameters and quality attributes are explored using a Bayesian network-based probabilistic knowledge graph
- A Shapley value-based sensitivity analysis accurately quantifies the variation contribution from each input factor on outputs such as productivity and product quality
- Model uncertainty is quantified using a Bayesian posterior distribution derived from limited process observations
- The Shapley value-based sensitivity analysis evaluates the impact of estimation uncertainty from each set of model coefficients
- The proposed bioprocess risk and sensitivity analyses help identify bottlenecks, guide reliable process specifications, determine informative data collection strategies, and improve production stability
- These analyses provide valuable insights into understanding and managing risks associated with biomanufacturing processes
- They optimize production processes by identifying critical factors that influence productivity and product quality
- Decision-makers can make informed choices regarding process specifications based on reliable data collection strategies
- This research enhances the stability and reliability of biomanufacturing processes while ensuring consistent product quality.

## Summary

Biomanufacturing plays a crucial role in supporting the economy and ensuring public health. However, it faces several challenges including complexity, high variability, lengthy lead time and limited process data, especially for personalized new cell and gene biotherapeutics. To address these challenges, the authors propose an interpretable semantic bioprocess probabilistic knowledge graph and develop a game theory-based risk and sensitivity analysis for production processes to facilitate quality-by-design and stability control. The authors first explore the causal relationships and interactions of critical process parameters and quality attributes (CPPs/CQAs) by creating a Bayesian network-based probabilistic knowledge graph. This graph characterizes the complex causal interdependencies of all factors involved in the bioprocess. By leveraging this graph, they introduce a Shapley value-based sensitivity analysis that accurately quantifies the variation contribution from each input factor on the outputs such as productivity and product quality. Since the coefficients of the bioprocess model are learned from limited process observations, the authors derive a Bayesian posterior distribution to quantify model uncertainty. They further develop the Shapley value-based sensitivity analysis to evaluate the impact of estimation uncertainty from each set of model coefficients. This approach allows for identifying bottlenecks in the bioprocess, guiding reliable process specifications, determining informative data collection strategies, and improving production stability. The proposed bioprocess risk and sensitivity analyses provide valuable insights into understanding and managing risks associated with biomanufacturing processes. These analyses can help optimize production processes by identifying critical factors that influence productivity and product quality. Additionally, they enable decision-makers to make informed choices regarding process specifications based on reliable data collection strategies. Overall, this research contributes to enhancing the stability and reliability of biomanufacturing processes while ensuring consistent product quality.