A method for predicting abnormalities during the hypoxia expansion phase of stem cells and related equipment

Abstract

The present application relates to the technical field of stem cell preparation, and provides an abnormality prediction method and related equipment for the hypoxic expansion stage of stem cells. The method includes: calculating the data features of each variable data, and calculating the global features of each historical moment based on all data features; calculating the similarity relationship between every two variable data at the historical moment based on the global features, and constructing a directed graph based on all similarity relationships; calculating the correlation decay rate between every two nodes in the directed graph based on all historical moments, and obtaining the final features of each variable data based on all correlation decay rates; aggregating the final features of all variable data corresponding to the modality at each historical moment to obtain the aggregated features of the modality at each historical moment; updating the aggregated features to obtain the final aggregated features, and performing abnormal prediction based on the final aggregated features to obtain abnormal prediction results. The method of the present application can improve the accuracy of abnormal prediction.

Description

Technical Field

[0001] The present application relates to the technical field of stem cell preparation, and in particular to a method for predicting abnormalities during the hypoxia expansion phase of stem cells and related equipment. Background Art

[0002] The preparation of stem cells is a highly sophisticated job, which is reflected in key steps such as raw material collection, passage separation, and process parameter control. In the current process flow, conditions that affect cell quality and expansion efficiency can be regarded as abnormal conditions, and changes in observable variables of environmental parameters and biological properties can well reflect abnormalities. The hypoxia expansion stage involves continuous and sophisticated multi-time series and multi-modal variables. Changes in a single variable can be monitored in a timely manner through various methods, but they cannot be used as a signal of abnormality. The reason is that the abnormality of a single variable does not directly affect cell growth, but rather causes a chain reaction by affecting other related variables, resulting in abnormal conditions.

[0003] There are differences between individual stem cells. These differences are due to genetic diversity, epigenetic differences, cell metabolic state, etc., which means that different stem cells have different requirements for the growth environment and the differences are difficult to determine. In addition, stem cells are very sensitive to the perception of the culture environment. Even small fluctuations may trigger the stress response of cells, which makes professionals face a series of serious problems in cell culture. First, the adjustment of process parameters according to the predetermined plan cannot fully guarantee the efficient expansion of cells, and may even cause abnormal cell quality. Second, the control of related parameters of microcarriers will lead to abnormalities. With the increase in the number of passages, multiple scale bioreactors will be used. In different bioreactors, different stirring speeds or fluid dynamic conditions may cause microcarrier aggregation or sedimentation, which will cause micro-changes in the local environment and also affect its associated variables. Third, embodied intelligence will be applied in the future scenarios of stem cell process flow. Its long-term non-fixed program operation may have perception errors, resulting in a certain deviation between the actual operation time and the planned operation time. At the same time, robots may also over-adapt to specific conditions. When the process requirements change, their performance may be poor or even abnormal. These problems are analyzed by capturing the characteristics of time series data such as environmental and biological properties. The complexity and dimensionality of multi-time series and multi-modal variables increase with the intelligence of the process flow. The cost of automatically monitoring these data is getting higher and higher, and the system latency will also change. In addition, given the large scale of cells in the expansion stage and the high cost of subculture, although the occurrence of abnormalities is a low-probability event, it is still indispensable to predict abnormalities in multivariate time series data in this scenario.

[0004] As far as we know, there is no anomaly prediction for the hypoxic expansion stage of stem cells. In traditional time series data anomaly detection algorithms, most research works model by capturing the spatial and temporal dependencies between factory data streams. On this basis, graph neural network-based methods have been widely adopted. Some studies model dependencies based on distance or similarity methods. A significant disadvantage is the symmetry of the dependency relationship. Failure to comprehensively consider direct or indirect unidirectional dependencies, dynamic dependencies between variables, etc. will lead to a lack of stability. In the environment of the stem cell preparation process, there are a variety of observed variables in categories such as microbial safety, biological safety, and cell biology that can intuitively reflect the growth status of stem cells. There are highly complex dependencies between these variables. For example, a certain concentration of carbon dioxide It will react with water to produce carbonic acid, which will affect Value; dissolved oxygen The level of affects the metabolic activity of cells, which produces and water, the result will be the opposite effect Level and Judging abnormalities only by considering the changes in each single variable will obviously result in errors, which leads to low accuracy in abnormal prediction of the environment during the hypoxic expansion stage of stem cells. Summary of the invention

[0005] The present application provides a method and related equipment for predicting abnormalities in the hypoxia expansion stage of stem cells, which can solve the problem of low accuracy of abnormality prediction.

[0006] In a first aspect, an embodiment of the present application provides a method for predicting abnormalities in a hypoxic expansion phase of stem cells, the method comprising:

[0007] Acquire multiple variable data of the environment in which the target stem cells are located at multiple historical moments during the hypoxia expansion stage; the multiple variable data correspond to multiple modes, and each mode corresponds to at least one variable data;

[0008] Calculate the data features of each variable data at each historical moment, and calculate the global features of each historical moment based on all data features;

[0009] Based on the global features of each historical moment, the similarity relationship between each two variable data at each historical moment is calculated, and a directed graph is constructed based on all similarity relationships; multiple nodes of the directed graph correspond to multiple variable data one by one, and the edge between two nodes is the similarity relationship between the corresponding two variable data;

[0010] Calculate the correlation decay rate between every two nodes in the directed graph according to all historical moments, and obtain the final features of each variable data based on all correlation decay rates;

[0011] For each mode, aggregate the final features of all variable data corresponding to the mode at each historical moment to obtain the aggregated features of the mode at each historical moment;

[0012] All aggregate features are updated to obtain multiple final aggregate features, and anomaly prediction is performed based on all final aggregate features to obtain anomaly prediction results; the anomaly prediction results are used to describe the abnormal conditions of multiple variable data of the environment in a future time period.

[0013] Optionally, calculate the data features of each variable data at each historical moment, including:

[0014] By formula:

[0015]

[0016] Calculate the The historical moment Data characteristics of variable data ;

[0017] in, Represents the sensor characteristics, , , represents convolution, Indicates The variable data matrix of historical moments, Indicates The first variable data of a historical moment, Indicates The second variable data of the historical moment, Indicates The historical moment Variable data, , Indicates the number of variable data at each historical moment, , represents the number of historical moments, Indicates The historical moment Positional encoding of variable data:

[0018]

[0019] in, represents an integer index, Indicates the oscillation frequency.

[0020] Optionally, global features are calculated for each historical moment based on all data features, including:

[0021] By formula:

[0022]

[0023]

[0024] Calculate the The global characteristics of a historical moment ;

[0025] in, Indicates The historical moment The information centrality of the variable data is Indicates The historical moment The data characteristics of the variable data are Indicates The historical moment The variable data and The historical moment The degree of information sharing between variable data express The entropy of express The entropy of express and The entropy between them; the information centrality is the sum of the information sharing degrees between the variable data and each other variable data at the historical moment.

[0026] Optionally, the similarity relationship between each two variable data at each historical moment is calculated based on the global features at each historical moment, including:

[0027] For each historical moment, perform the following steps:

[0028] Calculate the comprehensive similarity between every two variable data at historical moments based on the global characteristics of historical moments;

[0029] The initial similarity relationship is obtained according to the comprehensive similarity between every two variable data, and the initial similarity relationship is updated to obtain the similarity relationship between every two variable data.

[0030] Optionally, the comprehensive similarity between each two variable data at the historical moments is calculated based on the global features of the historical moments, including:

[0031] By formula:

[0032]

[0033]

[0034] Calculate the The historical moment The variable data and The historical moment The comprehensive similarity between variable data ;

[0035] in, , and Both represent saturation rate, Indicates The variable data and The degree of information sharing between variable data Indicates The variable data and The degree of information sharing between the global features of historical moments, Indicates The variable data and The cosine similarity between variable data, Indicates The historical moment The data characteristics of the variable data are Indicates The historical moment The data characteristics of the variable data are , , Indicates the number of variable data at each historical moment, , represents the number of historical moments;

[0036] The initial similarity relationship is obtained based on the comprehensive similarity between each two variable data, including:

[0037] By formula:

[0038]

[0039]

[0040] Get the The variable data and The initial similarity relationship between the variable data ;

[0041] in, represents the exponential function value, Indicates The variable data and The normalized similarity between variable data, represents the similarity threshold;

[0042] The initial similarity relationship is updated to obtain the similarity relationship between every two variable data, including:

[0043] By formula:

[0044]

[0045] Initial similarity relationship Update to get The variable data and Similarity between variable data ;

[0046] in, Represents the prior knowledge in the directed graph The variable data and The adjacency relationship between variable data, Represents the activation function.

[0047] Optionally, calculate the correlation decay rate between every two nodes in the directed graph based on all historical moments, including:

[0048] For each historical moment, the correlation decay rate between the node at that historical moment and the node at every other historical moment is defined as ; The value of is the difference between the historical moment and other historical moments minus one. is a constant;

[0049] The final features of each variable data are obtained based on all correlation decay rates, including:

[0050] By formula:

[0051]

[0052]

[0053] Calculate the The historical moment The final characteristics of the variable data ;

[0054] in, represents the learnable parameters, Indicates The historical moment The variable data and The historical moment The comprehensive adjacency relationship between variable data, Indicates The historical moment The data characteristics of the variable data are represents the number of historical moments, Indicates the number of variable data at each historical moment, Indicates The historical moment The variable data and The historical moment The adjacency relationship between variable data is ,but , Indicates The historical moment The variable data and The historical moment The comprehensive similarity between variable data is ,but , Indicates The node of the historical moment and the The correlation decay rate between nodes at a historical moment; the comprehensive adjacency relationship is the adjacency relationship between two variable data after the introduction of the correlation decay rate for adjustment.

[0055] Optionally, the final features of all variable data corresponding to the modality at each historical moment are aggregated to obtain the aggregate features of the modality at each historical moment, including:

[0056] By formula:

[0057]

[0058] Calculate the The mode is in Aggregate features of historical moments ;

[0059] in, Indicates The mode is in The set of all variable data corresponding to a historical moment, Indicates The historical moment The information centrality of the variable data.

[0060] Optionally, all aggregate features are updated to obtain multiple final aggregate features, including:

[0061] Calculate the correlation between each two aggregated features;

[0062] All aggregate features are updated according to all correlation degrees to obtain multiple final aggregate features.

[0063] Optionally, calculate the correlation between each two aggregated features, including:

[0064] By formula:

[0065]

[0066] Calculate the The mode and Between the modes The degree of relevance of historical moments ;

[0067] in, Indicates The mode is in The set of all variable data corresponding to a historical moment, Indicates The historical moment The variable data and The historical moment The relationship between the variable data, , Indicates The historical moment The variable data and The historical moment The comprehensive similarity between variable data;

[0068] All aggregate features are updated according to all association degrees to obtain multiple final aggregate features, including:

[0069] By formula:

[0070]

[0071] Update all aggregate features;

[0072] in, Indicates The intermediate aggregate feature matrix of the layer, represents the correlation matrix, Indicates The learnable weight matrix of the layer, Indicates The intermediate aggregate feature matrix of the layer, represents the activation function, , represents the update of the last layer, when hour, represents the aggregate feature matrix, when hour, Represents the final aggregated feature matrix.

[0073] In a second aspect, the present application provides an abnormality prediction device for a hypoxic expansion stage of stem cells, comprising:

[0074] An acquisition module is used to acquire multiple variable data of the environment in which the target stem cells are located at multiple historical moments during the hypoxia expansion stage; the multiple variable data correspond to multiple modes, and each mode corresponds to at least one variable data;

[0075] A first calculation module calculates the data features of each variable data at each historical moment, and calculates the global features of each historical moment based on all the data features;

[0076] The construction module calculates the similarity relationship between each two variable data at each historical moment based on the global features of each historical moment, and constructs a directed graph based on all similarity relationships; multiple nodes of the directed graph correspond to multiple variable data one by one, and the edge between two nodes is the similarity relationship between the corresponding two variable data;

[0077] The second calculation module calculates the correlation decay rate between every two nodes in the directed graph according to all historical moments, and obtains the final features of each variable data based on all correlation decay rates;

[0078] The aggregation module aggregates the final features of all variable data corresponding to each mode at each historical moment to obtain the aggregated features of the mode at each historical moment.

[0079] The anomaly prediction module updates all aggregated features to obtain multiple final aggregated features, and performs anomaly prediction based on all final aggregated features to obtain anomaly prediction results; the anomaly prediction results are used to describe the abnormal conditions of multiple variable data of the environment in the future time period.

[0080] In a third aspect, an embodiment of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for predicting abnormalities in the hypoxia expansion stage of stem cells when executing the above-mentioned computer program.

[0081] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned method for predicting abnormalities in the hypoxia expansion stage of stem cells.

[0082] The above solution of the present application has the following beneficial effects:

[0083] In an embodiment of the present application, the data features of each variable data at each historical moment are calculated, and the global features of each historical moment are calculated based on all the data features, and then the similarity relationship between every two variable data at each historical moment is calculated based on the global features of each historical moment, and a directed graph is constructed based on all similarity relationships, and then the correlation decay rate between every two nodes in the directed graph is calculated based on all historical moments, and the final features of each variable data are obtained based on all correlation decay rates, and then for each modality, the final features of all variable data corresponding to the modality at each historical moment are aggregated to obtain the aggregated features of the modality at each historical moment, and finally all the aggregated features are updated to obtain multiple final aggregated features, and anomaly prediction is performed based on all the final aggregated features to obtain anomaly prediction results. Among them, calculating the global features based on the data features of multiple variable data can comprehensively analyze multiple variables of the target stem cell environment and improve the information richness of the global features. Obtaining the similarity relationship between the variable data based on the global features can improve the accuracy and globality of the similarity relationship. The final features of the variable data are calculated using the correlation decay rate, taking into account the correlation between historical moments, improving the accuracy of the final features, and then improving the accuracy of the final aggregated features of the modality obtained by the aggregation of the final features. Anomaly prediction based on accurate final aggregated features can effectively improve the accuracy of anomaly prediction.

[0084] Other beneficial effects of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0086] Figure 1 A flow chart of a method for predicting abnormalities in the hypoxia expansion phase of stem cells provided in one embodiment of the present application;

[0087] Figure 2 A schematic diagram of the structure of an abnormality prediction device for stem cell hypoxia expansion stage provided in one embodiment of the present application;

[0088] Figure 3 A schematic diagram of the structure of a terminal device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0089] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0090] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0091] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0092] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0093] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0094] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0095] In response to the problem of low accuracy of existing abnormality predictions, an embodiment of the present application provides an abnormality prediction method for the hypoxic expansion stage of stem cells. The abnormality prediction method calculates the data features of each variable data at each historical moment, and calculates the global features of each historical moment based on all data features, and then calculates the similarity relationship between every two variable data at each historical moment based on the global features of each historical moment, and constructs a directed graph based on all similarity relationships, and then calculates the correlation decay rate between every two nodes in the directed graph based on all historical moments, and obtains the final features of each variable data based on all correlation decay rates, and then, for each modality, respectively, aggregates the final features of all variable data corresponding to the modality at each historical moment to obtain the aggregated features of the modality at each historical moment, and finally updates all aggregated features to obtain multiple final aggregated features, and performs abnormality prediction based on all final aggregated features to obtain abnormality prediction results. Among them, calculating the global features based on the data features of multiple variable data can comprehensively analyze multiple variables of the target stem cell environment and improve the information richness of the global features. Obtaining the similarity relationship between the variable data based on the global features can improve the accuracy and globality of the similarity relationship. The final features of the variable data are calculated using the correlation decay rate, taking into account the correlation between historical moments, improving the accuracy of the final features, and then improving the accuracy of the final aggregated features of the modality obtained by the aggregation of the final features. Anomaly prediction based on accurate final aggregated features can effectively improve the accuracy of anomaly prediction.

[0096] Next, the abnormality prediction method of the hypoxic expansion stage of stem cells provided in the present application is exemplified.

[0097] like Figure 1 As shown, the abnormal prediction method of stem cell hypoxia expansion stage provided in the present application comprises the following steps:

[0098] Step 11, obtaining multiple variable data of the environment in which the target stem cells are located at multiple historical moments during the hypoxia expansion stage.

[0099] The above-mentioned multiple variable data correspond to multiple modes, and each mode corresponds to at least one variable data. The above-mentioned target stem cells are stem cells that are growing and in a hypoxic expansion stage. The multiple variable data can be data that affect the growth of stem cells, such as the temperature value, carbon dioxide concentration value, oxygen concentration value of the environment, etc. The variable data is divided into multiple modes, such as mode: temperature (temperature value), gas (carbon dioxide concentration value, oxygen concentration value), etc. The number of modes is less than or equal to the number of variable data at each historical moment. The above-mentioned environment is the environment for stem cell preparation, such as the internal environment of a cell culture incubator.

[0100] In some embodiments of the present application, multiple variable data may be acquired using devices such as sensors.

[0101] Step 12, calculate the data features of each variable data at each historical moment, and calculate the global features of each historical moment based on all the data features.

[0102] Specifically, through the formula:

[0103]

[0104] Calculate the The historical moment Data characteristics of variable data .

[0105] in, Represents the sensor characteristics, , , represents convolution, Indicates The variable data matrix of historical moments, Indicates The first variable data of a historical moment, Indicates The second variable data of the historical moment, Indicates The historical moment Variable data, , Indicates the number of variable data at each historical moment, , represents the number of historical moments, Indicates The historical moment Positional encoding of variable data:

[0106]

[0107] in, represents the integer index that determines the sine and cosine functions of different frequencies, Represents the oscillation frequency of the sine and cosine functions in the positional encoding.

[0108] By formula:

[0109]

[0110]

[0111] Calculate the The global characteristics of a historical moment .

[0112] in, Indicates The historical moment The information centrality of the variable data is Indicates The historical moment The data characteristics of the variable data are Indicates The historical moment The variable data and The historical moment The degree of information sharing between variable data express The entropy of express The entropy of express and The entropy between them; the information centrality is the sum of the information sharing degrees between the variable data and each other variable data at the historical moment.

[0113] It should be noted that in general industrial scenarios, the variables monitored by multiplexers have symmetrical interdependencies in time and space. Existing studies on anomaly detection only consider the characteristics of adjacent variables, while different types of sensors collect variable data of different modes, which have non-uniformity, complex dependencies and different sampling frequencies. Although professionals have certain prior knowledge in biology or chemistry, most of this knowledge is limited to the variables represented by adjacent variables. There may be hidden interactions between other micro-variables, and there may be unobserved variables that affect the relationship between two observed variables.

[0114] It is worth mentioning that the present application takes into account multiple variable data and calculates global features based on the data features of multiple variable data. It can conduct a comprehensive analysis of multiple variables in the environment of the target stem cells, rather than just considering the relationship between adjacent variables in the variable data, thereby effectively improving the information richness of the global features.

[0115] Step 13, based on the global features of each historical moment, calculate the similarity relationship between every two variable data at each historical moment, and construct a directed graph based on all similarity relationships.

[0116] The multiple nodes of the directed graph correspond one-to-one to the multiple variable data, and the edge between two nodes is the similarity relationship between the corresponding two variable data.

[0117] In some embodiments of the present application, the above steps of calculating the similarity relationship between every two variable data at each historical moment based on the global features of each historical moment, and constructing a directed graph according to all similarity relationships are specifically as follows:

[0118] The first step is to perform the following steps for each historical moment:

[0119] First, the comprehensive similarity between every two variable data at historical moments is calculated based on the global characteristics of the historical moments.

[0120] Specifically, through the formula:

[0121]

[0122]

[0123] Calculate the The historical moment The variable data and The historical moment The comprehensive similarity between variable data ;

[0124] in, , and Both represent saturation rate, Indicates The variable data and The degree of information sharing between variable data Indicates The variable data and The degree of information sharing between the global features of historical moments, Indicates The variable data and The cosine similarity between variable data, Indicates The historical moment The data characteristics of the variable data are Indicates The historical moment The data characteristics of the variable data are , , Indicates the number of variable data at each historical moment, , represents the number of historical moments;

[0125] Then, the initial similarity relationship is obtained according to the comprehensive similarity between every two variable data, and the initial similarity relationship is updated to obtain the similarity relationship between every two variable data.

[0126] Specifically, through the formula:

[0127]

[0128]

[0129] Get the The variable data and The initial similarity relationship between the variable data .

[0130] in, represents the exponential function value, Indicates The variable data and The normalized similarity between variable data, Represents the similarity threshold.

[0131] By formula:

[0132]

[0133] Initial similarity relationship Update to get The variable data and Similarity between variable data .

[0134] in, Represents the prior knowledge in the directed graph The variable data and The adjacency relationship between variable data, Represents the activation function.

[0135] It should be noted that the multiple nodes in the above-mentioned prior knowledge directed graph correspond to multiple variable data one by one, and the edges between the nodes are the prior knowledge relationships between the corresponding two variable data. The prior knowledge relationship is used to describe the association between the variable data, which can be obtained by manually sorting out the experience of professionals and combining professional knowledge analysis in the fields of biology and chemistry. If the value of the prior knowledge relationship is 1, it is considered that there is an association relationship between the two variable data. If the value of the prior knowledge relationship is 0, it is considered that there is no association relationship between the two variable data.

[0136] In the second step, a directed graph is constructed based on all similarity relations.

[0137] Specifically, multiple nodes are generated, the multiple nodes correspond to multiple variable data one by one, and the edges between the nodes are corresponding similarity relationships.

[0138] Exemplarily, if the value of the similarity relationship is 0, it means that there is no similarity relationship between the corresponding two variable data, that is, there is no edge between the corresponding two nodes.

[0139] It is worth mentioning that obtaining the similarity relationship between variable data based on global features can improve the accuracy and globality of the similarity relationship. By constructing a directed graph based on the similarity relationship, the similarity relationship between variable data can be expressed, which facilitates the subsequent steps to analyze and calculate the similarity relationship between variable data.

[0140] Step 14, calculate the correlation decay rate between every two nodes in the directed graph according to all historical moments, and obtain the final features of each variable data based on all correlation decay rates.

[0141] The above correlation decay rate is used to describe the degree of influence between two historical moments corresponding to every two nodes.

[0142] In some embodiments of the present application, the above calculation of the correlation decay rate between every two nodes in the directed graph according to all historical moments and obtaining the final features of each variable data based on all correlation decay rates are specifically:

[0143] In the first step, the correlation decay rate between every two nodes in the directed graph is calculated based on all historical moments.

[0144] Specifically, for each historical moment, the correlation decay rate between the node at that historical moment and the node at every other historical moment is defined as ; The value of is the difference between the historical moment and other historical moments minus one. is a constant.

[0145] In the second step, the final features of each variable data are obtained based on all correlation decay rates.

[0146] Specifically, through the formula:

[0147]

[0148]

[0149] Calculate the The historical moment The final characteristics of the variable data ;

[0150] in, represents the learnable parameters, Indicates The historical moment The variable data and The historical moment The comprehensive adjacency relationship between variable data, Indicates The historical moment The data characteristics of the variable data are represents the number of historical moments, Indicates the number of variable data at each historical moment, Indicates The historical moment The variable data and The historical moment The adjacency relationship between variable data is ,but , Indicates The historical moment The variable data and The historical moment The comprehensive similarity between variable data is ,but , Indicates The node of the historical moment and the The correlation decay rate between nodes at a historical moment; the comprehensive adjacency relationship is the adjacency relationship between two variable data after the introduction of the correlation decay rate for adjustment.

[0151] It should be noted that in the scenario of stem cell intelligent preparation, there will be unidirectional dependencies between complex variables that span both time and space. historical moment Will be and historical moment The values ​​have different impacts.

[0152] It is worth mentioning that by calculating the correlation decay rate and obtaining the final features based on the correlation decay rate, the mutual influence between the variable data in time series and the degree of influence are taken into account, so that the information reflected by the final features is consistent with the actual situation.

[0153] Step 15, for each mode, respectively, aggregate the final features of all variable data corresponding to the mode at each historical moment to obtain the aggregated features of the mode at each historical moment.

[0154] Specifically, through the formula:

[0155]

[0156] Calculate the The mode is in Aggregate features of historical moments .

[0157] in, Indicates The mode is in The set of all variable data corresponding to the historical moment , indicating the The historical moment The information centrality of the variable data.

[0158] Step 16: Update all aggregated features to obtain multiple final aggregated features, and perform anomaly prediction based on all final aggregated features to obtain anomaly prediction results.

[0159] The abnormal prediction results are used to describe the abnormal conditions of multiple variable data of the environment in the future time period (such as the abnormal probability of variable data being abnormal among all variable data).

[0160] In some embodiments of the present application, the above steps of updating all aggregate features to obtain multiple final aggregate features, and performing anomaly prediction based on all final aggregate features to obtain anomaly prediction results are specifically as follows:

[0161] In the first step, the correlation between each two aggregated features is calculated.

[0162] Specifically, through the formula:

[0163]

[0164] Calculate the The mode and Between the modes The degree of relevance of historical moments .

[0165] in, Indicates The mode is in The set of all variable data corresponding to a historical moment, Indicates The historical moment The variable data and The historical moment The relationship between the variable data, , Indicates The historical moment The variable data and The historical moment The comprehensive similarity between the variable data.

[0166] In the second step, all aggregate features are updated according to all correlation degrees to obtain multiple final aggregate features.

[0167] Specifically, through the formula:

[0168]

[0169] Updates are made to all aggregate features.

[0170] in, Indicates The intermediate aggregate feature matrix of the layer, Represents the correlation matrix (i.e., the matrix composed of all correlations. Each layer uses the same correlation matrix for calculation. When the variable data at a historical moment changes, it is updated using the above formula for calculating the correlation). Indicates The learnable weight matrix of the layer, Indicates The intermediate aggregate feature matrix of the layer, represents the activation function, , represents the update of the last layer, when hour, represents the aggregate feature matrix (i.e., the matrix composed of all aggregate features obtained in the previous step), when hour, Represents the final aggregated feature matrix.

[0171] It should be noted that the learnable weight matrix can be optimized using optimization algorithms such as the gradient descent method, thereby improving the computational performance of the above formula for updating all aggregate features.

[0172] The third step is to perform anomaly prediction based on all the final aggregated features to obtain the anomaly prediction results.

[0173] Specifically, the matrix composed of all the final aggregated features can be input into the prediction model (such as the softmax activation function layer, etc.). The prediction model calculates the anomaly probability of all variable data occurring abnormally in the future time window consistent with the time window size composed of all historical moments based on all the final aggregated features, and obtains the anomaly prediction result.

[0174] It should be noted that after obtaining the abnormal probability, an early warning can be issued based on the value of the abnormal probability, and corresponding measures can be taken in time to maintain the values ​​of multiple variables in the environment where the target stem cells are located during the hypoxic expansion stage. If the value of the abnormal probability is greater than the abnormal probability threshold, an early warning is issued. If it is less than the abnormal probability threshold, no early warning is required. If an early warning occurs, the temperature and gas concentration of the environment need to be managed and adjusted to maintain the environmental state at the state required by the stem cells during the hypoxic expansion stage. The future moment is the moment after the current moment. If the current moment is 9 o'clock, the future moment can be 10 o'clock, and multiple historical moments can be 7 o'clock, 7:30, and 8 o'clock.

[0175] In order to ensure the accuracy of the prediction model for advance prediction, before performing the above steps, the prediction model needs to be trained first, the data anomalies of each mode at multiple moments after each historical moment are obtained, and the prediction model is trained based on all the data anomalies. Specifically, the prediction model is used to historical moments to make predictions, and obtain The future of a historical moment The probability of abnormal window at a moment , dynamic window ,Will As a fixed Historical moment prediction window ,in Obviously, the greater the abnormal probability, the closer it is to the occurrence of an abnormal situation. Therefore, a dynamic window mechanism is adopted. The dynamic size is ,but and The relationship is as follows:

[0176]

[0177] in, It is an adjustable parameter, which is used to control the window size based on experience. The true label value at the moment , In the At a historical moment, if the collection If there are abnormal labels in is an accurate prediction. If there is no abnormal label in The calculation is not accurate.

[0178] Abnormal Probability Too high but actually No anomalies occur in the set, or the probability of anomalies Low but Anomalies occurred in the set, and both of these situations should result in higher losses. Therefore, a loss function is constructed to train the prediction model. The loss function is:

[0179]

[0180]

[0181]

[0182] in, represents the value of the loss function, Indicates The weight of a historical moment, Indicates The deviation loss of historical moments, Indicates the actual abnormal condition. , , Indicates the deviation value.

[0183] Then the prediction model is trained according to the loss function. If the value of the loss function is less than the preset value of the loss function, the prediction model training is considered complete, and the above-mentioned steps of abnormal prediction based on all the final aggregated features to obtain the abnormal prediction results can be performed. If the value of the loss function is greater than or equal to the preset value of the loss function, the parameters of the prediction model are adjusted and training continues.

[0184] It is worth mentioning that calculating global features based on data features of multiple variable data can comprehensively analyze multiple variables of the target stem cell environment and improve the information richness of global features. Obtaining similar relationships between variable data based on global features can improve the accuracy and globality of similar relationships. Calculating the final features of variable data using the correlation decay rate takes into account the correlation between historical moments, thereby improving the accuracy of the final features and thus improving the accuracy of the final aggregated features of the modality obtained by aggregating the final features. Anomaly prediction based on accurate final aggregated features can effectively improve the accuracy of anomaly prediction.

[0185] The abnormality prediction device for the hypoxia expansion stage of stem cells provided in the present application is exemplified below.

[0186] like Figure 2 As shown, the embodiment of the present application provides an abnormality prediction device for the hypoxia expansion stage of stem cells. The abnormality prediction device 200 for the hypoxia expansion stage of stem cells includes:

[0187] An acquisition module 201 acquires multiple variable data of the environment in which the target stem cells are located at multiple historical moments during the hypoxia expansion stage; the multiple variable data correspond to multiple modes, and each mode corresponds to at least one variable data;

[0188] A first calculation module 202 calculates the data features of each variable data at each historical moment, and calculates the global features of each historical moment based on all the data features;

[0189] A construction module 203 calculates the similarity relationship between each two variable data at each historical moment based on the global features of each historical moment, and constructs a directed graph according to all similarity relationships; multiple nodes of the directed graph correspond to multiple variable data one by one, and an edge between two nodes is a similarity relationship between the corresponding two variable data;

[0190] The second calculation module 204 calculates the correlation decay rate between every two nodes in the directed graph according to all historical moments, and obtains the final feature of each variable data based on all correlation decay rates;

[0191] Aggregation module 205, for each mode, aggregates the final features of all variable data corresponding to the mode at each historical moment to obtain the aggregated features of the mode at each historical moment;

[0192] The abnormal prediction module 206 updates all the aggregated features to obtain multiple final aggregated features, and performs abnormal prediction based on all the final aggregated features to obtain abnormal prediction results; the abnormal prediction results are used to describe the abnormal conditions of multiple variable data of the environment in the future time period.

[0193] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0194] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0195] like Figure 3 As shown, an embodiment of the present application provides a terminal device. The terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 3 Only one processor is shown in the figure), a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 implements the steps of any of the above-mentioned method embodiments when executing the computer program D102.

[0196] Specifically, when the processor D100 executes the computer program D102, it calculates the data features of each variable data at each historical moment, and calculates the global features of each historical moment based on all the data features, and then calculates the similarity relationship between every two variable data at each historical moment based on the global features of each historical moment, and constructs a directed graph based on all similarity relationships, and then calculates the correlation decay rate between every two nodes in the directed graph based on all historical moments, and obtains the final features of each variable data based on all correlation decay rates, and then respectively for each mode, aggregates the final features of all variable data corresponding to the mode at each historical moment to obtain the aggregated features of the mode at each historical moment, and finally updates all the aggregated features to obtain multiple final aggregated features, and performs anomaly prediction based on all the final aggregated features to obtain anomaly prediction results. Among them, calculating the global features based on the data features of multiple variable data can comprehensively analyze multiple variables of the target stem cell environment and improve the information richness of the global features. Obtaining the similarity relationship between the variable data based on the global features can improve the accuracy and globality of the similarity relationship. The final features of the variable data are calculated using the correlation decay rate, taking into account the correlation between historical moments, improving the accuracy of the final features, and then improving the accuracy of the final aggregated features of the modality obtained by the aggregation of the final features. Anomaly prediction based on accurate final aggregated features can effectively improve the accuracy of anomaly prediction.

[0197] The processor D100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0198] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may also be an external storage device of the terminal device D10, such as a plug-in hard disk, a smart memory card (SMC, SmartMedia Card), a secure digital (SD, Secure Digital) card, a flash card (Flash Card), etc. equipped on the terminal device D10. Further, the memory D101 may also include both an internal storage unit of the terminal device D10 and an external storage device. The memory D101 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program, etc. The memory D101 may also be used to temporarily store data that has been output or is to be output.

[0199] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.

[0200] An embodiment of the present application provides a computer program product. When the computer program product runs on a terminal device, the terminal device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0201] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium that can carry the computer program code to the abnormal prediction method device / terminal device of the hypoxia expansion stage of stem cells. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk.

[0202] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0203] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0204] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for predicting abnormalities in the hypoxic expansion phase of stem cells, characterized in that: include: Acquire multiple variable data of the environment in which the target stem cells are located at multiple historical moments during the hypoxic expansion stage; Multiple variable data correspond to multiple modes, and each mode corresponds to at least one variable data; Calculate the data features of each variable data at each of the historical moments, and calculate the global features of each of the historical moments based on all the data features; Based on the global features of each of the historical moments, a similarity relationship between each two variable data at each of the historical moments is calculated, and a directed graph is constructed according to all the similarity relationships; a plurality of nodes of the directed graph correspond one-to-one to a plurality of variable data, and an edge between two nodes is a similarity relationship between the corresponding two variable data; Calculate the correlation decay rate between every two nodes in the directed graph according to all historical moments, and obtain the final features of each variable data based on all correlation decay rates; The correlation decay rate is used to describe the degree of influence between two historical moments corresponding to each two nodes; For each mode, aggregate the final features of all variable data corresponding to the mode at each historical moment to obtain the aggregated features of the mode at each historical moment; All aggregated features are updated to obtain multiple final aggregated features, and anomaly prediction is performed based on all final aggregated features to obtain anomaly prediction results; the anomaly prediction results are used to describe the abnormal conditions of multiple variable data of the environment in a future time period; The calculating of the data features of each variable data at each historical moment includes: By formula: Calculate the The historical moment Data characteristics of variable data ; in, Represents the sensor characteristics, , , represents convolution, Indicates the The variable data matrix of historical moments, Indicates the The first variable data of a historical moment, Indicates the The second variable data of the historical moment, Indicates the The historical moment Variable data, , Indicates the number of variable data at each historical moment, , represents the number of historical moments, Indicates the The historical moment Positional encoding of variable data: in, represents an integer index, represents the oscillation frequency; The calculating of the global features of each historical moment based on all data features includes: By formula: Calculate the The global characteristics of a historical moment ; in, Indicates the The historical moment The information centrality of the variable data is Indicates the The historical moment The data characteristics of the variable data are Indicates the The historical moment The variable data is related to the The historical moment The degree of information sharing between variable data express The entropy of express The entropy of express and The entropy between them; the information centrality is the sum of the information sharing degree between the variable data and each other variable data at the historical moment; The calculating the correlation decay rate between every two nodes in the directed graph according to all historical moments includes: For each of the historical moments, the correlation decay rate between the node at the historical moment and the node at each other historical moment is defined as ; The value of is the difference between the historical moment and other historical moments minus one, is a constant; The final features of each variable data are obtained based on all correlation decay rates, including: By formula: Calculate the The historical moment The final characteristics of the variable data ; in, represents the learnable parameters, Indicates The historical moment The variable data and The historical moment The comprehensive adjacency relationship between variable data, Indicates the The historical moment The data characteristics of the variable data are represents the number of historical moments, Indicates the number of variable data at each historical moment, Indicates The historical moment The variable data and The historical moment The adjacency relationship between variable data is ,but , Indicates the The historical moment The variable data is related to the The historical moment The comprehensive similarity between variable data is ,but , Indicates The node of the historical moment and the The correlation decay rate between nodes at each historical moment; the comprehensive adjacency relationship is the adjacency relationship between two variable data after the introduction of the correlation decay rate for adjustment; The step of aggregating the final features of all variable data corresponding to the modality at each historical moment to obtain the aggregated features of the modality at each historical moment includes: By formula: Calculate the The mode is in Aggregate features of historical moments ; in, Indicates the The mode is in The set of all variable data corresponding to a historical moment, Indicates the The historical moment The information centrality of the variable data.

2. The abnormality prediction method according to claim 1, characterized in that: The calculating the similarity relationship between each two variable data at each historical moment based on the global features of each historical moment comprises: For each of the historical moments, perform the following steps: Calculate the comprehensive similarity between every two variable data at the historical moment based on the global features of the historical moment; An initial similarity relationship is obtained according to the comprehensive similarity between every two variable data, and the initial similarity relationship is updated to obtain a similarity relationship between every two variable data.

3. The abnormality prediction method according to claim 2, characterized in that: The calculating the comprehensive similarity between every two variable data at the historical moments based on the global features of the historical moments comprises: By formula: Calculate the The historical moment The variable data is related to the The historical moment The comprehensive similarity between variable data ; in, , and Both represent saturation rate, Indicates the The variable data is related to the The degree of information sharing between variable data Indicates the The variable data is related to the The degree of information sharing between the global features of historical moments, Indicates the The variable data is related to the The cosine similarity between variable data, Indicates the The historical moment The data characteristics of the variable data are Indicates the The historical moment The data characteristics of the variable data are , , Indicates the number of variable data at each historical moment, , Represents the number of historical moments; The step of obtaining the initial similarity relationship according to the comprehensive similarity between each two variable data includes: By formula: Get the The variable data is related to the The initial similarity relationship between the variable data ; in, represents the exponential function value, Indicates the The variable data is related to the The normalized similarity between variable data, represents the similarity threshold; The updating of the initial similarity relationship to obtain the similarity relationship between every two variable data includes: By formula: Initial similarity relationship Update and obtain the The variable data is related to the Similarity between variable data ; in, Represents the prior knowledge in the directed graph The variable data is related to the The adjacency relationship between variable data, Represents the activation function.

4. The abnormality prediction method according to claim 1, characterized in that: The updating of all aggregate features to obtain multiple final aggregate features includes: Calculate the correlation between each two aggregated features; All aggregate features are updated according to all correlation degrees to obtain multiple final aggregate features.

5. The abnormality prediction method according to claim 4, characterized in that: The calculating the correlation degree between each two aggregated features includes: By formula: Calculate the The mode and Between the modes The degree of relevance of historical moments ; in, Indicates the The mode is in The set of all variable data corresponding to a historical moment, Indicates the The historical moment The variable data is related to the The historical moment The relationship between the variable data, , Indicates the The historical moment The variable data is related to the The historical moment The comprehensive similarity between variable data; The method updates all aggregate features according to all association degrees to obtain multiple final aggregate features, including: By formula: Update all aggregate features; in, Indicates The intermediate aggregate feature matrix of the layer, represents the correlation matrix, Indicates The learnable weight matrix of the layer, Indicates The intermediate aggregate feature matrix of the layer, represents the activation function, , represents the update of the last layer, when hour, represents the aggregate feature matrix, when hour, Represents the final aggregated feature matrix.

6. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the abnormality prediction method of the hypoxia expansion stage of stem cells according to any one of claims 1 to 5 is implemented.

Images