Predictive Cell Culture Control via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in bioreactors for producing biopharmaceutical products lies in maintaining balanced and consistent cellular metabolic concentrations, which is difficult due to the complex, nonlinear nature of the cell culture process, lacking relevant measurements, and insufficient experimental data, making it hard to optimize growth, yield, and product quality control.
Innovation Solution
A dynamic model is built using historical measurements and machine learning to predict future cell culture attributes, enabling a model-predictive controller to adjust inputs such as glucose levels in real-time, thereby optimizing cell culture processes without manual sampling or adjusting control set-points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If manual or PID controllers are used to control nutrient levels, then the control process is simple to implement, but the manufacturing precision and productivity are limited
Solution Approach 1:
The patent replaces traditional mechanical control systems (manual controllers and PID controllers) with a data-driven machine learning model. The system uses historical and real-time measurements fed into a trained model that automatically predicts optimal control actions, substituting human-operated mechanical systems with an intelligent computational system that achieves superior manufacturing precision while maintaining ease of operation.
Solution Approach 2:
The system dynamically changes control parameters (nutrient feed rates, environmental conditions) based on predictions from the machine learning model. Rather than using fixed control strategies, the system continuously adapts parameters based on real-time measurements and historical data, enabling precise control of cell culture attributes while responding to dynamic process conditions.
2Reliability
If mechanistic mathematical models are used to model bioprocesses, then the model can capture process dynamics, but the device complexity and difficulty of detecting and measuring increase
Solution Approach 1:
The patent substitutes complex mechanistic mathematical models with a data-driven machine learning model. Instead of using first-principles equations that require extensive process knowledge and complex hybrid modeling, the system trains a model on historical measurements, allowing it to capture process dynamics empirically. This reduces device complexity and measurement requirements while maintaining or improving modeling reliability.
Solution Approach 2:
The machine learning model learns process dynamics autonomously from historical data without requiring explicit mechanistic knowledge or manual model construction. The system serves itself by automatically adapting to process variations through continuous learning from measurements, eliminating the need for complex model development and maintenance that characterizes mechanistic approaches.
3Manufacturing precision
If extensive process knowledge is applied to create mechanistic models, then the model accuracy improves, but the ease of manufacture and adaptability decrease
Solution Approach 1:
The patent implements a dynamic machine learning model that continuously adapts to changing process conditions through real-time measurements. Unlike static mechanistic models that require complete process knowledge to handle variations, this system dynamically adjusts its predictions based on current measurements and historical patterns, maintaining high accuracy while being highly adaptable to new conditions without requiring extensive process knowledge modifications.
Solution Approach 2:
The machine learning model serves multiple functions: it predicts future cell culture attributes, identifies optimal control actions, and adapts to different process conditions all within a single unified framework. This universal approach replaces the need for multiple specialized mechanistic models, improving both manufacturing precision across different scenarios and adaptability to new conditions without requiring extensive domain expertise for each specific application.
Data Source
AI summary
A method of controlling a cell culture process includes, for each time interval of one or more time intervals during the cell culture process, obtaining current values of one or more cell culture attributes associated with a cell culture, predicting one or more future values of a particular cell culture attribute associated with the cell culture, and controlling one or more physical inputs to the cell culture process. Predicting the future value(s) includes applying the current values of the cell culture attribute(s), and an earlier value of at least one of the cell culture attributes, as inputs to a data-driven predictive model using historical data. Controlling the physical input(s) includes applying the future value(s) as inputs to a model predictive controller.