Cell culture monitoring data management system and method

Abstract

The invention relates to the technical field of cell culture monitoring data management, and particularly discloses a cell culture monitoring data management system and method.Three-dimensional visual monitoring of the culture process is achieved by constructing an initial digital twinborn model and combining real-time monitoring data to generate a dynamic digital twinborn model; meanwhile, three-dimensional imaging data are further fused, a fused three-dimensional model containing a cell structure and biochemical parameters is generated, an optimization strategy is generated based on an intelligent decision algorithm, and management is improved from passive response to active prediction; the cell growth state is comprehensively and visually monitored, the monitoring depth is enhanced, through a virtual environment preview strategy, the actual adjustment risk is reduced, a closed loop feedback mechanism is established, cross-batch learning and self-optimization are supported, meanwhile, man-machine interaction is supported, the experience of an operator is fused into a decision, the adaptability and robustness are improved, and the method is suitable for popularization and application. Finally, the stability and the product quality of the culture process are improved.

Description

Technical Field

[0001] This invention belongs to the field of cell culture monitoring data management technology, and relates to a cell culture monitoring data management system and method. Background Technology

[0002] Cell culture technology is a core component of modern biopharmaceutical, cell therapy, and basic life science research. The stability of the process and the quality of the products directly determine the success or failure of the final product. Monitoring and management systems for the cell culture process are crucial for ensuring that cells obtain a suitable growth environment within bioreactors and other devices. These systems achieve efficient cell proliferation and high-level expression of target products by real-time monitoring and regulation of key parameters during the culture process.

[0003] Currently, existing cell culture monitoring methods mainly rely on various sensors deployed in the culture device to acquire macroscopic physicochemical parameters such as temperature, pH, and dissolved oxygen. These are then combined with offline sampling analysis or online two-dimensional microscopic imaging techniques to assess cell density and viability. Process management typically relies on this discrete, two-dimensional monitoring data, adjusting process parameters such as feed rate and gas flow rate through pre-programmed control logic or manual intervention by operators based on experience.

[0004] However, the aforementioned existing technologies have significant technical shortcomings in practical applications. First, monitoring based on discrete sensors cannot reflect the heterogeneity of the microenvironment inside the culture vessel, and two-dimensional imaging cannot fully capture the true distribution and interaction states of cells in three-dimensional space. Second, the diversity and unstructured nature of data sources lead to severe information silos, making it difficult to form a holistic understanding of the culture process. Finally, control decisions heavily rely on the operator's experience, lacking foresight and optimization, often only able to respond to problems that have already occurred, making it difficult to effectively cope with complex biological process fluctuations, resulting in large batch-to-batch variations and poor process stability and reproducibility. Summary of the Invention

[0005] In view of this, in order to solve the problems mentioned in the background art, a cell culture monitoring data management system and method are proposed.

[0006] The objective of this invention can be achieved through the following technical solution: The first aspect of this invention provides a cell culture monitoring data management system, comprising: an initial digital twin model construction module, which acquires environmental parameters and cell characteristic parameters of cell culture, and constructs an initial digital twin model based on the environmental parameters and cell characteristic parameters.

[0007] The dynamic digital twin model generation module acquires real-time monitoring data and updates the initial digital twin model based on the real-time monitoring data to generate a dynamic digital twin model.

[0008] The fusion 3D model generation module acquires 3D imaging data and fuses it with a dynamic digital twin model to generate a fused 3D model.

[0009] The cell culture optimization strategy generation module, based on the fusion of a 3D model and a dynamic digital twin model, applies intelligent decision-making algorithms to generate cell culture optimization strategies.

[0010] The second aspect of the present invention provides a method for managing cell culture monitoring data, comprising: S1, initial digital twin model construction: acquiring environmental parameters and cell characteristic parameters of cell culture, and constructing an initial digital twin model based on the environmental parameters and cell characteristic parameters.

[0011] S2. Dynamic Digital Twin Model Generation: Acquire real-time monitoring data and update the initial digital twin model based on the real-time monitoring data to generate a dynamic digital twin model.

[0012] S3. Fusion 3D Model Generation: Acquire 3D imaging data and fuse it with a dynamic digital twin model to generate a fused 3D model.

[0013] S4. Generation of cell culture optimization strategies: Based on the fusion of 3D model and dynamic digital twin model, intelligent decision-making algorithm is applied to generate cell culture optimization strategies.

[0014] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention achieves comprehensive and intuitive monitoring of the three-dimensional spatial state of the cell culture process by constructing a dynamic digital twin model and deeply integrating it with three-dimensional visualization technology. This method overcomes the one-sidedness of information brought about by traditional two-dimensional imaging or discrete sensor monitoring, and can display the cell growth distribution, morphology and microenvironment field of key biochemical parameters in real time and three-dimensionally, so that the operator can accurately grasp the real state of the culture system from both macroscopic and microscopic levels, thereby significantly improving the depth and breadth of process monitoring.

[0015] (2) This invention introduces a model-based intelligent decision-making mechanism, elevating cell culture management from a passive response to an active prediction level. By rapidly evaluating and screening multiple optimization strategies in a virtual digital twin environment, the system can safely and efficiently select the optimal control scheme without interfering with the actual culture process. This forward-looking decision-making approach effectively avoids the risks and costs associated with traditional trial-and-error methods relying on human experience, ensuring the scientific rigor and accuracy of each parameter adjustment, thereby enhancing the stability and controllability of the culture process.

[0016] (3) This invention, by establishing a multi-level closed-loop feedback and learning mechanism, endows the system with the ability to self-evolve and continuously optimize. The system can not only make immediate corrections based on real-time data, but also conduct review learning after completing the entire training batch, fundamentally improving the underlying accuracy of the model. At the same time, the human-computer interaction design also enables the operator's expert experience to be effectively integrated into the automated decision-making process, realizing the synergistic effect of machine intelligence and human wisdom, enabling the entire system to continuously adapt to new process changes, and demonstrating strong robustness and adaptability. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the system structure connection of the present invention.

[0019] Figure 2 This is a schematic diagram of the method steps of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 The first aspect of the present invention provides a cell culture monitoring data management system, including: an initial digital twin model construction module, a dynamic digital twin model generation module, a fused three-dimensional model generation module, and a cell culture optimization strategy generation module.

[0022] The initial digital twin model construction module is connected to the dynamic digital twin model generation module, the dynamic digital twin model generation module is connected to the fused 3D model generation module, and both the dynamic digital twin model generation module and the fused 3D model generation module are connected to the cell culture optimization strategy generation module.

[0023] The initial digital twin model construction module acquires environmental parameters and cell characteristic parameters of cell culture, and constructs an initial digital twin model based on these parameters.

[0024] In a specific embodiment of the present invention, the specific steps for constructing an initial digital twin model based on environmental parameters and cell characteristic parameters include: collecting physicochemical information of the cell culture environment through a sensor network to obtain environmental parameters.

[0025] It should be noted that, firstly, a sensor network deployed inside and around the cell culture device collects key physicochemical information about the cell's microenvironment in real time. This key physicochemical information includes, but is not limited to, temperature, pH, and dissolved oxygen parameters. The sensor network converts the collected analog signals into digital signals, which, after filtering and calibration, form a structured set of environmental parameters. This set of environmental parameters not only includes the static values ​​at the initial moment but also reflects the minute fluctuations of these parameters in the initial stage, providing a data foundation for the model's realism.

[0026] Data matching the currently cultured cells is extracted from a database containing various cell biological characteristics to obtain cell characteristic parameters.

[0027] It should be noted that the system retrieves biological characteristic data matching the currently cultured cell line from a pre-set database containing various cell biological characteristics. This database is a knowledge base compiled from a large number of previous experimental studies and publicly available literature. Based on the input cell type identifier, the system obtains its unique cell growth rate parameters and metabolic characteristic parameters, which together constitute a set of cell characteristic parameters. Among these, the cell growth rate parameter is typically defined by a kinetic model describing the relationship between cell proliferation and substrate concentration, while the metabolic characteristic parameters quantify the stoichiometric relationships in the process of cells consuming nutrients and producing metabolic waste.

[0028] Environmental parameters and cell characteristic parameters are input into a computer simulation algorithm to construct an initial digital twin model.

[0029] It should be noted that, finally, the system utilizes computer simulation algorithms to deeply couple the aforementioned set of environmental parameters containing physicochemical information with the set of cell characteristic parameters containing biological principles, thereby constructing an initial digital twin model. The core of this algorithm is to establish a set of partial differential equations to describe the growth and metabolic dynamics of the cell population in three-dimensional space. For example, the local growth rate of cells can be expressed by the following formula: In this formula, The specific growth rate, representing the cell's growth rate, is one of the core variables that the model needs to calculate. This represents the cell's maximum specific growth rate, while These are the substrate affinity constants. Both parameters are obtained from the set of cell characteristic parameters and reflect the inherent biological properties of the cell. This represents the local concentration of a key substrate, such as glucose, and the initial distribution of this value is provided by a set of environmental parameters. and These are dimensionless correction functions, representing the influence factors of temperature and pH on cell growth. Their functional forms are determined based on the optimal growth conditions from the set of cell characteristic parameters, and are calculated in conjunction with real-time temperature and pH parameters from the set of environmental parameters. By solving this set of equations, the system can generate a three-dimensional gridded digital model. Each grid node contains state information such as cell density, substrate concentration, and metabolite concentration. This complete three-dimensional state field is the initial digital twin model.

[0030] It should also be noted that, and The function expressions are as follows: , ,in, Represents the natural constant. and These represent the real-time temperature and real-time pH value in the environmental parameter set, respectively. and These represent the optimal growth temperature and optimal growth pH value from the set of cell characteristic parameters, respectively. and These represent the maximum permissible temperature deviation range and the maximum pH deviation range, respectively.

[0031] The initial digital twin model constructed using the above method is technically effective because it is not a static or universal template, but rather a precise match between the initial physical environment of a specific culture batch and the intrinsic biological characteristics of a specific cell line. This combination ensures that the model possesses a high degree of realism and personalization from moment zero, laying a solid foundation for subsequent model evolution and accurate prediction based on real-time data. It solves the technical problem of accumulated deviations in subsequent prediction results caused by distortion of initial conditions in traditional simulations, significantly improving the reliability and accuracy of the entire digital twin system in monitoring and decision-making throughout the cell culture cycle.

[0032] The dynamic digital twin model generation module acquires real-time monitoring data and updates the initial digital twin model based on the real-time monitoring data to generate a dynamic digital twin model.

[0033] In a specific embodiment of the present invention, the specific steps of updating the initial digital twin model by combining real-time monitoring data to generate a dynamic digital twin model include: acquiring real-time monitoring data reflecting cell population status and environmental changes through an online monitoring system.

[0034] It should be noted that the method for updating the initial digital twin model to a dynamic digital twin model in this invention is a closed-loop correction process that synchronizes the model with reality. This method first acquires real-time monitoring data continuously or periodically through an online monitoring system. This data specifically includes two categories: one is cell density parameters that directly reflect the state of the cell population, which can be obtained through non-invasive sensors such as dielectric spectroscopy probes or optical densitometers; the other is environmental change parameters that describe the dynamic changes in the culture environment, such as real-time values ​​of parameters like temperature, pH, and dissolved oxygen continuously fed back by a sensor network.

[0035] Real-time monitoring data is compared with the predicted data from the initial digital twin model to generate model error parameters.

[0036] A model calibration algorithm is applied, and the internal parameters of the initial digital twin model are adjusted based on the generated model error parameters to generate a dynamic digital twin model.

[0037] It should be noted that after obtaining real-time monitoring data, the system immediately inputs it into the model calibration algorithm to adjust and optimize the initial digital twin model. The core idea of ​​this model calibration algorithm is to compare the deviation between the model's predictions and actual measurements, and use this deviation to correct the model's internal parameters. For example, the algorithm subtracts the cell density value predicted by the initial digital twin model at the current time point from the cell density parameter in the real-time monitoring data to obtain the prediction error. Subsequently, based on this prediction error, the algorithm automatically adjusts key parameters related to cell growth and metabolism in the model, such as maximum specific growth rate and substrate consumption coefficient, using a preset gain function. This adjustment process can be summarized by the following formula: In this formula, This represents the adjusted model parameters generated after calibration. These are the parameters used by the current model. It is real-time monitoring data obtained from sensors, and The model is based on The parameters represent the predicted data for the same point in time. This is the model calibration gain matrix, whose value is dynamically calculated by the algorithm based on the uncertainties of the model and measurement data. Its value typically ranges from 0 to 1. When the model prediction error is large, Approaching 1 to strengthen the correction, when the model prediction error is small, To avoid noise interference, the parameters are adjusted to approach zero, ensuring that both the magnitude and direction of the adjustment minimize the prediction error, and that the dimensions on both sides of the formula are matched. Finally, the system uses this set of optimized and adjusted model parameters to drive the next iteration of the initial digital twin model. This model, continuously corrected and evolved by real-time data, constitutes the dynamic digital twin model. Because its internal parameters have absorbed the latest real-world information, this dynamic digital twin model is no longer a static initial hypothesis, but a vibrant digital mapping that closely follows the changes in the actual culture process. Its most critical capability lies in its ability to make short-term, high-precision predictions of future cell growth trends based on the most accurate current state.

[0038] The technical advantage of this method lies in its fundamental solution to the long-standing technical pain point of the disconnect between simulation models and physical reality. By introducing a closed-loop feedback mechanism of real-time monitoring data and model calibration algorithms, the system achieves a leap from static construction to dynamic evolution of the digital twin model. This continuous self-correction capability ensures that the model maintains high fidelity throughout the entire cell culture cycle, avoiding prediction failures caused by the accumulation of initial errors. Ultimately, a highly reliable dynamic digital twin model synchronized in real time with the physical culture system is generated, providing a solid and reliable data foundation for subsequent accurate visualization analysis and forward-looking intelligent decision-making.

[0039] The fused 3D model generation module acquires 3D imaging data and fuses the 3D imaging data with a dynamic digital twin model to generate a fused 3D model.

[0040] In a specific embodiment of the present invention, the specific steps of fusing three-dimensional imaging data with a dynamic digital twin model to generate a fused three-dimensional model include: scanning the cell culture container with an optical scanning device to obtain three-dimensional imaging data.

[0041] It should be noted that the 3D imaging data is obtained by scanning the 3D space within the cell culture container layer by layer using high-resolution imaging equipment such as laser scanning microscopes or optical computed tomography. Each scan generates a 2D image slice, and by stacking a series of slices and performing 3D reconstruction, the raw 3D imaging data containing information on the precise spatial location, morphology, and aggregation state of the cells is constructed.

[0042] The spatial coordinates of the 3D imaging data and the dynamic digital twin model are aligned to generate spatially aligned data.

[0043] It should be noted that, subsequently, the system performs a crucial data fusion step, deeply integrating the aforementioned 3D imaging data containing real physical information with the generated dynamic digital twin model. This fusion is not a simple image overlay, but a process of multi-dimensional information matching and correction. First, the system uses a spatial registration algorithm to precisely align the coordinate system of the 3D imaging data with the virtual coordinate system of the dynamic digital twin model, ensuring that each 3D pixel has a one-to-one correspondence in both physical and virtual space.

[0044] Interpolation of the information dimensions of the spatially aligned data is performed to generate a fused 3D model.

[0045] In a specific embodiment of the present invention, the specific steps of interpolating the information dimensions of the spatially aligned data to generate a fused three-dimensional model include: identifying the dimensional differences between the structural information provided by the three-dimensional imaging data and the functional information provided by the dynamic digital twin model in the spatially aligned data, so as to determine the data gap.

[0046] Data interpolation algorithms are applied to fill data gaps in order to generate a dataset with consistent information dimensions.

[0047] Integrate datasets with unified information dimensions to generate fused 3D models.

[0048] It's important to note that the first step is precise alignment of spatial coordinates, specifically aligning the spatial coordinates of the 3D imaging data and the dynamic digital twin model. The system identifies a common reference point, such as the physical boundary of the culture container or pre-defined markers, and calculates a spatial transformation matrix. This matrix transforms the coordinates of each 3D pixel in the 3D imaging data into the virtual 3D coordinate system of the dynamic digital twin model, ensuring that the cell structure information from the real world and the simulation parameter information from the virtual world have a unified, mutually referential spatial basis. After coordinate alignment, the system initiates a data interpolation algorithm to fill the data gaps. These data gaps refer to the asymmetry in information dimensions between the two data sources. For example, the 3D imaging data provides information about the presence of cells at a specific 3D pixel location but lacks information about nutrient concentration at that location, while the dynamic digital twin model provides predictions of nutrient concentration, but its computational grid may be coarser than that of the imaging 3D pixels. To address this, a data interpolation algorithm, such as trilinear interpolation, calculates the simulation parameter values ​​at the precise location of a 3D pixel based on the predicted values ​​of the grid points surrounding that 3D pixel in the dynamic digital twin model. This method smoothly maps the biochemical parameter fields from the dynamic digital twin model onto a high-resolution structured mesh of the 3D imaging data. Finally, the system integrates the aligned and interpolated data to generate the final fused 3D model. This fused 3D model is structurally a multi-channel 3D matrix, where each 3D pixel carries an information vector. This vector contains not only structural information from the 3D imaging data, such as cell density or fluorescence intensity, but also various functional parameters from the dynamic digital twin model, calculated through interpolation, such as local pH, glucose concentration, and metabolic waste concentration. Thus, a single, information-rich fused 3D model tightly coupled with structure is constructed.

[0049] The technical advantage of this method lies in its innovative solution to the data barrier between physical measurement and digital simulation, deeply synergizing the strengths of these two heterogeneous data sources. It goes beyond simply overlaying two images; through rigorous spatial alignment and data interpolation, it achieves a one-to-one correspondence between the real cellular spatial conformation and the simulated biochemical microenvironment at the microscopic scale. The resulting fused 3D model contains far more information than the simple summation of any single data source, providing unprecedented depth of information for subsequent visualization and decision-making. This unified view of structure and function allows for a deeper understanding of cellular behavior, moving beyond surface observation to the level of mechanistic exploration, significantly enhancing the insight of the entire monitoring and management system.

[0050] The cell culture optimization strategy generation module generates cell culture optimization strategies based on a fusion of a 3D model and a dynamic digital twin model, and applies an intelligent decision-making algorithm.

[0051] In a specific embodiment of the present invention, the specific steps of applying the intelligent decision-making algorithm to generate cell culture optimization strategies include: performing collaborative analysis on the fused three-dimensional model and the dynamic digital twin model to generate identification results regarding cell growth status.

[0052] It should be noted that the generated fused 3D model and dynamic digital twin model are first analyzed collaboratively. The intelligent analysis module in the system automatically processes the generated fused 3D model, using image recognition algorithms to automatically quantify indicators such as the spatial distribution uniformity of cells, cluster size, and morphology. Simultaneously, it queries the dynamic digital twin model to obtain non-visualized biochemical parameters such as the predicted nutrient concentration field, metabolic waste concentration field, and pH distribution. When the intelligent analysis module detects that the cell aggregation degree in the fused 3D model exceeds a preset threshold, and simultaneously detects that the predicted nutrient concentration at the same spatial location in the dynamic digital twin model is below a critical value, the system will make a comprehensive judgment and generate a clear identification result: "abnormal cell growth trend caused by local nutrient depletion."

[0053] The identification results regarding cell growth status are input into a machine learning algorithm to generate multiple candidate optimization strategies.

[0054] In a specific embodiment of the present invention, the specific steps of inputting the identification results of cell growth status into a machine learning algorithm to generate multiple candidate optimization strategies include: training a prediction model for predicting cell response based on historical culture data.

[0055] It should be noted that a predictive model capable of rapid response is trained offline based on historical culture data. This historical culture data is a structured database containing continuously recorded sequences of environmental parameter changes, such as temperature, pH, and feed rate, from multiple successful or failed culture batches, along with corresponding actual cell response sequences, such as cell density, product concentration, and metabolite concentration. The system uses this data to train a predictive model, such as a recurrent neural network, which learns and internalizes the complex, nonlinear causal relationship between environmental parameter changes and cell growth. This predictive model can be abstracted as the following function: In this formula, This represents the completed training of the prediction model. The cell state at the current moment is represented by a dynamic digital twin model or real-time sensors. It represents a hypothetical sequence of environmental parameter changes over a future time period. This is the output of the prediction model, which is the predicted result of the future cellular response under this control sequence.

[0056] Predictive models are used to quickly simulate the effects of changes in various environmental parameters on cell growth, generating a large number of simulation results.

[0057] It's important to note that when the system needs to generate optimization strategies, it doesn't directly perform a brute-force search on the complex dynamic digital twin model. Instead, it first utilizes this lightweight, highly efficient predictive model to quickly simulate the impact of various environmental parameter changes on cell growth. The system automatically generates a set of parameter adjustment schemes containing numerous different assumptions, such as increasing or decreasing the feeding rate or adjusting the temperature at different slopes and magnitudes over the next six hours. The system uses each hypothetical scheme as... Input into the prediction model In this process, the corresponding prediction results are obtained within milliseconds. .

[0058] Based on a large number of simulation results, parameter change schemes with favorable prediction effects were selected and used as multiple candidate optimization strategies.

[0059] It should be noted that, finally, based on these numerous simulation results, the system automatically filters and generates candidate optimization strategies. The selection is based on preset performance metrics; for example, the system will select simulation schemes that result in the fastest predicted increase in live cell density while simultaneously producing the slowest predicted increase in lactate concentration. The input control sequences that produce favorable prediction results are then considered. This is then encapsulated by the system into specific, executable candidate optimization strategies.

[0060] In a dynamic digital twin model, the virtual effects of multiple candidate optimization strategies are evaluated, and the optimal strategy is selected as the cell culture optimization strategy based on the evaluation results.

[0061] It should be noted that these candidate optimization strategies are virtually evaluated without interfering with the actual cell culture process. The system applies each candidate optimization strategy to the current dynamic digital twin model for rapid forward simulation, predicting the cell growth status over a future period. The system uses a predefined optimization objective function to quantify the virtual effect of each strategy, the expression of which can be: In this formula, It is an optimization score calculated to evaluate the virtual effect. This represents the future live cell density predicted by the model after applying this strategy, while This represents the predicted concentration of inhibitory metabolites such as lactic acid. and These represent the future live cell density and the concentration of inhibitory metabolites, respectively, as set as a reference. and This is a weighting factor, whose value is preset based on the main objective of this batch of cell culture, used to balance the importance of different objectives and unify the dimensions. The system calculates a weighting factor for each candidate optimization strategy. The strategy with the highest score is selected as the final output cell culture optimization strategy.

[0062] In one specific embodiment of the present invention, and The selection of typical values ​​needs to be combined with the specific goals and priorities of cell culture. If the core goal is to maximize viable cell density, while also appropriately considering the inhibition of lactate accumulation, 0.8 is acceptable. The value is 0.2, which is determined by assigning... Higher weighting makes the model focus more on increasing viable cell density during optimization, while relatively weakening the consideration of lactate inhibition; if the culture objective is to seek a more balanced optimization effect between viable cell density and lactate inhibition, Take 0.6, A value of 0.4 is more appropriate, as the weight difference between the two is small, which allows both objectives to be fully reflected in the optimization process, avoiding one objective from becoming too dominant and neglecting the other.

[0063] The technological effects of this method are revolutionary, elevating cell culture management from a passive response and trial-and-error model to a new level of proactive prediction and intelligent optimization. By pre-simulating the consequences of different control strategies in a digital twin virtual environment, the system can safely and efficiently screen for the optimal intervention before potential problems escalate into serious failures. This decision-making mechanism based on predictive evaluation significantly reduces the risks and costs associated with directly adjusting parameters during actual culture, ensuring that every adjustment is targeted and effective, thereby significantly improving the stability, reproducibility, and yield and quality of the final product in the cell culture process.

[0064] In a specific embodiment of the present invention, the method further includes: acquiring actual cell growth data obtained through offline analysis after the culture process is completed.

[0065] It is important to note that, firstly, actual cell growth data is obtained after a complete cell culture process. This data is typically obtained offline using high-precision gold standard methods, such as precise live cell counting using a hemocytometer or flow cytometer after aseptic sampling, and analysis of key metabolite concentrations in the culture supernatant using high-performance liquid chromatography (HPLC). This yields reliable, real-world cell growth data covering the entire culture cycle.

[0066] The full-cycle prediction results of the dynamic digital twin model are compared with actual cell growth data to generate global model error parameters.

[0067] It should be noted that, next, the system will comprehensively compare the full-cycle prediction results output by the dynamic digital twin model during this batch of culture with the actual cell growth data obtained above. This comparison is not a comparison of instantaneous values, but rather an evaluation of the fit of the entire time series curve. The system generates quantified model error parameters by calculating the cumulative deviation between the two. These model error parameters can be defined by an integral form of the error function, as follows: In this formula, This refers to the final generated model error parameter, which is a scalar representing the overall predictive performance of the model throughout the entire training period. At any moment The actual cell growth data obtained, It is a dynamic digital twin model at any time The prediction results. Represents the training period. This is the total duration of the culture batch. The formula calculates the integral of the squared difference between the predicted curve and the actual data curve over the entire time domain, comprehensively reflecting the degree of model deviation.

[0068] A feedback algorithm is applied, and the biological parameters used to construct the initial digital twin model are adjusted based on the generated global model error parameters.

[0069] It should be noted that, finally, the system inputs the model error parameters into a feedback algorithm. This feedback algorithm, such as a gradient descent-based optimizer, aims to minimize the model error parameters. By backpropagating this error, the feedback algorithm adjusts the fundamental biological parameters relied upon in constructing the initial digital twin model, such as the maximum specific growth rate of cells, substrate affinity constant, or inhibition constant. The feedback algorithm iteratively and automatically calculates the adjustment gradients of these fundamental parameters and updates their values ​​accordingly. The new set of parameters optimized by this feedback algorithm will be stored and used to construct a more accurate initial digital twin model for the next batch of cell cultures.

[0070] The technical effect of this method is that it endows digital twin systems with the ability to learn across batches and self-evolve. By reviewing each complete culture cycle, the system can identify and correct systematic biases in the model at a macro level, making the digital twin model not merely a passive data mirror, but an intelligent agent capable of continuously learning and accumulating experience. This long-term feedback optimization mechanism ensures that the model's understanding of specific cell lines and processes deepens with increased production experience, thus providing more accurate predictions and more reliable decision support from the outset in future culture batches, forming a virtuous cycle of continuous optimization.

[0071] In a specific embodiment of the present invention, it further includes: receiving user input parameters input by the operator through a graphical user interface for defining the current training objective.

[0072] It should be noted that the graphical user interface (GUI) receives specific user input parameters from the operator. This GUI not only displays the cell growth curves and key parameters predicted by the dynamic digital twin model in real time, but also provides the operator with interactive control options, such as specifying the primary objective of the current culture stage by adjusting virtual sliders or inputting numerical values. User input parameters can adjust the weights of the objective function; for example, when the objective is to rapidly expand cells, the user can increase the weight of viable cell density in the objective function; when the objective is to produce a specific protein, the user can increase the weight of the target product concentration in the function.

[0073] User input parameters are integrated into intelligent decision-making algorithms to generate adjusted optimization strategies that reflect the operator's intentions.

[0074] It should be noted that the system then integrates these user input parameters provided by the operator into the intelligent decision-making algorithm in real time. Specifically, if the user adjusts the optimization objective, the system will directly use the weights input by the user to modify the optimization objective function. For example, the modified function becomes: In this formula, This is the adjusted, optimized score. and Instead of fixed values ​​preset by the system, the weights are weight parameters input by the user through a graphical user interface. These two parameters directly reflect the operator's judgment and preference for the current training objective. Upon receiving the new weights, the intelligent decision-making algorithm re-evaluates all candidate optimization strategies and bases its decisions on the adjusted optimization scores. The optimal strategy is selected to generate an adjusted, optimized strategy that reflects the operator's intent.

[0075] The dynamic digital twin model is updated based on the adjusted optimization strategy, and the updated prediction results are displayed on the graphical user interface.

[0076] It should be noted that, finally, this newly generated and adjusted optimization strategy will be immediately used to update the dynamic digital twin model. The system will use this strategy as the new control input to drive the dynamic digital twin model to perform a rapid forward simulation to predict the future state evolution of the cell culture system after implementing this new strategy. The simulation results will be immediately displayed on the graphical user interface as new prediction curves, providing the operator with immediate and intuitive feedback on the possible consequences of their decisions.

[0077] The technical effects achieved by this method are significant. It bridges the gap between automated systems and human experts, forming a highly efficient human-machine collaborative decision-making loop. It doesn't simply allow users to override system decisions; instead, it quantifies the user's macro-level intentions into algorithmic parameters, enabling the system to perform refined intelligent optimization under human guidance. This decision-making model, combining machine computing power and human experience and wisdom, allows the system to flexibly cope with various complex and unexpected production scenarios. Its robustness and effectiveness far surpass those of a single automated system or purely manual operation, achieving synergistic efficiency through human-machine intelligence.

[0078] Reference Figure 2 The second aspect of the present invention provides a method for managing cell culture monitoring data, comprising: S1, initial digital twin model construction: acquiring environmental parameters and cell characteristic parameters of cell culture, and constructing an initial digital twin model based on the environmental parameters and cell characteristic parameters.

[0079] S2. Dynamic Digital Twin Model Generation: Acquire real-time monitoring data and update the initial digital twin model based on the real-time monitoring data to generate a dynamic digital twin model.

[0080] S3. Fusion 3D Model Generation: Acquire 3D imaging data and fuse it with a dynamic digital twin model to generate a fused 3D model.

[0081] S4. Generation of cell culture optimization strategies: Based on the fusion of 3D model and dynamic digital twin model, intelligent decision-making algorithm is applied to generate cell culture optimization strategies.

[0082] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A cell culture monitoring data management system, characterized in that, include: The initial digital twin model construction module acquires environmental parameters and cell characteristic parameters of cell culture, and constructs an initial digital twin model based on these parameters. The dynamic digital twin model generation module acquires real-time monitoring data and updates the initial digital twin model based on the real-time monitoring data to generate a dynamic digital twin model. The 3D model generation module acquires 3D imaging data and fuses it with a dynamic digital twin model to generate a fused 3D model. The cell culture optimization strategy generation module, based on the fusion of a 3D model and a dynamic digital twin model, applies intelligent decision-making algorithms to generate cell culture optimization strategies.

2. The cell culture monitoring data management system according to claim 1, characterized in that, The specific steps for constructing the initial digital twin model based on environmental parameters and cell characteristic parameters include: The physicochemical information of the cell culture environment is collected through a sensor network to obtain environmental parameters; Extract data that matches the currently cultured cells from a database containing various cell biological characteristics to obtain cell characteristic parameters; Environmental parameters and cell characteristic parameters are input into a computer simulation algorithm to construct an initial digital twin model.

3. The cell culture monitoring data management system according to claim 1, characterized in that, The specific steps for updating the initial digital twin model by combining real-time monitoring data to generate a dynamic digital twin model include: Real-time monitoring data reflecting cell population status and environmental changes are obtained through an online monitoring system; Real-time monitoring data is compared with the predicted data from the initial digital twin model to generate model error parameters; A model calibration algorithm is applied, and the internal parameters of the initial digital twin model are adjusted based on the generated model error parameters to generate a dynamic digital twin model.

4. The cell culture monitoring data management system according to claim 1, characterized in that, The specific steps for fusing 3D imaging data with a dynamic digital twin model to generate a fused 3D model include: The cell culture container is scanned using an optical scanning device to obtain three-dimensional imaging data; Align the spatial coordinates of the 3D imaging data with those of the dynamic digital twin model to generate spatially aligned data; Interpolation of the information dimensions of the spatially aligned data is performed to generate a fused 3D model.

5. A cell culture monitoring data management system according to claim 4, characterized in that, The specific steps for interpolating the spatially aligned data according to the information dimension to generate a fused 3D model include: Identify the dimensional differences between the structural information provided by 3D imaging data and the functional information provided by dynamic digital twin models in spatially aligned data to determine data gaps; Data interpolation algorithms are applied to fill data gaps in order to generate a dataset with uniform information dimensions; Integrate datasets with unified information dimensions to generate fused 3D models.

6. The cell culture monitoring data management system according to claim 1, characterized in that, The specific steps for applying intelligent decision-making algorithms to generate cell culture optimization strategies include: The fusion of 3D models and dynamic digital twin models is analyzed collaboratively to generate identification results about cell growth status; The identification results regarding cell growth status are input into a machine learning algorithm to generate multiple candidate optimization strategies; In a dynamic digital twin model, the virtual effects of multiple candidate optimization strategies are evaluated, and the optimal strategy is selected as the cell culture optimization strategy based on the evaluation results.

7. A cell culture monitoring data management system according to claim 6, characterized in that, The specific steps of inputting the identification results of cell growth status into the machine learning algorithm to generate multiple candidate optimization strategies include: A predictive model for predicting cell responses was trained based on historical culture data. Predictive models are used to quickly simulate the effects of various environmental parameter changes on cell growth, generating a large number of simulation results. Based on a large number of simulation results, parameter change schemes with favorable prediction effects were selected and used as multiple candidate optimization strategies.

8. The cell culture monitoring data management system according to claim 1, characterized in that, Also includes: After the culture process is completed, obtain actual cell growth data through offline analysis; The full-cycle prediction results of the dynamic digital twin model are compared with actual cell growth data to generate global model error parameters. A feedback algorithm is applied, and the biological parameters used to construct the initial digital twin model are adjusted based on the generated global model error parameters.

9. A cell culture monitoring data management system according to claim 1, characterized in that, Also includes: The system receives user input parameters from the operator to define the current training objective through a graphical user interface. Integrate user input parameters into intelligent decision-making algorithms to generate adjusted optimization strategies that reflect the operator's intent; The dynamic digital twin model is updated based on the adjusted optimization strategy, and the updated prediction results are displayed on the graphical user interface.

10. A method for managing cell culture monitoring data, characterized in that, include: S1. Initial digital twin model construction: Obtain environmental parameters and cell characteristic parameters for cell culture, and construct an initial digital twin model based on these parameters. S2. Dynamic Digital Twin Model Generation: Acquire real-time monitoring data and update the initial digital twin model in combination with the real-time monitoring data to generate a dynamic digital twin model. S3. Fusion 3D Model Generation: Acquire 3D imaging data and fuse the 3D imaging data with a dynamic digital twin model to generate a fused 3D model; S4. Generation of cell culture optimization strategies: Based on the fusion of 3D model and dynamic digital twin model, intelligent decision-making algorithm is applied to generate cell culture optimization strategies.

Images