Virtual Bioreactor Simulation for Metabolite Balance and Osmolarity Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Controlling and optimizing the complex interplay of major metabolites in cell culture bioreactors is challenging, as fluctuations in metabolites like glucose, glutamine, and lactate can negatively affect cell culture quality, making it difficult to achieve optimal culturing conditions.
Innovation Solution
A virtual bioreactor simulation apparatus using machine learning to predict and analyze the interaction between influence and injection variables, allowing for the derivation of optimal cell culturing conditions by simulating future values of important variables and adjusting injection variables to meet predetermined quality attribute conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If all major metabolites (glucose, glutamine, glutamate) are optimized to be depleted, then cell growth and product quality improve, but osmolarity increases and negatively affects cells
Solution Approach 1:
The system performs preliminary prediction of future metabolite concentrations and osmolarity using machine learning models before actual injection occurs. By simulating future states and predicting outcomes, the system can adjust injection variables in advance to achieve desired metabolite balance while preventing harmful osmolarity increases, rather than reacting after problems occur.
Solution Approach 2:
The system implements closed-loop feedback control where actual metabolite measurements are continuously compared with predicted values, and injection variables are adjusted based on prediction errors. This feedback mechanism enables real-time optimization of metabolite balance while maintaining osmolarity within safe ranges, resolving the contradiction between achieving depletion and avoiding harmful accumulation.
2Reliability
If excessive amounts of glucose, glutamine, and glutamate are injected through feed, then metabolite depletion is prevented, but osmolarity increases and has negative effects on cells
Solution Approach 1:
The system predicts future metabolite concentrations and osmolarity values before injection decisions are made. By using machine learning models to simulate upcoming states, the system can determine optimal injection amounts that prevent metabolite depletion while keeping osmolarity within safe limits, avoiding the need for excessive injection.
Solution Approach 2:
The system dynamically adjusts injection variables (amounts of glucose, glutamine, glutamate) based on real-time predictions and actual measurements. By continuously optimizing these parameters using prediction error feedback, the system maintains metabolite stability without causing harmful osmolarity increases, adapting to changing culture conditions.
3Measurement precision
If machine learning models are trained with large datasets from multiple bioreactors, then prediction accuracy improves, but data heterogeneity and distribution differences increase
Solution Approach 1:
The system divides the training process into two stages: first training individual bioreactor-specific models on their own data to capture unique characteristics, then training a meta-model on the predictions from multiple bioreactor models. This segmentation approach allows the system to leverage data from multiple bioreactors while preserving the unique data distribution characteristics of each bioreactor, improving overall prediction accuracy without suffering from data heterogeneity.
Solution Approach 2:
The meta-model acts as an intermediary that learns from the predictions of multiple bioreactor-specific models rather than directly from raw sensor data. This intermediate layer harmonizes the outputs from different bioreactors with varying data distributions, enabling the system to benefit from large datasets while maintaining prediction accuracy across heterogeneous data sources.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention relates to an apparatus for determining a cell culturing condition and a method of operating the same. The method includes acquiring a first quality attribute and a first influence variable at a first point in time from a cell culture medium, acquiring a first injection variable, applying at least one of the first quality attribute, the first influence variable, and the first injection variable to a culture simulation model and acquiring a second quality attribute and a second influence variable at a second point in time, and selectively changing the first injection variables so that the second quality attribute meets a predetermined condition.