Bioprocess Quality Prediction Using State-Space and Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current bioprocess monitoring and control methods are inadequate for real-time prediction of critical quality attributes, leading to delayed quality determination and inability to take corrective actions during the process due to slow offline measurements.
Innovation Solution
A method using a machine learning model trained with state variables and process conditions from a state space model to predict critical quality attributes in real-time, allowing for immediate corrective actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If offline measurement methods are used to determine critical quality attributes, then measurement accuracy can be maintained, but the measurement time delay increases significantly
Solution Approach 1:
The system performs preliminary analysis by continuously monitoring process parameters and using trained machine learning models to predict critical quality attributes in real-time during the bioprocess, rather than waiting for offline measurements after process completion. This allows quality assessment to be performed in advance, enabling timely corrective actions.
Solution Approach 2:
The patent introduces an intermediary prediction system that acts as a bridge between process parameters and final quality attributes. The machine learning model serves as an intermediary that translates real-time process data into predicted quality outcomes, avoiding the need for direct slow offline measurements while maintaining accuracy.
2Ease of operation
If statistical process analysis methods are used to identify critical process parameters, then process condition ranges can be established, but the connection between process parameters and quality attributes remains poorly understood
Solution Approach 1:
The system implements feedback by continuously comparing predicted quality attributes against target values and using this information to adjust process parameters in real-time. The trained models provide feedback loops that link process parameters to quality outcomes, enabling dynamic optimization and improving understanding of their relationships.
Solution Approach 2:
The patent employs parameter changes by using trained machine learning models to identify optimal ranges and trajectories for critical process parameters that lead to desired quality attributes. The system dynamically adjusts parameter settings based on real-time predictions, moving from static historical ranges to adaptive optimized parameters.
3Productivity
If real-time prediction of critical quality attributes is implemented, then corrective actions can be taken during the process, but the system complexity increases
Solution Approach 1:
The system achieves universality by using a single integrated machine learning framework that handles multiple quality attribute predictions simultaneously. The trained models can predict various critical quality attributes from the same set of process parameters, reducing the need for separate specialized systems for each quality metric.
Solution Approach 2:
The patent implements self-service through autonomous operation of the prediction system. Once trained, the machine learning models automatically predict quality attributes and trigger corrective actions without requiring manual intervention or complex operator judgment, simplifying the operational complexity despite the advanced technology used.
Data Source
AI summary
Methods for monitoring, controlling, optimising and simulating a bioprocess comprising a cell culture in a bioreactor are provided. The methods comprise obtaining the values of one or more state variables of a state space model at one or more maturities, and predicting the value of one or more critical quality attributes of a product of the bioprocess using a machine learning model trained to predict the value of the one or more critical quality attributes based on input variables comprising values of the one or more state variables or variables derived therefrom, at one or more maturities. The state space model comprises a kinetic growth model representing changes in the state of the cell culture and a material balance model representing changes in the bulk concentration of one or more metabolites in the bioreactor. Systems, computer readable media implementing such methods, and methods for providing tools to implement such methods, are also provided.