A method and system for automatic mining and analysis of cell culture process data

Abstract

The present application relates to the technical field of cell culture process data mining, and discloses a method and system for automatic mining and analysis of cell culture process data, the method comprising: extracting features based on the recorded data of multiple process variables of multiple culture groups in the cell culture process to obtain a feature set; determining the features to be evaluated and multiple feature subsets in the feature set; calculating the importance of indicators based on the number of features in the feature set, the number of features in each feature subset, and the common contribution and individual contribution of each feature subset to obtain indicator evaluation data; determining the target feature based on the indicator evaluation data, and using the process variable corresponding to the target feature as the target variable that causes the difference between the first performance data and the second performance data. The present application effectively reduces the complexity of data analysis of the cell culture process, and improves the credibility of the analysis results by outputting quantitative analysis results.

Description

Technical Field

[0001] The present application relates to the technical field of cell culture process data mining, and in particular to a method and system for automatic mining and analysis of cell culture process data. Background Art

[0002] In the fields of tumor treatment, anti-viral infection, anti-transplant rejection and autoimmune diseases, antibody drugs have been widely used in practical applications due to their targeting, low toxicity and multiple mechanisms of action. In the production process of antibody drugs, cell culture is an important process. The quality of the cell culture process will affect the key process indicators such as the expression amount and charge variants of the antibody protein, which will directly affect the production of antibody drugs.

[0003] However, in the actual cell culture process, key process indicators are jointly affected by multiple cell culture process variables, and the joint influence relationship of process variables on key process indicators cannot be directly obtained, and usually needs to be determined through experiments. Due to cost constraints, it is difficult to determine the joint influence relationship of process variables through experiments in large-scale cell culture, and the experimental results of small and medium-scale cell culture are not applicable to large-scale cell culture. Therefore, it is urgent to propose a method that can determine the joint influence relationship of process variables in order to optimize the cell culture process in a targeted manner. Summary of the invention

[0004] The present application provides a method and system for automatic mining and analysis of cell culture process data, which can automatically extract features and evaluate the importance of cell culture process data, effectively reducing the complexity of data analysis of the cell culture process, and outputting target variables that cause a decrease in culture group yield evaluation data through quantitative analysis, thereby improving the credibility of the analysis results.

[0005] In a first aspect, an embodiment of the present application provides a method for automatic mining and analysis of cell culture process data, the method comprising:

[0006] Extracting features based on recorded data of multiple process variables of multiple culture groups during cell culture to obtain a feature set; wherein the feature set includes data features of each process variable; the classification results of the multiple culture groups include a first culture group and a second culture group, and first performance data of the first culture group on a yield evaluation index is better than second performance data of the second culture group on the yield evaluation index;

[0007] Determining a feature to be evaluated and a plurality of feature subsets in the feature set; wherein any feature subset is any subset of all features in the feature set except the feature to be evaluated;

[0008] Determine the common contribution and individual contribution of any feature subset; wherein the common contribution is the contribution of the feature to be evaluated and any feature subset to the classification of the culture group together, and the individual contribution is the contribution of any feature subset to the classification of the culture group alone;

[0009] Calculating the index importance according to the number of features in the feature set, the number of features in each of the feature subsets, and the common contribution and individual contribution of each of the feature subsets, to obtain index evaluation data of the importance of the feature to be evaluated in the culture group classification;

[0010] A target feature is determined in the feature set according to the indicator evaluation data, and a process variable corresponding to the target feature is used as a target variable causing a difference between the first performance data and the second performance data.

[0011] The automatic mining and analysis method for cell culture process data proposed in the embodiment of the present application performs feature extraction on the recorded data of the process variables, and performs index importance calculation on the extracted data features, so as to determine the index evaluation data representing the importance of each data feature in the culture group classification, so as to determine the target variable causing the difference in yield evaluation index according to the index evaluation data. Compared with the related art, the present application sets the index evaluation data for representing the importance of the data features in the culture group classification, so as to automatically perform quantitative analysis on the influence of the data features on the cell culture process, so as to determine the target variable from all process variables, thereby reducing the complexity of data analysis on the recorded data of the cell culture process, and improving the efficiency of data analysis. In addition, since the target variable as the analysis result is obtained through quantitative analysis, the present application can also improve the credibility and accuracy of the analysis result, so as to provide a reliable data basis for the optimization of the cell culture process.

[0012] Optionally, the index importance calculation is performed according to the number of features in the feature set, the number of features in each of the feature subsets, and the common contribution and individual contribution of each of the feature subsets to obtain the index evaluation data of the importance of the feature to be evaluated in the culture group classification, including:

[0013] The common contribution and the individual contribution of each feature subset are subtracted to obtain the classification contribution of the feature to be evaluated to the classification of the culture group;

[0014] Obtaining a number of feature subsets based on the number of features in the feature set and the number of features in each of the feature subsets;

[0015] According to the number of feature subsets, weight adjustment is performed on the classification contribution of the feature to be evaluated to the classification of the culture group to obtain the adjusted contribution corresponding to each feature subset;

[0016] The adjusted contribution corresponding to each of the feature subsets is summed and averaged according to the number of features in the feature set to obtain index evaluation data of the importance of the feature to be evaluated in the culture group classification.

[0017] Optionally, the index evaluation data of the importance of the feature to be evaluated in the culture group classification is obtained by the following formula:

[0018]

[0019] in, Indicates indicator evaluation data; is the feature to be evaluated; is a feature set; is the number of features in the feature set; is a set of features to be evaluated that are excluded from the feature set; is the union of the feature subset and the feature to be evaluated; is a feature subset; is the number of features in the feature subset; For collection The number of feature subsets; is the common contribution of the feature subset and the feature to be evaluated; is the individual contribution of the feature subset.

[0020] Optionally, determining the common contribution and the individual contribution of any feature subset includes:

[0021] Inputting the data feature combination including the feature to be evaluated and the feature subset into a culture group classification model to obtain the common contribution output by the culture group classification model; wherein the culture group classification model is obtained by model training using the feature set;

[0022] The data features of the feature subset are individually input into the culture group classification model to obtain the individual contribution output by the culture group classification model.

[0023] Optionally, the recorded data includes sensor monitoring data corresponding to the sensor monitoring variable and offline inspection data corresponding to the offline inspection variable; the feature extraction is performed based on the recorded data of multiple process variables of multiple culture groups during the cell culture process to obtain a feature set, including:

[0024] Determining the variable type of the sensor monitoring variable according to the change trend of the sensor monitoring data;

[0025] Extracting features from the sensor monitoring data according to the variable type to obtain sensor data features;

[0026] The offline inspection data is used as offline data features; wherein the feature set includes the sensor data features and the offline data features.

[0027] Optionally, determining the variable type of the sensor monitoring variable according to the change trend of the sensor monitoring data includes:

[0028] Cutting and trend sampling the recorded data corresponding to the sensor monitoring variable to obtain a plurality of data subsequences with the same time length;

[0029] Selecting any number of the data subsequences to form multiple training data sets, using the multiple training data sets to perform model training respectively, and obtaining multiple classification models of variables to be screened corresponding to the multiple training data sets;

[0030] Screening the classification results of the sensor monitoring variables according to the classification models of the variables to be screened to obtain a target variable classification model;

[0031] The variable type of the sensor monitoring variable is determined according to the classification result of the target variable classification model on the sensor monitoring variable.

[0032] Optionally, the variable type includes a stable variable, a step variable, a slowly changing variable and a pulse variable; and extracting features from the sensor monitoring data according to the variable type to obtain sensor data features includes:

[0033] extracting the mean and standard deviation of the stable variable during the cell culture process as the sensing data features of the stable variable;

[0034] Performing piecewise linear regression on the step variable, obtaining the mutation node of the step variable in the cell culture process according to the fitting slope of each of the data subsequences; dividing the sensor monitoring data corresponding to the step variable according to the mutation node to obtain step segments; extracting the mean value and linear fitting slope of each step segment as the sensor data feature of the step variable;

[0035] Performing basis function fitting on the slowly varying variable to obtain a slowly varying fitting result; extracting coefficients of each basis function in the slowly varying fitting result as sensing data features of the slowly varying variable;

[0036] A cumulative value curve is obtained according to the pulse variable, and basis function fitting is performed on the cumulative value curve to obtain a curve fitting result; the sensor monitoring data corresponding to the pulse variable is divided using the nodes of each basis function in the curve fitting result to obtain pulse segments; the coefficients of each basis function in the curve fitting result and the mean value of each pulse segment are extracted as the sensor data features of the pulse variable.

[0037] Optionally, the method further comprises:

[0038] Performing normalization and Fourier transformation on the recorded data to obtain a logarithmic amplitude spectrum of the recorded data;

[0039] Smoothing the logarithmic amplitude spectrum to obtain a smoothed logarithmic amplitude spectrum;

[0040] Performing exponential transformation and inverse Fourier transformation on the smoothed logarithmic amplitude spectrum and the smoothed loss vector between the logarithmic amplitude spectrum to obtain a deviation index;

[0041] According to the comparison result of the deviation index and the preset threshold value, the occurrence time of the data deviation point in the process data is determined to eliminate and correct the data deviation point.

[0042] In a second aspect, an embodiment of the present application provides a system for automatic mining and analysis of cell culture process data, the system comprising:

[0043] A data feature extraction module, used for extracting features based on the recorded data of multiple process variables of multiple culture groups during the cell culture process to obtain a feature set; wherein the feature set includes the data features of each process variable; the classification results of the multiple culture groups include a first culture group and a second culture group, and the first performance data of the first culture group on the yield evaluation index is better than the second performance data of the second culture group on the yield evaluation index;

[0044] A feature subset determination module, used to determine a feature to be evaluated and a plurality of feature subsets in the feature set; wherein any feature subset is any subset of all features in the feature set except the feature to be evaluated;

[0045] A contribution determination module, used to determine the common contribution and individual contribution of any feature subset; wherein the common contribution is the contribution of the feature to be evaluated and any feature subset to the classification of the culture group, and the individual contribution is the contribution of any feature subset to the classification of the culture group alone;

[0046] An index evaluation calculation module is used to calculate the index importance according to the number of features in the feature set, the number of features in each of the feature subsets, and the common contribution and individual contribution of each of the feature subsets, to obtain index evaluation data of the importance of the feature to be evaluated in the culture group classification;

[0047] A target variable determination module is used to determine a target feature in the feature set according to the indicator evaluation data, and use a process variable corresponding to the target feature as a target variable that causes a difference between the first performance data and the second performance data.

[0048] In a third aspect, an embodiment of the present application provides an automatic mining and analysis system for cell culture process data, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the method described in any one of the above embodiments by executing the computer instructions.

[0049] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to enable a computer to execute any one of the methods in the above embodiments.

[0050] In a fifth aspect, an embodiment of the present application provides a computer program product, including computer instructions, and the computer instructions are used to enable a computer to execute any one of the methods in the above embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the specific implementation methods of the present application or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0052] Figure 1 A step diagram of the method for automatic mining and analysis of cell culture process data provided in an embodiment of the present application;

[0053] Figure 2 A diagram showing the steps for obtaining index evaluation data in an embodiment of the present application;

[0054] Figure 3 A diagram showing the steps for obtaining the common contribution and the individual contribution in the embodiment of the present application;

[0055] Figure 4 A diagram showing the steps of obtaining a feature set in an embodiment of the present application;

[0056] Figure 5 A diagram showing the steps of determining the variable type of the sensor monitoring variable in an embodiment of the present application;

[0057] Figure 6 A diagram showing the steps of detecting and correcting data deviation points in an embodiment of the present application;

[0058] Figure 7 This is a schematic diagram of the yield evaluation indicators of each culture group in the application scenario example of this application;

[0059] Figure 8 A schematic diagram of the index evaluation data of each feature in the application scenario example of this application;

[0060] Fig. 9 for Figure 8 A magnified schematic diagram of the indicator evaluation data of the pH value;

[0061] Fig.10 This is a graph showing the changing trend of pH values ​​of each culture group during the cell culture process in the application scenario example of this application;

[0062] Fig.11 A module diagram of the automatic mining and analysis system for cell culture process data provided in an embodiment of the present application;

[0063] Fig.12 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0065] In the fields of tumor treatment, anti-viral infection, anti-transplant rejection and autoimmune diseases, antibody drugs have been widely used in practical applications due to their targeting, low toxicity and multiple mechanisms of action, and have a large market size. In the production process of antibody drugs, cell culture is an important process. The quality of the cell culture process will affect key process indicators such as antibody protein expression and charge variants. The expression of antibody protein is related to the output of antibody drugs. The decrease in the expression of antibody protein will directly reduce the output of drugs under a fixed batch quantity, resulting in the production efficiency of pharmaceutical manufacturers being affected.

[0066] The actual cell culture process contains a large number of biochemical reactions and physical changes, and involves a variety of process variables. Different process variables may have synergistic or interactive effects, which in turn have a highly nonlinear effect on the cell culture results. Therefore, in order to determine the influence relationship between process variables and cell culture results, a large number of experiments are usually required for specific analysis. However, due to cost constraints, it is difficult to determine the influence relationship between process variables and cell culture results through experiments in large-scale cell culture, and the experimental results of small- and medium-scale cell culture are not applicable to large-scale cell culture.

[0067] In the related art, the data analysis method of the cell culture process usually requires professional analysts to clean and preprocess the cell culture process data, and select appropriate feature extraction methods based on the experience of the analysts to extract features from the cell culture process data. Based on the obtained data features, the analysts determine the main factors causing the decrease in production through cluster analysis or comparative analysis. However, since many steps in the above process rely on the professional experience of the analysts themselves, the analysis process has a certain degree of randomness and is greatly interfered by human factors, resulting in reduced accuracy and effectiveness of the analysis results, and manual data analysis also limits the efficiency of data analysis.

[0068] Based on the above problems, the present application provides a method and system for automatic mining and analysis of cell culture process data, wherein the method includes: extracting features based on the recorded data of process variables of the culture group during the cell culture process to obtain a feature set; determining the features to be evaluated and multiple feature subsets in the feature set; determining the common contribution and individual contribution of any feature subset; calculating the importance of indicators based on the number of features in the feature set, the number of features in each feature subset, and the common contribution and individual contribution of each feature subset to obtain indicator evaluation data; determining the target feature based on the indicator evaluation data, and taking the process variable corresponding to the target feature as the target variable that causes the difference between the first performance data and the second performance data.

[0069] The automatic mining and analysis method for cell culture process data proposed in the present application extracts features from the recorded data of process variables and calculates the indicator importance of the extracted data features, thereby determining the indicator evaluation data representing the importance of each data feature in the culture group classification, and determining the target variable that causes the difference in yield evaluation indicators based on the indicator evaluation data.

[0070] Compared with the related art, the present application sets indicator evaluation data to represent the importance of data features in culture group classification, so as to automatically perform quantitative analysis on the impact of data features on the cell culture process, so as to determine the target variable from all process variables, thereby reducing the complexity of data analysis on the recorded data of the cell culture process and improving the efficiency of data analysis.

[0071] In addition, since the target variable as the analysis result is obtained through quantitative analysis, the present application can also improve the credibility and accuracy of the analysis results to provide a reliable data basis for the optimization of the cell culture process.

[0072] The cell culture process data automatic mining and analysis method provided in this specification can be applied to automatic data analysis of the cell culture process, and the object of cell culture can be a cell culture group for producing therapeutic antibody drugs, etc. It is understandable that this application can also perform data analysis on cell culture processes in other fields after adaptively modifying the selection of process variables or the calculation method of indicator evaluation data.

[0073] According to an embodiment of the present application, an embodiment of a method for automatic mining and analysis of cell culture process data is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0074] In this embodiment, a method for automatic mining and analysis of cell culture process data is provided, which can be used to automatically analyze the data of the cell culture process. Figure 1 As shown, the method includes:

[0075] S100. Perform feature extraction based on the recorded data of multiple process variables of multiple culture groups during the cell culture process to obtain a feature set; wherein the feature set includes data features of each process variable; the classification results of the multiple culture groups include a first culture group and a second culture group, and the first performance data of the first culture group on the yield evaluation index is better than the second performance data of the second culture group on the yield evaluation index.

[0076] S200. Determine a feature to be evaluated and multiple feature subsets in the feature set; wherein any feature subset is any subset of all features in the feature set except the feature to be evaluated.

[0077] S300. Determine the common contribution and individual contribution of any feature subset; wherein the common contribution is the contribution of the feature to be evaluated and any feature subset to the classification of the culture group, and the individual contribution is the contribution of any feature subset to the classification of the culture group alone.

[0078] S400. Calculate the index importance according to the number of features in the feature set, the number of features in each feature subset, and the common contribution and individual contribution of each feature subset to obtain index evaluation data of the importance of the feature to be evaluated in the culture group classification.

[0079] S500. Determine a target feature in a feature set according to the indicator evaluation data, and use a process variable corresponding to the target feature as a target variable causing a difference between the first performance data and the second performance data.

[0080] Among them, the culture group can be a group of cells and culture fluid that undergo the same cell culture process. The cells in any culture group have the same process variable changes in the same cell culture batch and correspond to the same cell culture results. The process variable can be a variable whose value changes over time during the cell culture process. The value of the process variable can be obtained by collecting data from the culture group. The first culture group and the second culture group can be obtained by dividing all culture groups based on the yield threshold according to the performance data of the culture group on the yield evaluation index. The first culture group and the second culture group correspond to different cell culture batches respectively. It can be understood that the performance data of the cell culture batch corresponding to the first culture group is different from the performance data of the cell culture batch corresponding to the second culture group.

[0081] Among them, the yield evaluation index can be the actual protein expression level of the culture group after the cell culture process ends. The actual protein expression level of each culture group is detected to obtain the performance data of each culture group on the yield evaluation index, and the culture group whose performance data exceeds the yield threshold is divided into the first culture group, and the culture group whose performance data does not exceed the yield threshold is divided into the second culture group.

[0082] A feature subset is any subset of all features in the feature set except the feature to be evaluated. There can be multiple feature subsets, including combinations of different numbers of other features. The common contribution can be the contribution of the union of the feature to be evaluated and any feature subset in the culture group classification, which indicates the degree of influence of any feature subset and the feature to be evaluated on the culture group classification when the feature to be evaluated is included. The individual contribution can be the contribution of any feature subset in the culture group classification, which indicates the degree of influence of any feature subset independently on the culture group classification when the feature to be evaluated is not included.

[0083] Specifically, during the cell culture process, data is collected for the process variables of each culture group to obtain recorded data with a time span covering the complete cell culture process. After the cell culture process is completed, the cell culture results of each culture group are analyzed to determine the performance data of each culture group on the yield evaluation index, and the culture groups are classified according to the performance data to obtain the first culture group and the second culture group. Feature extraction is performed on the recorded data of the first culture group to obtain the feature set of the first culture group; feature extraction is performed on the recorded data of the second culture group to obtain the feature set of the second culture group. It can be understood that the data features in the feature set of the first culture group have the label of the first culture group, and the data features in the feature set of the second culture group have the label of the second culture group. Each data feature in the feature set corresponds to a process variable, which is used to represent the change characteristics of the process variable during the cell culture process. Each process variable can correspond to multiple data features, and the number of data features corresponding to different process variables can be different.

[0084] Furthermore, when evaluating the importance of any feature to be evaluated, remove the feature to be evaluated from the feature set, thereby obtaining the feature complement corresponding to the feature to be evaluated. All subsets of the feature complement are respectively used as feature subsets. It should be noted that the number of feature subsets is related to the number of features in the feature set and the number of features in each feature subset, and the number of feature subsets can represent the dependence of the culture group classification on any feature to be evaluated. The more feature subsets there are, the more data features there are still for the culture group classification after removing the feature to be evaluated, indicating that the dependence of the culture group classification on the feature to be evaluated is lower; the fewer feature subsets there are, the more data features used for the culture group classification after removing the feature to be evaluated are significantly reduced, indicating that the dependence of the culture group classification on the feature to be evaluated is higher.

[0085] Further, the common contribution of any feature subset is determined based on the feature to be evaluated and any feature subset, and the individual contribution of any feature subset is determined based on the any feature subset. It is understandable that the method for determining the common contribution and the individual contribution can be any one of the methods such as decision tree and random forest, gradient boosting tree, linear model, support vector machine and naive Bayes classifier. According to the common contribution and the individual contribution of any feature subset, the contribution of the feature to be evaluated compared with the any feature subset in the classification of the culture group can be obtained. On the basis of obtaining the common contribution and the individual contribution of each feature subset, the feature to be evaluated is calculated with reference to the number of features in the feature set and the number of features in each feature subset, and the index evaluation data of the importance of the feature to be evaluated in the classification of the culture group is obtained.

[0086] Furthermore, the index evaluation data of any data feature in the feature set in each culture group are summed and averaged to obtain the mean value of the index evaluation data of any data feature in all culture groups. According to the numerical value of the mean value, the data feature with the highest importance in the culture group classification is determined as the target feature, and the process variable corresponding to the target feature is used as the target variable that causes the difference between the first performance data and the second performance data.

[0087] The automatic mining and analysis method for cell culture process data provided in this embodiment extracts features from the recorded data of process variables, and calculates the indicator importance of the extracted data features, thereby determining indicator evaluation data representing the importance of each data feature in the culture group classification, and determining the target variable that causes the difference in yield evaluation indicators based on the indicator evaluation data.

[0088] Compared with the related art, the present application sets indicator evaluation data to represent the importance of data features in culture group classification, so as to automatically perform quantitative analysis on the impact of data features on the cell culture process, so as to determine the target variable from all process variables, thereby reducing the complexity of data analysis on the recorded data of the cell culture process and improving the efficiency of data analysis.

[0089] In addition, since the target variable as the analysis result is obtained through quantitative analysis, the present application can also improve the credibility and accuracy of the analysis results to provide a reliable data basis for the optimization of the cell culture process.

[0090] Reference Figure 2 As shown, as an embodiment of the present application, the indicator importance is calculated according to the number of features in the feature set, the number of features in each feature subset, and the common contribution and individual contribution of each feature subset, to obtain the indicator evaluation data of the importance of the feature to be evaluated in the culture group classification, including:

[0091] S410. Calculate the difference between the common contribution and the individual contribution of each feature subset to obtain the classification contribution of the feature to be evaluated to the classification of the culture group.

[0092] S420. Obtain the number of feature subsets based on the number of features in the feature set and the number of features in each feature subset.

[0093] S430. Adjust the weight of the classification contribution of the features to be evaluated to the classification of the training group according to the number of feature subsets to obtain the adjusted contribution corresponding to each feature subset.

[0094] S440. According to the number of features in the feature set, the adjusted contribution corresponding to each feature subset is summed and averaged to obtain indicator evaluation data of the importance of the feature to be evaluated in the culture group classification.

[0095] Specifically, the common contribution represents the contribution of any feature subset in the classification of the culture group when the feature to be evaluated is included. The individual contribution represents the contribution of any feature subset in the classification of the culture group when the feature to be evaluated is not included. By taking the difference between the common contribution and the individual contribution of each feature subset, the classification contribution of the feature to be evaluated to the classification of the culture group can be obtained. It should be noted that the classification contribution of the feature to be evaluated is based on the synergistic effect of other features other than the feature to be evaluated and the feature to be evaluated in the classification of the culture group. According to the classification contribution of each feature to be evaluated, the common influence relationship of different data features in the feature set in the classification of the culture group can be analyzed.

[0096] Furthermore, the number of data features in each feature subset may be different, and the number of non-empty subsets is calculated based on the number of features in the feature set and the number of features in each feature subset to obtain the number of feature subsets. It is understandable that for the same feature to be evaluated, feature subsets with the same number of features have the same number of feature subsets.

[0097] It should be noted that the number of feature subsets is the number of all feature subsets that can be obtained based on all other data features after removing the feature to be evaluated from the feature set. The number of feature subsets can represent the dependence of the culture group classification on any feature to be evaluated. The more feature subsets there are, the more data features there are still for the culture group classification after removing any feature to be evaluated, indicating that the dependence of the culture group classification on any feature to be evaluated is lower; the fewer feature subsets there are, the more data features used for the culture group classification after removing any feature to be evaluated are significantly reduced, indicating that the dependence of the culture group classification on any feature to be evaluated is higher.

[0098] Furthermore, the feature weight of the feature to be evaluated is obtained according to the number of feature subsets, and the classification contribution of the feature to be evaluated is weighted and adjusted according to the number of feature subsets to obtain the adjusted contribution corresponding to each feature subset. The adjusted contribution corresponding to the feature subsets obtained from the same feature to be evaluated is added, and the added result is averaged according to the number of features in the feature set to obtain the index evaluation data of the importance of the feature to be evaluated in the culture group classification. It should be noted that the index evaluation data of any feature to be evaluated includes the contribution of any feature to be evaluated in the culture group classification, and the dependence of the culture group classification on any feature to be evaluated, thereby comprehensively evaluating the importance of the feature to be evaluated in the culture group classification and improving the accuracy of data analysis of the cell culture process data.

[0099] As an embodiment of the present application, the index evaluation data of the importance of the feature to be evaluated in the culture group classification is obtained by the following formula:

[0100]

[0101] in, Indicates indicator evaluation data; is the feature to be evaluated; is a feature set; is the number of features in the feature set; is a set of features to be evaluated that are excluded from the feature set; is the union of the feature subset and the feature to be evaluated; is a feature subset; is the number of features in the feature subset; For collection The number of feature subsets; is the common contribution of the feature subset and the feature to be evaluated; is the individual contribution of the feature subset.

[0102] Specifically, the feature weight of the feature to be evaluated is the inverse of the number of feature subsets. The feature weight is used to adjust the classification contribution of the feature to be evaluated to the classification of the culture group, and the adjusted contribution corresponding to each feature subset is obtained. The adjusted contributions corresponding to the obtained feature subsets are added, and the addition results are averaged according to the number of features in the feature set to obtain indicator evaluation data of the importance of the feature to be evaluated in the culture group classification.

[0103] Reference Figure 3 As shown, as an embodiment of the present application, determining the common contribution and individual contribution of any feature subset includes:

[0104] S310. Input the data feature combination including the feature to be evaluated and the feature subset into the culture group classification model to obtain the common contribution output by the culture group classification model; wherein the culture group classification model is obtained by model training using the feature set.

[0105] S320. Input the data features of the feature subset separately into the training group classification model to obtain the individual contribution output by the training group classification model.

[0106] Specifically, the culture group classification model can be a decision tree, and the output result of the culture group classification model includes the classification result of the culture group and the contribution of the features input into the culture group classification model to the classification of the culture group. The culture group classification model can be established by a boosting learning method based on a decision tree, and the process may include: training the decision tree through the feature sets of the first culture group and the second culture group, and modifying the parameters of the decision tree according to the first culture group label or the second culture group label of each data feature in the feature set and the classification result of the decision tree, until the training is terminated when the decision tree can accurately classify the first culture group and the second culture group, and the culture group classification model is obtained.

[0107] Furthermore, the data features including the feature to be evaluated and any feature subset are input into the culture group classification model, and the output result of the culture group classification model is the common contribution of any feature subset. The data features of any feature subset are input into the culture group classification model individually, and the output result of the culture group classification model is the individual contribution of any feature subset.

[0108] Reference Figure 4As shown, as an embodiment of the present application, the recorded data includes sensor monitoring data corresponding to the sensor monitoring variable and offline inspection data corresponding to the offline inspection variable; feature extraction is performed based on the recorded data of multiple process variables of multiple culture groups during the cell culture process to obtain a feature set, including:

[0109] S110. Determine the variable type of the sensor monitoring variable according to the change trend of the sensor monitoring data.

[0110] S120. Extract features of the sensor monitoring data according to the variable type to obtain sensor data features.

[0111] S130. Use the offline inspection data as offline data features; wherein the feature set includes sensor data features and offline data features.

[0112] The sensor monitoring data is the recorded data of the sensor monitoring variables, which is obtained by regularly obtaining the data through the sensor instrument on the cell culture tank during the cell culture process. The sensor monitoring data can be one or more of the following data: pH value, culture temperature, dissolved oxygen content, bottom layer air flow rate, bottom layer oxygen flow rate, bottom layer carbon dioxide flow rate, surface layer air flow rate, culture tank pressure, stirring rate, and culture liquid volume.

[0113] Offline test data is the recorded data of offline test variables, which is obtained by taking culture fluid samples through the sampling valve of the culture tank at regular intervals during the cell culture process, and then using analytical equipment to detect and obtain the culture fluid samples. Offline test data can be one or more of cell status data such as cell density, cell viability, cell diameter, aggregation rate, and concentrations of metabolites such as glucose, lactic acid, inorganic ions, and amino acids.

[0114] It is understandable that the acquisition frequency of sensor monitoring data is higher than that of offline test data. The acquisition frequency of sensor monitoring data can be several seconds or several minutes, etc., while the acquisition frequency of offline test data can be several times a day. The time length of sensor monitoring data is related to the duration of the cell culture process; while the time length of offline test data is usually the same due to the fixed acquisition frequency.

[0115] Specifically, the variable type of the sensor monitoring variable indicates the type of driving factor that affects the sensor monitoring variable during the cell culture process, such as exogenous factors or endogenous factors that affect the sensor monitoring variable during the cell culture process. The sensor monitoring variables affected by different driving factors show different change trends during the cell culture process. Therefore, the sensor monitoring variables can be classified according to the change trend of the sensor monitoring data to determine the variable type of the sensor monitoring variable.

[0116] It should be noted that the sensor monitoring variables of each variable type have different data characteristics, so it is necessary to use the corresponding feature extraction method according to the characteristics of the variable type to extract features of the sensor monitoring variables of each variable type to obtain sensor data features. Compared with the technical solution in the related art in which the feature extraction method is manually determined by technicians, the sensor monitoring variables are automatically classified according to the changing trend of the sensor monitoring data in this application, and the corresponding feature extraction method is used to extract features according to the variable type of the sensor monitoring variable, thereby realizing the automation and standardization of the data analysis of the cell culture process, reducing the impact of human subjectivity on data analysis, and improving the accuracy of the data analysis results.

[0117] Furthermore, since the acquisition frequency of offline inspection data is lower than that of sensor monitoring data, its data volume is less than that of sensor monitoring data. When extracting offline data features, the numerical value of offline inspection data is used as offline data features. According to the fusion of sensor data features and offline data features, a feature set is obtained. The data features in the feature set can be in the form of a feature matrix, which includes a sensor data feature matrix and an offline data feature matrix.

[0118] Reference Figure 5 As shown, as an embodiment of the present application, the variable type of the sensor monitoring variable is determined according to the change trend of the sensor monitoring data, including:

[0119] S112. Perform cropping and trend sampling on the recorded data corresponding to the sensor monitoring variables to obtain multiple data subsequences with the same time length.

[0120] S114. Select any number of data subsequences to form multiple training data sets, use the multiple training data sets to perform model training respectively, and obtain multiple classification models of variables to be screened corresponding to the multiple training data sets.

[0121] S116. Screen the classification results of the sensor monitoring variables according to the classification models of the variables to be screened to obtain the classification model of the target variable.

[0122] S118. Determine the variable type of the sensor monitoring variable according to the classification result of the sensor monitoring variable by the target variable classification model.

[0123] Specifically, the start and end time of cell culture in each culture group may be different, resulting in different time lengths of sensor monitoring data. In order to improve the efficiency of data analysis, the recorded data of all culture groups are trimmed based on the culture group with the shortest time length among all culture groups to obtain a first subsequence with the same time length.

[0124] It should be noted that the classification feature space corresponding to the first subsequence is relatively large, which increases the complexity of variable classification. Therefore, in order to reduce the complexity of variable classification, in this embodiment, trend sampling is performed on the first subsequence, including fixed window sampling and random sampling strategies of random forest. Among them, fixed window sampling can be performed on the first subsequence using a fixed window with a window length of a power of 2 to obtain a second subsequence; the random sampling strategy of random forest can be performed on the second subsequence to obtain a data subsequence. The classification feature space of the data subsequence is reduced by trend sampling, which effectively reduces the complexity of variable classification.

[0125] Furthermore, any number of data subsequences are selected to form multiple training data sets for model training, and multiple classification models of variables to be screened are obtained. The classification model of variables to be screened can be a decision tree, and its splitting criterion can be combined with entropy gain and distance metric. By adding a distance metric item, the problem that the splitting results of different nodes in the decision tree may have the same entropy gain is solved.

[0126] Furthermore, based on the classification results of the sensor monitoring variables by the classification model of the variable to be screened, multiple classification models of the variable to be screened are screened, and the classification model of the variable to be screened with the highest classification accuracy is obtained as the target variable classification model, and the variable type of the sensor monitoring variable is determined according to the classification results of the target variable classification model.

[0127] As an embodiment of the present application, the variable types include stable variables, step variables, slowly changing variables and pulse variables; according to the variable types, the sensor monitoring data is feature extracted to obtain sensor data features, including:

[0128] The mean and standard deviation of the stable variables during the cell culture process were extracted as the sensing data features of the stable variables.

[0129] A piecewise linear regression is performed on the step variable, and the mutation nodes of the step variable in the cell culture process are obtained according to the fitting slopes of each data subsequence. The sensor monitoring data corresponding to the step variable is divided according to the mutation nodes to obtain the step segments. The mean and linear fitting slope of each step segment are extracted as the sensor data features of the step variable.

[0130] The slowly varying variables are fitted with basis functions to obtain slowly varying fitting results; the coefficients of each basis function in the slowly varying fitting results are extracted as the sensing data features of the slowly varying variables.

[0131] A cumulative value curve is obtained according to the pulse variable, and basis functions are fitted to the cumulative value curve to obtain a curve fitting result; the sensor monitoring data corresponding to the pulse variable is divided using the nodes of each basis function in the curve fitting result to obtain pulse segments; the coefficients of each basis function in the curve fitting result and the mean of each pulse segment are extracted as the sensor data features of the pulse variable.

[0132] Among them, a stable variable can be a sensor monitoring variable whose value remains basically stable during the cell culture process and does not change significantly. A step variable can be a sensor monitoring variable whose value remains basically stable during the cell culture process and only changes significantly for a short time at a specific time point. A slowly changing variable can be a sensor monitoring variable whose value changes all the time during the cell culture process, but the rate of change is low. A pulse variable can be a sensor monitoring variable whose value remains basically stable during the cell culture process, only changes significantly for a short time at a specific time point, and quickly returns to the state before the change.

[0133] It should be noted that, during the cell culture process, the driving factors that cause changes in sensor monitoring variables include exogenous factors and endogenous factors, among which exogenous factors may include aeration, stirring and feeding, and endogenous factors may include cell growth metabolism. Stable variables are not affected by driving factors, and their numerical values ​​in the cell culture process show stable characteristics. Step variables and pulse variables are mainly affected by exogenous factors, and their numerical values ​​in the cell culture process show characteristics corresponding to the variable type. Slowly changing variables are mainly affected by endogenous factors, and their numerical values ​​in the cell culture process show slowly changing characteristics. The sensor monitoring variables of each variable type have different data characteristics. Therefore, it is necessary to extract features of the sensor monitoring variables of each variable type using the corresponding feature extraction method according to the characteristics of the variable type to obtain sensor data features.

[0134] Reference Figure 6 As shown, as an embodiment of the present application, the method further includes:

[0135] S610. Perform normalization and Fourier transformation on the recorded data to obtain a logarithmic amplitude spectrum of the recorded data.

[0136] S620. Smooth the logarithmic amplitude spectrum of the recorded data to obtain a smoothed logarithmic amplitude spectrum.

[0137] S630. Perform exponential transformation and inverse Fourier transformation on the smoothed logarithmic amplitude spectrum and the smoothed loss vector between the logarithmic amplitude spectrum to obtain a deviation index.

[0138] S640. According to the comparison result between the deviation index and the preset threshold value, determine the occurrence time of the data deviation point in the process data to eliminate and correct the data deviation point.

[0139] Specifically, in the process of acquiring recorded data, due to the drift or communication interruption of the sensor used to collect the sensor monitoring data, short-time pulse deviation points appear in the recorded data, thereby affecting the data analysis results. In this embodiment, by performing Fourier transform on the recorded data, the frequency domain information of the recorded data can be fully utilized, reducing the influence of the change trend of the process variable in the recorded data on the detection of data deviation points.

[0140] Furthermore, the process of eliminating and correcting data deviation points includes: determining the occurrence time of the data deviation point and replacing the data deviation point with a null value; using linear interpolation to obtain a correction value of the data deviation point based on the recorded data before and after the data deviation point, and using the corrected value to replace the null value to obtain the corrected recorded data.

[0141] The deviation point detection algorithms in the related art include the standard score method and the K-nearest neighbor algorithm, etc., and the above algorithms are easily affected by the changing trend of the recorded data, resulting in misjudgment or missed judgment. In the present application, the recorded data is subjected to Fourier transform, thereby utilizing the frequency domain information of the recorded data, reducing the influence of the changing trend of the recorded data and improving the accuracy of the data deviation point detection.

[0142] Based on the automatic mining and analysis method of cell culture process data provided in the above embodiment, a practical application scenario of the method is described below. In this practical application scenario, the collected data is the cell culture process data of an enterprise producing antibody drugs on a commercial scale, and the volume of the culture tank is 2500L. The main reason for the difference between the first performance data and the second performance data is analyzed based on the cell culture process data.

[0143] In this actual application scenario, the sensor monitoring data is collected from the cell culture production workshop, and the sensor monitoring variables include 10 variables. The offline inspection data is collected from the sample testing laboratory, and the offline inspection variables include 6 variables. The specific types of sensor monitoring variables and offline inspection variables are shown in Table 1.

[0144] Table 1 List of process variables

[0145]

[0146] In this actual application scenario, the cell culture process lasts for 14 days, the sensor monitoring data is acquired every 30 seconds, and the offline test data is acquired once a day, with an acquisition interval of about 24 hours. During the complete cell culture process, the number of acquisitions of any sensor monitoring variable in any culture group is 40,000 to 41,000 times, and the number of acquisitions of any offline test variable in any culture group is 16 times.

[0147] In this actual application scenario, a total of 20 culture group records were collected. The yield evaluation data of the culture group is the protein expression level of the culture group after the cell culture process is completed. Figure 7 As shown, Figure 7 The yield evaluation index of each culture group in this actual application scenario is shown. The dotted line at 1.0 in the figure represents the yield threshold. The culture groups are classified according to the yield threshold, and it can be found that there are 12 first culture groups and 8 second culture groups in the 20 culture groups. The first culture group is recorded as cluster 1 and the second culture group is recorded as cluster 2. The first performance data of the first culture group in cluster 1 on the yield evaluation index is better than the second performance data of the second culture group in cluster 2 on the yield evaluation index.

[0148] The data deviation points of the recorded data of each culture group were detected and eliminated, and the sensor monitoring variables were classified. When classifying the variables, the recorded data were cut to 40,000 sampling points, and 500 data subsequences were randomly selected from all the recorded data with a window length of 200 sampling points. Trend sampling was performed on the data subsequences, 1,000 decision trees were trained according to the trend characteristics of the data subsequences, and voting was performed to determine the target variable classification model from the decision trees, thereby determining the variable type of the sensor monitoring variable. The variable type of the sensor monitoring variable can be referred to as shown in Table 2.

[0149] Table 2 Variable types of sensor monitoring variables

[0150]

[0151] According to the variable type of the sensor monitoring variable, the sensor monitoring data is feature extracted to obtain the sensor data features. For the step variable, the maximum and minimum values ​​of each culture group are first normalized, and then a piecewise linear regression with a size of 200 and a step size of 1 is created to extract the fitting slope of each data subsequence as the mutation node judgment indicator, and the judgment threshold is set to 0.0003. After determining the mutation node, the sensor monitoring data is divided according to the mutation node to obtain the step segmentation, and the mean and linear fitting slope of each step segment are extracted as the sensor data features of the step variable.

[0152] For slowly varying variables, 100 B-spline basis functions are used to fit the sensor monitoring data to obtain the slowly varying fitting results, and the coefficients of each basis function in the slowly varying fitting results are extracted as the sensor data features of the slowly varying variables.

[0153] For pulse variables, the cumulative value curve of the pulse variables is first obtained, and then the cumulative value curve is fitted using 100 B-spline basis functions to obtain the curve fitting result. The sensor monitoring data is divided using the nodes of each basis function in the curve fitting result to obtain 100 pulse segments, and the coefficients of each basis function in the curve fitting result and the mean of each pulse segment are extracted as the sensor data features of the pulse variable.

[0154] Based on the above steps, 530 data features can be obtained for the sensor monitoring variables of any culture group, and 6 data features can be obtained for the offline inspection variables of any culture group. The number of features of each process variable can be shown in Table 3. It can be obtained that the dimension of the feature matrix obtained according to the data features of the first culture group is 12×536, and the dimension of the feature matrix obtained according to the data features of the second culture group is 8×536.

[0155] Table 3. Number of characteristics of process variables

[0156]

[0157] In this practical application scenario, the Gradient Boosting with Categorical Features Support (CatBoost) model is used as the culture group classification model to output the importance of data features in the culture group classification and obtain the indicator evaluation data. The indicator evaluation data finally obtained according to the culture group classification model can be referred to Figure 8 As shown in the figure, it can be seen that the index evaluation data of the data feature near the feature number 200 is significantly higher than that of other data features. It can be seen that the corresponding data features are the coefficients of the 78th to 99th basis functions of the pH value. Figure 8 The above part is enlarged to obtain Fig. 9 ,like Fig. 9 As shown, the data features with feature numbers 177-225 correspond to the pH curve in the later stage of the cell culture process. Therefore, it can be determined that in the later stage of the cell culture process, the pH value may be the target variable that causes the difference between the first performance data and the second performance data.

[0158] Reference Fig.10 As shown, Fig.10 The changing trend of pH value in the later stage of cell culture process is shown. It can be seen that in the later stage of cell culture process, the pH value of the second culture group in cluster 2 is significantly higher than the pH value of the first culture group in cluster 1. Therefore, the main reason for the difference between the first performance data and the second performance data is that the pH value in the later stage of cell culture process is too high.

[0159] Accordingly, please refer to Fig.11The present application embodiment provides a cell culture process data automatic mining and analysis system, the system comprising:

[0160] The data feature extraction module 1110 is used to extract features according to the recorded data of multiple process variables of multiple culture groups during the cell culture process to obtain a feature set; wherein the feature set includes the data features of each process variable; the classification results of the multiple culture groups include a first culture group and a second culture group, and the first performance data of the first culture group on the yield evaluation index is better than the second performance data of the second culture group on the yield evaluation index;

[0161] The feature subset determination module 1120 is used to determine the feature to be evaluated and multiple feature subsets in the feature set; wherein any feature subset is any subset of all features in the feature set except the feature to be evaluated;

[0162] Contribution determination module 1130, used to determine the common contribution and individual contribution of any feature subset; wherein the common contribution is the contribution of the feature to be evaluated and any feature subset to the classification of the culture group, and the individual contribution is the contribution of any feature subset to the classification of the culture group alone;

[0163] An index evaluation calculation module 1140 is used to calculate the index importance according to the number of features in the feature set, the number of features in each feature subset, and the common contribution and individual contribution of each feature subset, to obtain index evaluation data of the importance of the feature to be evaluated in the culture group classification;

[0164] The target variable determination module 1150 is used to determine the target feature in the feature set according to the indicator evaluation data, and use the process variable corresponding to the target feature as the target variable that causes the difference between the first performance data and the second performance data.

[0165] In some optional implementations, the indicator evaluation calculation module 1140 includes:

[0166] The classification contribution calculation unit is used to calculate the difference between the common contribution and the individual contribution of each feature subset to obtain the classification contribution of the feature to be evaluated to the classification of the culture group.

[0167] The feature subset quantity calculation unit is used to obtain the number of feature subsets based on the number of features in the feature set and the number of features in each feature subset.

[0168] The classification contribution adjustment unit is used to adjust the weight of the classification contribution of the feature to be evaluated to the classification of the training group according to the number of feature subsets, so as to obtain the adjusted contribution corresponding to each feature subset.

[0169] The index evaluation data calculation unit is used to sum and average the adjusted contribution corresponding to each feature subset according to the number of features in the feature set to obtain the index evaluation data of the importance of the feature to be evaluated in the culture group classification.

[0170] In some optional embodiments, the index evaluation data of the importance of the feature to be evaluated in the culture group classification is obtained in the index evaluation data calculation unit by the following formula:

[0171]

[0172] in, Indicates indicator evaluation data; is the feature to be evaluated; is a feature set; is the number of features in the feature set; is a set of features to be evaluated that are excluded from the feature set; is the union of the feature subset and the feature to be evaluated; is a feature subset; is the number of features in the feature subset; For collection The number of feature subsets; is the common contribution of the feature subset and the feature to be evaluated; is the individual contribution of the feature subset.

[0173] In some optional implementations, the contribution determination module 1130 includes:

[0174] The common contribution output unit is used to input the data feature combination including the feature to be evaluated and the feature subset into the training group classification model to obtain the common contribution output by the training group classification model; wherein the training group classification model is obtained by model training using the feature set.

[0175] The individual contribution output unit is used to input the data features of the feature subset into the training group classification model separately to obtain the individual contribution output by the training group classification model.

[0176] In some optional implementations, the recorded data includes sensor monitoring data corresponding to the sensor monitoring variable and offline inspection data corresponding to the offline inspection variable; the data feature extraction module 1110 includes:

[0177] The variable type determination unit is used to determine the variable type of the sensor monitoring variable according to the change trend of the sensor monitoring data.

[0178] The data feature extraction unit is used to extract features from the sensor monitoring data according to the variable type to obtain sensor data features.

[0179] The offline feature extraction unit is used to use the offline inspection data as offline data features; wherein the feature set includes sensor data features and offline data features.

[0180] In some optional implementations, the variable type determination unit includes:

[0181] The data sequence processing subunit is used to perform cropping and trend sampling on the recorded data corresponding to the sensor monitoring variables to obtain multiple data subsequences with the same time length.

[0182] The classification model training subunit is used to select any number of data subsequences to form multiple training data sets, use the multiple training data sets to perform model training respectively, and obtain multiple classification models of variables to be screened corresponding to the multiple training data sets.

[0183] The classification model screening subunit is used to screen the classification results of the sensor monitoring variables according to the classification models of the variables to be screened, so as to obtain the classification model of the target variable.

[0184] The variable type determination subunit is used to determine the variable type of the sensor monitoring variable according to the classification result of the sensor monitoring variable by the target variable classification model.

[0185] In some optional implementations, the variable types include stable variables, step variables, slowly changing variables, and pulse variables; and the data feature extraction unit includes:

[0186] The stable variable feature extraction subunit is used to extract the mean and standard deviation of the stable variable during the cell culture process as the sensing data features of the stable variable.

[0187] The step variable feature extraction subunit is used to perform piecewise linear regression on the step variable, and obtain the mutation nodes of the step variable in the cell culture process according to the fitting slopes of each data subsequence; divide the sensor monitoring data corresponding to the step variable according to the mutation nodes to obtain step segments; extract the mean and linear fitting slope of each step segment as the sensor data features of the step variable.

[0188] The slowly varying variable feature extraction subunit is used to perform basis function fitting on the slowly varying variable to obtain a slowly varying fitting result; and extract the coefficients of each basis function in the slowly varying fitting result as the sensor data feature of the slowly varying variable.

[0189] The pulse variable feature extraction subunit is used to obtain a cumulative value curve based on the pulse variable, perform basis function fitting on the cumulative value curve, and obtain a curve fitting result; use the nodes of each basis function in the curve fitting result to divide the sensor monitoring data corresponding to the pulse variable to obtain pulse segments; extract the coefficients of each basis function in the curve fitting result and the mean of each pulse segment as the sensor data feature of the pulse variable.

[0190] In some optional embodiments, the system further includes a deviation point detection and rejection module, including:

[0191] The data processing unit is used to perform normalization processing and Fourier transformation on the recorded data to obtain the logarithmic amplitude spectrum of the recorded data.

[0192] The smoothing processing unit is used to smooth the logarithmic amplitude spectrum of the recorded data to obtain a smoothed logarithmic amplitude spectrum.

[0193] The deviation calculation unit is used to perform exponential change and inverse Fourier transformation on the smoothed logarithmic amplitude spectrum and the smoothed loss vector between the logarithmic amplitude spectrum to obtain a deviation index.

[0194] The deviation point determination unit is used to determine the occurrence time of the data deviation point in the process data according to the comparison result of the deviation index and the preset threshold value, so as to eliminate and correct the data deviation point.

[0195] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0196] The cell culture process data automatic mining and analysis system in this embodiment is presented in the form of functional units, where the units refer to ASIC (Application Specific Integrated Circuit) circuits, processors and memories that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0197] See also Fig.12 , Fig.12 It is a structural diagram of a computer device provided by an embodiment of the present application. As shown in the figure, the computer device includes: one or more processors 10, a memory 20, and an interface for connecting various components, including a high-speed interface and a low-speed interface. Each component uses different buses to communicate with each other, and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Fig.12 A processor 10 is taken as an example.

[0198] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0199] The memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0200] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0201] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0202] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0203] The embodiment of the present application also provides a computer-readable storage medium. The above method according to the embodiment of the present application can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0204] The embodiment of the present application provides a computer program product, which includes computer instructions, which are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method of any embodiment of the present application.

[0205] Although the embodiments of the present application are described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present application, and such modifications and variations are all within the scope defined by the appended claims.

[0206] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0207] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0208] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0209] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1A device that provides the functions specified in a block or multiple blocks.

[0210] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0211] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0212] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0213] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0214] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

[0215] Although the embodiments of the present application have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present application, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for automatic mining and analysis of cell culture process data, characterized in that: The method comprises: Extracting features based on recorded data of multiple process variables of multiple culture groups during cell culture to obtain a feature set; wherein the feature set includes data features of each process variable; the classification results of the multiple culture groups include a first culture group and a second culture group, and first performance data of the first culture group on a yield evaluation index is better than second performance data of the second culture group on the yield evaluation index; Determining a feature to be evaluated and a plurality of feature subsets in the feature set; wherein any feature subset is any subset of all features in the feature set except the feature to be evaluated; Determine the common contribution and individual contribution of any feature subset; wherein the common contribution is the contribution of the feature to be evaluated and any feature subset to the classification of the culture group together, and the individual contribution is the contribution of any feature subset to the classification of the culture group alone; Calculating the index importance according to the number of features in the feature set, the number of features in each of the feature subsets, and the common contribution and individual contribution of each of the feature subsets, to obtain index evaluation data of the importance of the feature to be evaluated in the culture group classification; A target feature is determined in the feature set according to the indicator evaluation data, and a process variable corresponding to the target feature is used as a target variable causing a difference between the first performance data and the second performance data.

2. The method according to claim 1, characterized in that: The index importance calculation is performed according to the number of features in the feature set, the number of features in each of the feature subsets, and the common contribution and individual contribution of each of the feature subsets to obtain the index evaluation data of the importance of the feature to be evaluated in the culture group classification, including: The common contribution and the individual contribution of each feature subset are subtracted to obtain the classification contribution of the feature to be evaluated to the classification of the culture group; Obtaining a number of feature subsets based on the number of features in the feature set and the number of features in each of the feature subsets; According to the number of feature subsets, weight adjustment is performed on the classification contribution of the feature to be evaluated to the classification of the culture group to obtain the adjusted contribution corresponding to each feature subset; The adjusted contribution corresponding to each of the feature subsets is summed and averaged according to the number of features in the feature set to obtain index evaluation data of the importance of the feature to be evaluated in the culture group classification.

3. The method according to claim 2, characterized in that The index evaluation data of the importance of the feature to be evaluated in the culture group classification is obtained by the following formula: in, Indicates indicator evaluation data; is the feature to be evaluated; is a feature set; is the number of features in the feature set; is a set of features to be evaluated that are excluded from the feature set; is the union of the feature subset and the feature to be evaluated; is a feature subset; is the number of features in the feature subset; For collection The number of feature subsets; is the common contribution of the feature subset and the feature to be evaluated; is the individual contribution of the feature subset.

4. The method according to claim 1, characterized in that: The determining of the common contribution and the individual contribution of any feature subset comprises: Inputting the data feature combination including the feature to be evaluated and the feature subset into a culture group classification model to obtain the common contribution output by the culture group classification model; wherein the culture group classification model is obtained by model training using the feature set; The data features of the feature subset are individually input into the culture group classification model to obtain the individual contribution output by the culture group classification model.

5. The method according to claim 1, characterized in that The recorded data includes sensor monitoring data corresponding to the sensor monitoring variable and offline inspection data corresponding to the offline inspection variable; the feature set is obtained by extracting features based on the recorded data of multiple process variables of multiple culture groups during the cell culture process, including: Determining the variable type of the sensor monitoring variable according to the change trend of the sensor monitoring data; Extracting features from the sensor monitoring data according to the variable type to obtain sensor data features; The offline inspection data is used as offline data features; wherein the feature set includes the sensor data features and the offline data features.

6. The method according to claim 5, characterized in that Determining the variable type of the sensor monitoring variable according to the change trend of the sensor monitoring data includes: Cutting and trend sampling the recorded data corresponding to the sensor monitoring variable to obtain a plurality of data subsequences with the same time length; Selecting any number of the data subsequences to form multiple training data sets, using the multiple training data sets to perform model training respectively, and obtaining multiple classification models of variables to be screened corresponding to the multiple training data sets; Screening the classification results of the sensor monitoring variables according to the classification models of the variables to be screened to obtain a target variable classification model; The variable type of the sensor monitoring variable is determined according to the classification result of the target variable classification model on the sensor monitoring variable.

7. The method according to claim 6, characterized in that The variable types include stable variables, step variables, slowly changing variables and pulse variables; the feature extraction of the sensor monitoring data according to the variable types to obtain sensor data features includes: extracting the mean and standard deviation of the stable variable during the cell culture process as the sensing data features of the stable variable; Performing piecewise linear regression on the step variable, obtaining the mutation node of the step variable in the cell culture process according to the fitting slope of each of the data subsequences; dividing the sensor monitoring data corresponding to the step variable according to the mutation node to obtain step segments; extracting the mean value and linear fitting slope of each step segment as the sensor data feature of the step variable; Performing basis function fitting on the slowly varying variable to obtain a slowly varying fitting result; extracting coefficients of each basis function in the slowly varying fitting result as sensing data features of the slowly varying variable; A cumulative value curve is obtained according to the pulse variable, and basis function fitting is performed on the cumulative value curve to obtain a curve fitting result; the sensor monitoring data corresponding to the pulse variable is divided using the nodes of each basis function in the curve fitting result to obtain pulse segments; the coefficients of each basis function in the curve fitting result and the mean value of each pulse segment are extracted as the sensor data features of the pulse variable.

8. The method according to any one of claims 1 to 7, characterized in that: The method further comprises: Performing normalization and Fourier transformation on the recorded data to obtain a logarithmic amplitude spectrum of the recorded data; Smoothing the logarithmic amplitude spectrum to obtain a smoothed logarithmic amplitude spectrum; Performing exponential transformation and inverse Fourier transformation on the smoothed logarithmic amplitude spectrum and the smoothed loss vector between the logarithmic amplitude spectrum to obtain a deviation index; According to the comparison result of the deviation index and the preset threshold value, the occurrence time of the data deviation point in the process data is determined to eliminate and correct the data deviation point.

9. A cell culture process data automatic mining and analysis system, characterized in that: The system comprises: A data feature extraction module, used for extracting features based on the recorded data of multiple process variables of multiple culture groups during the cell culture process to obtain a feature set; wherein the feature set includes the data features of each process variable; the classification results of the multiple culture groups include a first culture group and a second culture group, and the first performance data of the first culture group on the yield evaluation index is better than the second performance data of the second culture group on the yield evaluation index; A feature subset determination module, used to determine a feature to be evaluated and a plurality of feature subsets in the feature set; wherein any feature subset is any subset of all features in the feature set except the feature to be evaluated; A contribution determination module, used to determine the common contribution and individual contribution of any feature subset; wherein the common contribution is the contribution of the feature to be evaluated and any feature subset to the classification of the culture group, and the individual contribution is the contribution of any feature subset to the classification of the culture group alone; An index evaluation calculation module is used to calculate the index importance according to the number of features in the feature set, the number of features in each of the feature subsets, and the common contribution and individual contribution of each of the feature subsets, to obtain index evaluation data of the importance of the feature to be evaluated in the culture group classification; A target variable determination module is used to determine a target feature in the feature set according to the indicator evaluation data, and use a process variable corresponding to the target feature as a target variable that causes a difference between the first performance data and the second performance data.

10. A cell culture process data automatic mining and analysis system, comprising: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method according to any one of claims 1 to 8 by executing the computer instructions.

Images