Cell culture solution metabolite concentration determination method and apparatus, device and medium

Abstract

A cell culture solution metabolite concentration determination method and apparatus, a computer device and a computer-readable storage medium. The method comprises: acquiring spectral data of a preset length of a target detection cell culture solution at a target moment and a preset cell culture environment parameter, and inputting the spectral data and the preset cell culture environment parameter into a target concentration prediction model to obtain a metabolite concentration prediction result corresponding to the target detection cell culture solution at the target moment, wherein the target concentration prediction model comprises a plurality of data feature extraction modules, and the plurality of data feature extraction modules are respectively used for performing feature extraction on the spectral data and the preset cell culture environment parameter from different feature analysis dimensions. The apparatus corresponding to the method comprises a data acquisition unit and a concentration prediction unit. The computer device and the computer-readable storage medium comprise a processor for executing the method, and a computer program. Data feature analysis is carried out on the basis of multi-modal data of the tested cell culture solution, such that a more accurate cell culture solution concentration prediction result can be obtained.

Description

Method, device, equipment and medium for determining metabolite concentration in cell culture fluid

[0001] This application is based on the application with CN application number 202311548419.6 and application date November 20, 2023, and claims its priority. The disclosed content of the CN application is hereby introduced as a whole into this application. Technical Field

[0002] The present disclosure relates to the field of biological detection technology, and in particular to a method, device, equipment and medium for determining the concentration of metabolites in a cell culture fluid. Background Art

[0003] In biotechnology research and application, monitoring the concentration of cellular metabolites in cell culture fluids during cell culture is often necessary. Raman spectroscopy, with its advantages of being non-destructive, highly sensitive, and label-free, is increasingly being used in the detection of cell culture metabolites.

[0004] Summary of the Invention

[0005] In a first aspect, the present disclosure provides a method for determining the concentration of metabolites in a cell culture fluid, the method comprising:

[0006] Acquiring spectral data of a preset length of a target detection cell culture fluid at a target time and preset cell culture environment parameters;

[0007] Inputting the spectral data and the preset cell culture environment parameters into a target concentration prediction model to obtain a metabolite concentration prediction result corresponding to the target detection cell culture fluid at the target time;

[0008] The target concentration prediction model includes a plurality of data feature extraction modules, and the plurality of data feature extraction modules are respectively used to extract features of the spectral data and the preset cell culture environment parameters from different feature analysis dimensions.

[0009] In a second aspect, the present disclosure provides a device for determining the concentration of metabolites in a cell culture fluid, the device comprising:

[0010] A data acquisition unit, configured to acquire spectral data of a preset length of a target detection cell culture fluid at a target time and preset cell culture environment parameters;

[0011] a concentration prediction unit, configured to input the spectral data and the preset cell culture environment parameters into a target concentration prediction model to obtain a predicted result of the metabolite concentration of the target detection cell culture fluid corresponding to the target time;

[0012] The target concentration prediction model includes a plurality of data feature extraction modules, and the plurality of data feature extraction modules are respectively used to extract features of the spectral data and the preset cell culture environment parameters from different feature analysis dimensions.

[0013] In a third aspect, the present disclosure further provides a computer device, comprising:

[0014] one or more processors;

[0015] a memory for storing one or more programs;

[0016] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for determining the metabolite concentration in the cell culture fluid as provided in any embodiment of the present disclosure.

[0017] In a fourth aspect, the present disclosure further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for determining the concentration of metabolites in a cell culture fluid as provided in any embodiment of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] FIG1 is a flow chart of a method for determining metabolite concentrations in a cell culture fluid according to some embodiments of the present disclosure;

[0019] FIG2 is a flow chart of a method for determining metabolite concentrations in a cell culture fluid according to some embodiments of the present disclosure;

[0020] FIG3 is a schematic diagram of a training process of a target concentration prediction model according to some embodiments of the present disclosure;

[0021] FIG4 is a graph showing component trend changes in a tank 6 during an experiment according to some embodiments of the present disclosure;

[0022] FIG5 is a graph showing component trend changes in a tank 9 during an experiment according to some embodiments of the present disclosure;

[0023] FIG6 is a flow chart of a method for determining metabolite concentrations in a cell culture fluid according to some embodiments of the present disclosure;

[0024] FIG7 is a schematic diagram of an application example of a target concentration prediction model according to some embodiments of the present disclosure;

[0025] FIG8 is a schematic diagram of automatically controlled glucose concentration matching a peristaltic pump according to some embodiments of the present disclosure;

[0026] FIG9 is a schematic diagram of manually controlled glucose concentration matched to a peristaltic pump according to some embodiments of the present disclosure;

[0027] FIG10 is a schematic structural diagram of a device for determining the concentration of metabolites in a cell culture fluid according to some embodiments of the present disclosure;

[0028] FIG11 is a schematic structural diagram of a computer device according to some embodiments of the present disclosure. DETAILED DESCRIPTION

[0029] The present disclosure will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate the present disclosure and are not intended to limit the present disclosure. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions relevant to the present disclosure, not all structures.

[0030] In the process of implementing the present disclosure, it was found that due to the high dimensional complexity of Raman spectral data and the diversity between data, the use of simple single-modal machine learning or deep learning (such as text convolution model) to predict metabolite concentrations has certain limitations and large prediction errors.

[0031] The present disclosure provides a method, apparatus, device, and medium for determining the concentration of metabolites in a cell culture fluid, which can perform data feature analysis based on multimodal data of the measured cell culture fluid to obtain more accurate cell culture fluid concentration prediction results.

[0032] Specifically, the present disclosure obtains spectral data of a preset length and preset cell culture environment parameters of a target detection cell culture fluid at a target moment, and inputs the spectral data and preset cell culture environment parameters into a target concentration prediction model to obtain a prediction result of the metabolite concentration corresponding to the target detection cell culture fluid at a target moment. The target concentration prediction model includes a plurality of data feature extraction modules, and the plurality of data feature extraction modules are respectively used to extract features from the spectral data and the preset cell culture environment parameters from different feature analysis dimensions. The technical solution of the embodiment of the present disclosure solves the problem of inaccurate prediction results of the existing cell culture fluid concentration analysis model, and can perform data feature analysis based on the multimodal data of the tested cell culture fluid to obtain a more accurate prediction result of the cell culture fluid concentration.

[0033] Figure 1 is a flow chart of a method for determining metabolite concentrations in cell culture fluids according to some embodiments of the present disclosure. This embodiment is applicable to biological experiment scenarios, particularly when predicting cellular metabolite concentrations during cell culture in a bioreactor. This method can be performed by a device for determining metabolite concentrations in cell culture fluids, which can be implemented using software and / or hardware and integrated into a computer device with application development capabilities.

[0034] As shown in FIG1 , the method for determining the concentration of metabolites in a cell culture fluid of this embodiment includes the following steps S110 and S120 .

[0035] In step S110 , spectrum data of a preset length of a target detection cell culture fluid at a target time and preset cell culture environment parameters are acquired.

[0036] Specifically, the target detection cell culture fluid can be any cell culture fluid in a biological reaction process in which the concentration of cell metabolites in the cell culture fluid needs to be analyzed. The target moment can be any moment in the biological reaction process in which the target detection cell culture fluid is detected.

[0037] The spectral data can be any type of spectral information capable of reflecting concentration differences of different substances in the cell culture fluid. For example, the spectral data can be Raman spectral data. Online spectral data of the target cell culture fluid at a target time can be obtained using a Raman spectrometer. The preset length of the spectral data can be an optimal data format determined experimentally during the training of the target concentration prediction model, or can be determined based on the experience of relevant technicians.

[0038] In the past, the detection of metabolite concentrations in cell culture fluids was usually based solely on spectral data for concentration analysis to obtain a metabolite concentration prediction result. However, considering that the environmental parameters in the cell culture process will also have a certain impact on the biological reaction process, when performing metabolite concentration analysis in this embodiment, the environmental parameters will also be combined to perform metabolite concentration prediction analysis to obtain a more accurate prediction result. The metabolite concentration in the target cell culture fluid can be analyzed based on multimodal data of different dimensions obtained by different detection methods of the target cell culture fluid.

[0039] Specifically, the preset cell culture environment parameters may include, but are not limited to, discrete metadata related to the cell culture experiment, such as cell line, clone, culture process, and other parameters, such as cell line type, culture medium type, temperature and pH during the cell culture process, and other parameters. The preset cell culture environment parameters may be one or a combination of the above parameters. In some embodiments, the preset cell culture environment parameters include at least one parameter selected from the group consisting of cell line parameters, cell clone parameters, culture process parameters, and various parameters during the cell culture process.

[0040] In step S120 , the spectral data and the preset cell culture environment parameters are input into the target concentration prediction model to obtain a metabolite concentration prediction result corresponding to the target detection cell culture fluid at the target time.

[0041] Specifically, the target concentration prediction model can be a pre-trained neural network model that can predict metabolite concentrations in the cell culture fluid based on input data and output corresponding metabolite concentration prediction results. The target concentration prediction model can more accurately predict metabolite concentrations in the cell culture fluid because it can include multiple data feature extraction modules, each of which is used to extract features from spectral data and preset cell culture environment parameters from different feature analysis dimensions.

[0042] In some embodiments, the target concentration prediction model further includes a feature fusion module and a concentration prediction module. The feature fusion module is configured to fuse the data features extracted by the multiple data feature extraction modules to obtain a target fusion feature. The concentration prediction module is configured to analyze the target fusion feature to obtain the metabolite concentration prediction result.

[0043] In some embodiments, the target concentration prediction model is trained by using a Raman spectrometer to obtain online spectral data on the time-varying concentrations of different components in the cell culture fluid of a bioreactor. The corresponding culture environment parameters are then collected. During the experiment, samples are taken periodically and offline detection equipment is used to obtain offline target values ​​for each batch. This process can, for example, combine data feature extraction modules and feature fusion modules designed for different data characteristics to provide rich contextual information to the concentration prediction module of the entire model, thereby improving the prediction and analysis accuracy of the target concentration prediction model.

[0044] Therefore, by inputting the acquired spectral data and the preset cell culture environment parameters into the target concentration prediction model, the metabolite concentration prediction result corresponding to the target detection cell culture fluid at the target time can be obtained.

[0045] In some embodiments, the data feature extraction module in the target concentration prediction model includes a first data feature extraction module, a second data feature extraction module, and a third data feature extraction module. The first data feature extraction module is used to perform feature coding analysis on the preset cell culture environment parameters to obtain target environment parameter characteristics. The second data feature extraction module is used to extract spectral features between spectral data of different frequency bands in the spectral data. The third data feature extraction module is used to extract at least one spectral feature of the preset spectral statistics feature, the preset spectral feature, and the preset spectral peak analysis feature of the spectral data.

[0046] It should be noted here that the preset spectral statistics (such as mean, median, etc.), preset spectral features (such as spectral energy, spectral entropy, etc.), and preset spectral peak analysis features (such as the number of peaks, average peak height, distance between peaks, etc.) of the spectral data are added to the algorithm because not only are the differences in optical signals between Raman devices large, but differences also exist within the same device (with varying degrees of drift). Differences between devices often require the establishment of a set of optical physics and mathematical strategies for conversion based on the optical signal data. This approach is time-consuming, labor-intensive, and lacks accuracy. The current approach to differences between different channels within a device is to calibrate optical fibers, optical coupling sensors, and laser light sources through optical standards, and calibrate different optical channels to the same order of magnitude. This approach will bring unnecessary culture risks such as contamination during the experiment. The embodiment of the present disclosure can effectively eliminate spectral differences between devices and within devices by adding these statistics, thereby making the predicted results more accurate.

[0047] In some embodiments, the second data feature extraction module can be used to extract spectral features between spectral data of different frequency bands in the spectral data using a neural network module with a self-attention mechanism. The self-attention mechanism can be introduced during feature processing of spectral data. In the Transformer model, the self-attention mechanism, as a core component, can improve the model's parallel processing capabilities and depth of understanding.

[0048] The self-attention mechanism is based on the similarities between the characteristics of Raman spectroscopy data and natural language data. First, the correlation between spectral intensities in different frequency bands in Raman spectroscopy can be compared to the semantic and grammatical relationships between words or phrases in natural language. In Raman spectroscopy, different frequency bands may reflect the characteristics of the same or similar chemical components, thus inherently interrelated. Similarly, in natural language text, different words or phrases may have semantic dependencies and influence each other. Second, the analysis of both Raman spectroscopy and natural language data requires capturing these complex internal relationships. Methods based on the self-attention mechanism precisely emphasize this point. By dynamically assigning weights to different components, they can capture the correlations between different frequency bands in Raman spectroscopy. This characteristic is similar to the self-attention mechanism's ability to handle long-range dependencies in natural language. Furthermore, neural network models based on the self-attention mechanism can effectively overcome the limitations of spectral band distance and effectively capture the correlations between metabolites and spectral values ​​in each band. This analysis can be performed automatically, without the need for manual steps such as feature selection, significantly improving the efficiency of the modeling and inference process. Finally, both Raman spectroscopy and natural language data can, to a certain extent, be represented as sequences. Precisely because of this similarity, the self-attention mechanism can bridge these two domains, providing a unified and efficient approach for both quantitative analysis of Raman spectroscopy and deep understanding of natural language. The target concentration prediction model trained based on these comprehensive data features has been validated in practice to achieve more accurate predictions.

[0049] The technical solution of this embodiment obtains spectral data of a preset length and preset cell culture environment parameters of the target detection cell culture fluid at the target moment; inputs the spectral data and preset cell culture environment parameters into a target concentration prediction model to obtain a predicted result of the metabolite concentration of the target detection cell culture fluid corresponding to the target moment; the target concentration prediction model includes multiple data feature extraction modules, and the multiple data feature extraction modules are respectively used to extract features of the spectral data and preset cell culture environment parameters from different feature analysis dimensions. The technical solution of the disclosed embodiment solves the problem of inaccurate prediction results of the existing cell culture fluid concentration analysis model, and can perform data feature analysis based on the multimodal data of the tested cell culture fluid to obtain a more accurate prediction result of the cell culture fluid concentration. In particular, the analysis of the spectral data of the preset length of the target detection cell culture fluid at the target moment based on the self-attention mechanism can capture deeper data features to reflect the final prediction result.

[0050] Figure 2 is a flow chart of a method for determining metabolite concentrations in a cell culture fluid according to some embodiments of the present disclosure. This embodiment, which shares the same inventive concept as the method for determining metabolite concentrations in a cell culture fluid described above, further describes the process of training a target concentration prediction model. This method can be performed by a device for determining metabolite concentrations in a cell culture fluid. This device can be implemented using software and / or hardware and integrated into a computer device capable of application development.

[0051] As shown in FIG2 , the method for determining the concentration of metabolites in a cell culture fluid of this embodiment includes the following steps S210 to S240 .

[0052] In step S210 , spectrum sample data of the concentrations of different components of the cell culture solution at different times, corresponding preset cell culture environment parameter sample data, and offline concentration detection values ​​are obtained.

[0053] For example, the process of obtaining the target concentration prediction model in the above embodiment can be roughly divided into several steps: data processing, model building, data training, and model verification.

[0054] In this step, data processing is first performed to construct sample data for model training.

[0055] Specifically, online spectral data of the concentrations of different components of the cell culture fluid changing over time can be obtained from the cell culture fluid of the bioreactor by, for example, using a Raman spectrometer, and collecting culture environment parameter values. At the same time, samples can be taken at regular intervals during the experiment and offline detection equipment can be used to obtain offline target values ​​for the concentrations of different components of the cell culture fluid.

[0056] Exemplarily, spectral data were obtained by installing a pre-set Raman spectrometer probe in a bioreactor and immersing it directly in the cell culture medium. Raman spectra were recorded for different bioreactors throughout the experiment. For a single recorded spectrum, 30 subsequent spectra were captured with a 10 s exposure time and averaged, with a sampling interval of approximately 5 minutes for each bioreactor. The laser excitation wavelength during the spectrometer operation was 785 nm, providing a 100-3425 cm -1 spectral coverage (Raman shift).

[0057] For example, the preset cell culture environment parameter sample data may include: the conditions when the cells are cultured in a shaker during the seed stage, such as temperature, shaker speed, carbon dioxide concentration level, culture medium parameters, the reactor size used in the production culture stage (such as 3L and 250L), and the initial culture volume, culture temperature, pH setting value, dissolved oxygen saturation, initial inoculation density, and feed medium parameters of the cell culture medium.

[0058] The offline concentration detection values ​​of different components can be the target values ​​detected after sampling 5 times a day, such as but not limited to the viable cell density (VCD), cell viability (Viability) and average cell diameter (AvgDiam) detected by a cell counter (such as Vi-CELL XR); glucose (glucose), lactate (lactate), ammonium ion (NH4 + ), glutamate, glutamine, iron, phosphate, lactate dehydrogenase (LDH), and target protein concentration (titer); osmotic pressure (Osmolality) measured by an osmometer (such as OsmoPRO); and acidity (pH), carbon dioxide partial pressure (pCO2), sodium ion (Na) measured by a blood gas analyzer. + ) and potassium ions (K + ).

[0059] In step S220, the spectral sample data and the preset cell culture environment parameter sample data are used as model training input data, and the offline concentration detection value at the corresponding time is used as model training label data to train the preset initial concentration prediction model to obtain a target concentration prediction model.

[0060] Each model training sample uses the spectral sample data and the preset cell culture environment parameter sample data as model training input data, and the offline concentration detection value at the corresponding time as model training label data. The detection time of the spectrum and the detection time of the offline concentration detection value can correspond one to one.

[0061] To analyze the acquired multimodal data, when constructing the model, the preset initial concentration prediction model includes multiple data feature extraction modules for extracting data features from different dimensions to obtain higher-order data features. In some embodiments, the data feature extraction modules include a first data feature extraction module, a second data feature extraction module, and a third data feature extraction module. The first data feature extraction module is configured to perform feature encoding analysis on the preset cell culture environment parameters to obtain target environment parameter features; the second data feature extraction module is configured to extract spectral features between spectral data in different frequency bands within the spectral data; and the third data feature extraction module is configured to extract at least one spectral feature from the spectral data selected from the group consisting of preset spectral statistics, preset spectral features, and preset spectral peak analysis features. In some examples, the second data feature extraction module is a neural network module with a self-attention mechanism that can extract spectral features between spectral data in different frequency bands within the spectral data. The main idea of ​​the attention mechanism is to assign different weights to input signals at different locations when processing sequential data, allowing the model to focus more on important inputs. In deep learning models, attention mechanisms are typically implemented by adding additional network layers that learn how to calculate weights and apply these weights to the input signal.

[0062] Exemplarily, the process of training the preset initial concentration prediction model can refer to the process shown in Figure 3. In Figure 3, the data in the historical database is the sample data for model training. The sample data is input into the model, and the sample data is subjected to feature extraction through three paths (corresponding to three data feature extraction modules). Among them, after the preset cell culture environment parameter sample data is uniquely encoded, the high-order features of the environmental parameters are obtained through a dual-channel normalization layer and a fully connected network, which corresponds to the first data feature extraction module. The spectral signal is channel amplified, and then passed through a dual-channel normalization layer, a multi-head attention network, and a fully connected network and then pooled once to obtain high-order features of the frequency band spectrum, which corresponds to the second data feature extraction module. In addition, the global features of the spectral data are calculated and selected, and the high-order features of the global spectrum are obtained through a dual-channel normalization layer and a fully connected network, which corresponds to the third data feature extraction module.

[0063] Multi-head attention is an extension of the attention mechanism that can more effectively extract information when processing sequential data. In multi-head attention, multiple sets of attention weights are used, each of which can learn different semantic information and generate a context vector. Finally, these context vectors are concatenated and then subjected to a linear transformation to produce the final output.

[0064] Furthermore, after extracting features of spectral data and cell culture environmental parameters from multiple dimensions as shown in Figure 3, multimodal feature fusion is performed on the extracted high-order features of environmental parameters, high-order features of frequency band spectra, and high-order features of global spectra. The fusion results are then passed through the concentration prediction module to predict the components of multiple detection items, that is, to predict the concentrations of each component in the cell culture medium and obtain the final prediction results.

[0065] In addition, it should be noted that, in the model training stage, in addition to the model structure shown in FIG3 (multimodal Raman spectrum converter model based on attention, RAFORMER-multimodal), a variety of machine learning models are also constructed in this embodiment, including partial least squares regression method (Partial Least Squares Regression, PLSR), gradient boosting decision tree (Gradient Boosting Decision Tress, GBDT), text convolution (TextCNN) model, and only using Raman spectrum data based on attention Raman spectrum converter model (RAFORMER-single modal). Further, each model structure of the structure is trained based on the model training sample, and verification and test comparison are performed. Specifically, first, according to the verification results of the test set, it is ensured that each model of the training has stable reasoning ability. Then, on the basis of the stable performance of the test set, in order to ensure that the production of the model is available, all model effects are evaluated by wet experiments. The wet experiment process is described as follows: Under the same culture conditions as the historical training data, multiple batches of cell culture experiments were tested, and Raman spectral data were collected by scanning the bioreactors. Four points were collected daily, with samples collected simultaneously for offline data collection throughout the entire culture cycle. Model evaluation was then conducted based on the collected data. The root mean square error (RMSE) between the model predictions and offline measurements showed that the attention-based multimodal Raman spectral converter model demonstrated higher accuracy and superior performance. Furthermore, the model did not require hyperparameter selection to tailor the model to a single application scenario, significantly reducing the need for manual intervention in the modeling process. This enhances the versatility of RAFORMER and its potential for large-scale industrial application. Comparative analysis of multiple model architectures confirmed the superior performance of the RAFORMER-multimodal model after accounting for environmental parameters relevant to cell culture experiments. Furthermore, given the diverse culture processes and other aspects of the tested bioreactors, the multimodal model has greater potential for general applicability to a wide range of clones and cell lines compared to single-modal models. This effectively overcomes the limitations of single-modal models that focus solely on spectral data analysis, minimizing the loss of prediction accuracy.

[0066] Furthermore, the predicted trends of the RAFORMER-Multimodal model were compared with those of the baseline model. The Pearson correlation coefficient was used to assess the goodness of fit of this trend. Comparing the performance of the RAFORMER-Multimodal model with the two traditional models, GBDT and PLSR, on the various bioreactors described above revealed that the RAFORMER-Multimodal model showed improvements in Pearson correlation coefficients compared to GBDT and PLSR for all concentration measurements.

[0067] In summary, it can be determined that the target concentration prediction model trained in this embodiment can obtain more accurate prediction results of cell fluid metabolite concentrations based on the analysis of multimodal data.

[0068] Specifically, the detailed process of evaluating the above model is as follows:

[0069] The trained RAFORMER-multimodal model, RAFORMER-unimodal model, GBDT model, TextCNN model, and PLSR model were used to predict and analyze the metabolite concentrations in the cell culture fluid of two selected bioreactors. The two bioreactors are labeled as tank 6 and tank 9. The evaluation results of these two bioreactors were selected as the results because they represent 3L and 200L culture processes, as well as the culture processes of enhanced fed-batch culture and traditional fed-batch culture, respectively, which can more comprehensively reflect the applicability of the model under different conditions. For the evaluation comparison results of tank 6 and tank 9, please see Table 1 and Table 2 respectively:

[0070] Table 1: Root mean square error between model predictions and offline measurements for bioreactor tank 6

[0071] Table 2: Root mean square error between model predictions and offline measurements for bioreactor tank 9

[0072] In the two tables, VCD represents viable cell density, Glucose represents glucose, Iron represents iron ion, Glutamine represents glutamine, PH represents acidity, pCO2 represents carbon dioxide partial pressure, and Na + represents sodium ion, K + represents potassium ion, Lactate represents lactic acid, NH4 + represents ammonium ion, Glutamate represents glutamate, OSMO represents osmotic pressure, Viability represents cell viability, AvgDiam represents average cell diameter, Titer represents target protein concentration, LDH represents lactate dehydrogenase, and Phosphate represents phosphate ion.

[0073] Comparing the performance of the RAFORMER-Multimodal model with two traditional models, GBDT and PLSR, on the two bioreactors, across all 17 test items, revealed that on the six-tank bioreactor, the RAFORMER-Multimodal model achieved an average RMS error improvement of 47.57% and 36.86% over GBDT and PLSR, respectively. On the nine-tank bioreactor, the RAFORMER-Multimodal model achieved an average RMS error improvement of 14.67% and 17.47% over GBDT and PLSR, respectively. Specifically, RAFORMER achieved the best performance on 16 of the 17 test items on the six-tank bioreactor and on 12 of the nine-tank bioreactor. Best performance refers to achieving the lowest RMS error when comparing the RAFORMER-Multimodal model to the baseline models, GBDT and PLSR. It's also worth noting that, compared to the baseline models GBDT and PLSR, the RAFORMER-Multimodal model doesn't require hyperparameter selection or other steps to tune the model for a single application scenario, significantly reducing the effort required for manual intervention in the modeling process. This also enhances the versatility of the RAFORMER-Multimodal model and its potential for large-scale industrial application.

[0074] To validate the superiority of the multimodal model and ensure a fair comparison of model performance, we further compared the prediction performance of the RAFORMER-multimodal model with that of the RAFORMER-unimodal model, which only uses spectral data as input, on two bioreactors, Tank 6 and Tank 9. The evaluation comparison results for Tank 6 and Tank 9 are shown in Tables 3 and 4, respectively:

[0075] Table 3: Root mean square error between model predictions and offline measurements for bioreactor tank 6

[0076] Table 4 Root mean square error between model predictions and offline measurements for bioreactor tank 9

[0077] Comparing the performance of the RAFORMER-unimodal model and the RAFORMER-multimodal model on the two bioreactors, across all 17 test items, the RAFORMER-multimodal model achieved an average 30.40% improvement in root mean square error (RMS) over the RAFORMER-unimodal model on the 6-tank model, and an average 41.75% improvement in RMS over the RAFORMER-unimodal model on the 9-tank model. Specifically, the RAFORMER-multimodal model achieved the best performance in 16 of the 17 test items on the 6-tank model, and in 15 of the 9-tank model. Optimal performance refers to achieving the lowest RMS error between the multimodal RAFORMER and the RAFORMER-unimodal model.

[0078] This result strongly demonstrates the superior performance of the RAFORMER-multimodal model after accounting for environmental parameters relevant to cell culture experiments. Furthermore, given the diversity of culture processes across the two test bioreactors, the RAFORMER-multimodal model has greater potential for universal application across multiple clones and cell lines compared to single-modal models. This effectively overcomes the limitations of single-modal models, which focus solely on spectral data analysis, and reduces the loss of prediction accuracy.

[0079] In addition to evaluating the model's root mean square error (RMSE), we also measured the difference between the predicted trend of the RAFORMER-multimodal model and the baseline model. We used the Pearson correlation coefficient to assess the fit of this trend. Detailed evaluation results are shown in Tables 5 and 6.

[0080] Table 5: Pearson correlation coefficients for model-predicted and offline measured values ​​for bioreactor tank 6

[0081] Table 6: Pearson correlation coefficients for model-predicted and offline measured values ​​for bioreactor tank 9

[0082] Comparing the performance of the RAFORMER-multimodal model with two traditional models, GBDT and PLSR, on the two bioreactors described above revealed that for all 17 test items, the RAFORMER-multimodal model achieved an average improvement of 0.34 and 0.19 in the Pearson correlation coefficient for Tank 6 compared to GBDT and PLSR, respectively. In Tank 9, the RAFORMER-multimodal model achieved an average improvement of 0.07 and 0.08 in the root mean square error compared to GBDT and PLSR, respectively. Typical parameter trend diagrams for Tanks 6 and 9 can be found in Figures 4 and 5 (trend lines represent model predictions, and crosses indicate true values ​​obtained through offline testing).

[0083] In step S230 , spectrum data of a preset length of the target detection cell culture fluid at a target time and preset cell culture environment parameters are acquired.

[0084] Specifically, the target detection cell culture fluid can be any cell culture fluid in a biological reaction process in which the concentration of cell metabolites in the cell culture fluid needs to be analyzed. The target moment can be any moment in the biological reaction process in which the target detection cell culture fluid is detected.

[0085] The spectral data of the preset length and the preset cell culture environment parameters can be model input data that has undergone data preprocessing and has the same input paradigm as that in the above-mentioned model training process.

[0086] In step S240 , the spectral data and the preset cell culture environment parameters are input into the target concentration prediction model to obtain a metabolite concentration prediction result corresponding to the target detection cell culture fluid at the target time.

[0087] The trained target concentration prediction model can be directly used in the same application scenario to predict the concentration of cell metabolites in the cell culture fluid of any bioreactor where concentration prediction is required.

[0088] The technical solution of this embodiment is to obtain spectral sample data of the concentrations of different components of the cell culture fluid at different times, corresponding preset cell culture environment parameter sample data and offline concentration detection values; use the spectral sample data and the preset cell culture environment parameter sample data as model training input data, and use the offline concentration detection values ​​at the corresponding time as model training label data, to train a preset initial concentration prediction model to obtain a target concentration prediction model, wherein the target concentration prediction model includes multiple data feature extraction modules for extracting data features from different dimensions; and then apply the trained model to obtain spectral data of a preset length and preset cell culture environment parameters of the target detection cell culture fluid at a target time, input the spectral data and preset cell culture environment parameters into the target concentration prediction model, and obtain a metabolite concentration prediction result corresponding to the target detection cell culture fluid at the target time. The technical solution of the disclosed embodiment solves the problem of inaccurate prediction results of existing cell culture fluid concentration analysis models. It can perform data feature analysis based on the multimodal data of the tested cell culture fluid to obtain more accurate cell culture fluid concentration prediction results. In particular, the self-attention mechanism is used to analyze the spectral data of a preset length of the target detection cell culture fluid at the target moment, which can capture deeper data features to reflect the final prediction results.

[0089] Figure 6 is a flow chart of a method for determining metabolite concentrations in a cell culture fluid according to some embodiments of the present disclosure. This embodiment, which shares the same inventive concept as the method for determining metabolite concentrations in a cell culture fluid described above, further illustrates the application of cell culture fluid concentration prediction results, particularly the use of bioreactors based on the predicted results. This method can be performed by a device for determining metabolite concentrations in a cell culture fluid, which can be implemented using software and / or hardware and integrated into a computer device with application development capabilities.

[0090] As shown in FIG6 , the method for determining the concentration of metabolites in a cell culture fluid of this embodiment includes the following steps S310 to S340 .

[0091] In step S310 , spectrum data of a preset length of a target detection cell culture fluid at a target time and preset cell culture environment parameters are acquired.

[0092] In step S320, the spectral data and the preset cell culture environment parameters are input into the target concentration prediction model to obtain the metabolite concentration prediction result corresponding to the target detection cell culture fluid at the target time.

[0093] Specifically, the target concentration prediction model can include multiple data feature extraction modules, each configured to extract features from spectral data and preset cell culture environment parameters from different feature analysis dimensions. In some embodiments, the multiple data feature extraction modules include a data feature extraction module configured to extract spectral features between spectral data in different frequency bands within the spectral data using a self-attention mechanism. In some embodiments, the multimodal high-order features can then be fused to predict the metabolite concentration of the target cell culture fluid at a target time.

[0094] In step S330 , a control strategy for operating the target bioreactor where the target detection cell culture fluid is located is determined based on the concentration prediction result of at least one substance component in the metabolite concentration prediction result.

[0095] For example, metabolite concentration predictions can include information such as carbon dioxide concentration, various nutrient concentrations, and cell concentration. This concentration information can also be used to determine osmotic pressure, temperature, and aeration and stirring conditions. When the corresponding adjustment parameter indicators for each concentration value do not meet standard conditions, adjustments can be made to generate a control strategy.

[0096] In some examples, the control strategy of the bioreactor may include a culture environment control strategy and a feeding strategy. For example, the culture environment control strategy may include a carbon dioxide partial pressure strategy and a temperature reduction strategy.

[0097] For example, during cell culture, carbon dioxide is both a metabolic product of the cells and an essential component for cell growth, and is also related to maintaining the pH of the culture medium. During the cell culture process, as the amount of carbon dioxide released by metabolism increases, the culture medium will become acidic, and alkaline solution is added or ventilation is increased to maintain the pH of the culture medium. On the contrary, the culture medium will be alkaline, and carbon dioxide can be actively introduced to maintain the pH. By using parameters such as carbon dioxide concentration and pH in the metabolite concentration prediction results, the carbon dioxide concentration can be automatically increased or decreased according to the corresponding strategy, so that the adjustment is more real-time and the adjustment accuracy is higher.

[0098] For example, when viable cell density reaches a certain level during cell culture, actively lowering the culture temperature can effectively maintain cell viability and increase target protein production. The temperature can be adjusted promptly based on parameters such as cell density from metabolite concentration prediction results.

[0099] In addition, the cell culture process requires the addition of corresponding nutrients based on the cell's nutrient consumption rate. Due to the lack of real-time feedback of offline detection data, the feeding strategy can only be set based on historical data and the experimenter's experience to set the number of feedings, feeding ratio, and glucose concentration during the cell culture cycle. After the operator calculates the feeding parameters according to the empirical formula, the operator manually adds the feed to the bioreactor using a scale and a peristaltic pump. Based on the real-time metabolite concentration prediction results in this embodiment, timely supplementation can be carried out according to the preset feeding rules to improve the success rate of cell culture.

[0100] At step S340 , the target bioreactor is operated based on the control strategy.

[0101] By synchronizing the control strategy determined in the above steps and executing it in the target bioreactor, the corresponding control objectives can be achieved, ensuring the smooth progress of the biological reaction. For a more detailed description, see the process shown in Figure 3. After predicting multiple test item components, a corresponding control strategy can be generated based on the prediction results. The corresponding instructions for the control strategy are then issued to the target bioreactor for feedback control. Simultaneously, sampling and testing of the cell culture fluid in the bioreactor can be performed to increase historical data in the database.

[0102] In some instances, Raman spectral data can be transmitted in real time via an IoT platform to a computer device (which can be an edge computing gateway device) equipped with a self-attention encoder model (target concentration prediction model). The model predicts component indicators to generate status values ​​for various indicators, and the bioreactor is automatically controlled according to the status values, such as automatic feedback control of the glucose concentration in the bioreactor. In some embodiments, the target concentration prediction model can be deployed on an edge device, and a Raman spectrometer, a Raman industrial computer, an edge computing gateway, an IoT platform, an industrial computer, and a bioreactor are integrated to form an artificial intelligence IoT device for monitoring and reverse control of the bioreactor, as shown in Figure 7. The artificial intelligence IoT device can capture Raman spectral data in real time to predict the concentration of various components, calculate the status value by combining the concentration with other relevant parameters, and implement the corresponding automated control strategy to achieve automated control of cell culture in the bioreactor.

[0103] Furthermore, for the comparison of changes in glucose concentration in the bioreaction process control performed by the system described in FIG7 and the manual bioreaction process control, reference can be made to FIG8 and FIG9. FIG8 shows a schematic diagram of the glucose concentration of the model feedback control and the peristaltic pump matching within 8 days. The glucose of the model feedback control follows the set value of 4.0 g / L. FIG9 shows a schematic diagram of the manually controlled glucose following the experimental design range of 6-8 g within 11 days. Within 11 days, the manual control feed was 7 times with a total amount of 270 g, and the automatic multiple small amounts of feed were a total amount of 216 g, so while controlling in real time, the amount of feed was reduced by about 20%. The yield of the target protein did not decrease, but increased by about 4.03%. It can be determined that the control strategy for automatically controlling the bioreactor based on the target concentration prediction model in this embodiment is better.

[0104] The technical solution of this embodiment obtains spectral data of a preset length and preset cell culture environment parameters of a target detection cell culture fluid at a target time; inputs the spectral data and preset cell culture environment parameters into a target concentration prediction model to obtain a metabolite concentration prediction result corresponding to the target detection cell culture fluid at the target time; determines a control strategy for operating a target bioreactor where the target detection cell culture fluid is located based on the concentration prediction result of at least one substance component in the metabolite concentration prediction result; and operates the target bioreactor based on the control strategy. The technical solution of the disclosed embodiment solves the problem of inaccurate prediction results of existing cell culture fluid concentration analysis models. It can perform data feature analysis based on multimodal data of the tested cell culture fluid to obtain a more accurate cell culture fluid concentration prediction result. In particular, the spectral data of the preset length of the target detection cell culture fluid at the target time is analyzed based on the self-attention mechanism, which can capture deeper data features to reflect the final prediction result. In addition, the bioreactor can be adjusted in real time, more accurately, and automatically to ensure the smooth progress of the biological reaction.

[0105] Figure 10 is a schematic structural diagram of a cell culture fluid metabolite concentration determination device according to some embodiments of the present disclosure. This embodiment can be applied to biological experiment scenarios, especially when it is necessary to predict the cell metabolite concentration during the cell culture process in a bioreactor. The cell culture fluid metabolite concentration determination device can be implemented by software and / or hardware and integrated into a computer terminal device with application development capabilities.

[0106] As shown in FIG10 , the apparatus for determining the concentration of metabolites in a cell culture fluid includes a data acquisition unit 410 and a concentration prediction unit 420 .

[0107] The data acquisition unit 410 is configured to acquire spectral data of a predetermined length and predetermined cell culture environment parameters of a target cell culture fluid at a target time. The concentration prediction unit 420 is configured to input the spectral data and predetermined cell culture environment parameters into a target concentration prediction model to obtain a predicted metabolite concentration of the target cell culture fluid at the target time. The target concentration prediction model includes multiple data feature extraction modules, each of which is configured to extract features from the spectral data and predetermined cell culture environment parameters using different feature analysis dimensions.

[0108] The technical solution of this embodiment obtains spectral data of a preset length and preset cell culture environment parameters of the target detection cell culture fluid at a target time; inputs the spectral data and preset cell culture environment parameters into a target concentration prediction model to obtain a predicted result of the metabolite concentration corresponding to the target detection cell culture fluid at the target time; the target concentration prediction model includes multiple data feature extraction modules, each of which is used to extract features from the spectral data and preset cell culture environment parameters from different feature analysis dimensions. The technical solution of the disclosed embodiment solves the problem of inaccurate prediction results of existing cell culture fluid concentration analysis models, and can perform data feature analysis based on multimodal data of the tested cell culture fluid to obtain a more accurate prediction result of the cell culture fluid concentration.

[0109] In some embodiments, the plurality of data feature extraction modules include a first data feature extraction module, a second data feature extraction module, and a third data feature extraction module;

[0110] Wherein, the first data feature extraction module is used to perform feature coding analysis on the preset cell culture environment parameters to obtain target environment parameter features;

[0111] The second data feature extraction module is used to extract spectral features between spectral data of different frequency bands in the spectral data;

[0112] The third data feature extraction module is used to extract at least one spectral feature of the spectral data from among a preset spectral statistic feature, a preset spectrum feature, and a preset spectral peak analysis feature.

[0113] In some embodiments, the second data feature extraction module is used to:

[0114] The spectral features between the spectral data of different frequency bands in the spectral data are extracted through a neural network module with a self-attention mechanism.

[0115] In some embodiments, the preset cell culture environment parameters include at least one parameter selected from cell line parameters, cell clone parameters, culture process parameters, and various parameters during the cell culture process.

[0116] In some embodiments, the target concentration prediction model further includes a feature fusion module and a concentration prediction module;

[0117] The feature fusion module is used to fuse the data features extracted by the multiple data feature extraction modules to obtain target fusion features;

[0118] The concentration prediction module is used to analyze the target fusion feature to obtain the metabolite concentration prediction result.

[0119] In some embodiments, the cell culture fluid metabolite concentration determination device further includes a model training module for training a target concentration prediction model. The training process of the target concentration prediction model includes:

[0120] Obtaining spectral sample data of the concentrations of different components of the cell culture solution at different times, corresponding preset cell culture environment parameter sample data, and offline concentration detection values;

[0121] The spectral sample data and the preset cell culture environment parameter sample data are used as model training input data, and the offline concentration detection value at the corresponding time is used as model training label data to train the preset initial concentration prediction model to obtain the target concentration prediction model.

[0122] In some embodiments, the cell culture fluid metabolite concentration determination device further includes a reactor operation strategy determination module, which is configured to:

[0123] A control strategy for operating a target bioreactor where the target detection cell culture fluid is located is determined based on the concentration prediction result of at least one substance component in the metabolite concentration prediction result.

[0124] The cell culture fluid metabolite concentration determination device provided in the embodiments of the present disclosure can execute the cell culture fluid metabolite concentration determination method provided in any embodiment of the present disclosure, and has the corresponding functional modules and beneficial effects of the execution method.

[0125] Figure 11 is a schematic diagram of the structure of a computer device according to some embodiments of the present disclosure. Figure 11 shows a block diagram of an exemplary computer device 12 suitable for implementing embodiments of the present disclosure. The computer device 12 shown in Figure 11 is merely an example and should not limit the functionality or scope of use of the embodiments of the present disclosure. The computer device 12 can be any terminal device with computing capabilities, such as an intelligent controller, server, mobile phone, or other terminal device.

[0126] As shown in Figure 11, computer device 12 is a general-purpose computing device. Components of computer device 12 may include, but are not limited to, one or more processors or processing units 16, system memory 28, and a bus 18 that connects various system components (including system memory 28 and processing unit 16).

[0127] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor, or a local bus using any of a variety of bus architectures. Examples of such architectures include, but are not limited to, an Industrial Standard Architecture (ISA) bus, a Micro-Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0128] The computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computer device 12, including volatile and non-volatile media, removable and non-removable media.

[0129] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The computer device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be configured to read and write to a non-removable, non-volatile magnetic medium (not shown in FIG. 11 , commonly referred to as a “hard drive”). Although not shown in FIG. 11 , a magnetic disk drive may be provided for reading and writing to a removable non-volatile magnetic disk (e.g., a “floppy disk”), as well as an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a compact disc read-only memory (CD-ROM), a digital versatile disc read-only memory (DVD-ROM), or other optical media). In these cases, each drive may be connected to the bus 18 via one or more data media interfaces. The system memory 28 may include at least one program product having a set (eg, at least one) of program modules configured to perform the functions of various embodiments of the present disclosure.

[0130] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28. Such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may include an implementation of a network environment. Program modules 42 generally implement the functions and / or methods of the embodiments described herein.

[0131] The computer device 12 can also communicate with one or more external devices 14 (e.g., a keyboard, pointing device, display 24, etc.), one or more devices that enable a user to interact with the computer device 12, and / or any device that enables the computer device 12 to communicate with one or more other computing devices (e.g., a network card, a modem, etc.). Such communication can occur via an input / output (I / O) interface 22. Furthermore, the computer device 12 can communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with the other modules of the computer device 12 via a bus 18. It should be understood that, although not shown in FIG. 11 , other hardware and / or software modules can be used in conjunction with the computer device 12, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, redundant independent disk arrays (RAID) systems, tape drives, and data backup storage systems.

[0132] The processing unit 16 executes various functional applications and data processing by running programs stored in the system memory 28, such as implementing the method for determining the concentration of metabolites in a cell culture fluid provided in an embodiment of the present disclosure, which includes:

[0133] Acquiring spectral data of a preset length of a target detection cell culture fluid at a target time and preset cell culture environment parameters;

[0134] Inputting the spectral data and preset cell culture environment parameters into a target concentration prediction model to obtain a metabolite concentration prediction result corresponding to the target detection cell culture fluid at the target time;

[0135] The target concentration prediction model includes a plurality of data feature extraction modules, and the plurality of data feature extraction modules are respectively used to extract features of the spectral data and preset cell culture environment parameters from different feature analysis dimensions.

[0136] The present disclosure also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the method for determining the concentration of a metabolite in a cell culture fluid as provided in any embodiment of the present disclosure is implemented. The method comprises:

[0137] Acquiring spectral data of a preset length of a target detection cell culture fluid at a target time and preset cell culture environment parameters;

[0138] Inputting the spectral data and preset cell culture environment parameters into a target concentration prediction model to obtain a metabolite concentration prediction result corresponding to the target detection cell culture fluid at the target time;

[0139] The target concentration prediction model includes a plurality of data feature extraction modules, and the plurality of data feature extraction modules are respectively used to extract features of the spectral data and preset cell culture environment parameters from different feature analysis dimensions.

[0140] The computer storage medium of the embodiment of the present disclosure may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination thereof. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device.

[0141] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0142] The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, radio frequency (RF), etc., or any suitable combination thereof.

[0143] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0144] Those skilled in the art will appreciate that the modules or steps of the present disclosure described above can be implemented using a general-purpose computing device. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Alternatively, they can be implemented using program code executable by a computer device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. Thus, the present disclosure is not limited to any particular combination of hardware and software.

[0145] Note that the above are only preferred embodiments of the present disclosure and the technical principles used. Those skilled in the art will understand that the present disclosure is not limited to the specific embodiments described herein, and that various obvious changes, readjustments, and substitutions can be made by those skilled in the art without departing from the scope of protection of the present disclosure. Therefore, although the present disclosure has been described in more detail through the above embodiments, the present disclosure is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present disclosure, and the scope of the present disclosure is determined by the scope of the appended claims.

Claims

1. A method for determining the concentration of metabolites in a cell culture fluid, comprising: Acquiring spectral data of a preset length of a target detection cell culture fluid at a target time and preset cell culture environment parameters; Inputting the spectral data and the preset cell culture environment parameters into a target concentration prediction model to obtain a metabolite concentration prediction result corresponding to the target detection cell culture fluid at the target time; Wherein, the target concentration prediction model includes a plurality of data feature extraction modules, and the plurality of data feature extraction modules are respectively used to extract features of the spectral data and the preset cell culture environment parameters from different feature analysis dimensions.

2. The method according to claim 1, wherein: The multiple data feature extraction modules include a first data feature extraction module, a second data feature extraction module and a third data feature extraction module; Wherein, the first data feature extraction module is used to perform feature coding analysis on the preset cell culture environment parameters to obtain target environment parameter features; The second data feature extraction module is used to extract spectral features between spectral data of different frequency bands in the spectral data; The third data feature extraction module is used to extract at least one spectral feature of the spectral data from among a preset spectral statistic feature, a preset spectrum feature, and a preset spectral peak analysis feature.

3. The method according to claim 2, wherein: The second data feature extraction module is used for: The spectral features between the spectral data of different frequency bands in the spectral data are extracted through a neural network module with a self-attention mechanism.

4. The method according to claim 1, wherein: The preset cell culture environment parameters include at least one parameter among cell line parameters, cell clone parameters, culture process parameters and various parameters in the cell culture process.

5. The method according to any one of claims 1 to 4, wherein: The target concentration prediction model also includes a feature fusion module and a concentration prediction module; The feature fusion module is used to fuse the data features extracted by the multiple data feature extraction modules to obtain target fusion features; The concentration prediction module is used to analyze the target fusion feature to obtain the metabolite concentration prediction result.

6. The method according to any one of claims 1 to 4, wherein: The training process of the target concentration prediction model includes: Obtaining spectral sample data of the concentrations of different components of the cell culture solution at different times, corresponding preset cell culture environment parameter sample data and offline concentration detection values; The spectral sample data and the preset cell culture environment parameter sample data are used as model training input data, and the offline concentration detection value at the corresponding time is used as model training label data to train the preset initial concentration prediction model to obtain the target concentration prediction model.

7. The method according to any one of claims 1 to 4, further comprising: A control strategy for operating a target bioreactor where the target detection cell culture fluid is located is determined based on the concentration prediction result of at least one material component in the metabolite concentration prediction result.

8. A device for determining the concentration of metabolites in a cell culture fluid, comprising: A data acquisition unit, used to acquire spectral data of a preset length of a target detection cell culture fluid at a target time and preset cell culture environment parameters; A concentration prediction unit, used for inputting the spectral data and the preset cell culture environment parameters into a target concentration prediction model to obtain a prediction result of the metabolite concentration of the target detection cell culture fluid corresponding to the target time; Wherein, the target concentration prediction model includes a plurality of data feature extraction modules, and the plurality of data feature extraction modules are respectively used to extract features of the spectral data and the preset cell culture environment parameters from different feature analysis dimensions.

9. A computer device, wherein: The computer device comprises: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method for determining the metabolite concentration of a cell culture fluid according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the method for determining the concentration of metabolites in a cell culture fluid according to any one of claims 1 to 7 is implemented.

Images