Physiological state category prediction model training method, system, equipment and medium

Abstract

The present application discloses a physiological state category prediction model training method, system, device and medium, which relates to the field of model training, including obtaining a gene expression matrix of a preset physiological state, extracting a gene expression feature vector in the gene expression matrix; obtaining a core feature of a preset physiological state, selecting one or more feature genes associated with the core feature, and combining the core feature and the feature gene to construct an attribute signature associated with the preset physiological state; projecting the gene expression feature vector into the attribute signature to obtain a predefined mapping matrix; inputting the predefined mapping matrix into the prediction model to obtain a trained prediction model. The present application projects the mapping matrix learned by the model in the training phase into the attribute space to match the feature vector with its corresponding attribute signature, thereby achieving that for new biological process categories, the model can accurately classify based only on the attribute description.

Description

Technical Field

[0001] The present application relates to the field of model training, and specifically to a physiological state category prediction model training method, system, device and medium. Background Art

[0002] Currently, the field of model-based prediction is generally based on traditional machine learning and relies on manually selected and optimized feature parameters. The selection and optimization of these feature parameters usually require professional knowledge and experience, and the selection of feature parameters is a time-consuming and labor-intensive process. Different feature parameter selections may lead to huge differences in prediction results, and manually selected feature parameters may not be able to capture all information.

[0003] In some related technologies, the above model prediction methods usually require manual participation in the selection of feature parameters and the model training process, which greatly limits the degree of automation of the method and increases the cost and time of manual intervention; the performance of the model is also highly dependent on the selection and optimization of feature parameters. Acquiring a large amount of high-quality annotated data is a time-consuming and costly process, which may limit the prediction process and results. In addition, these methods are also limited to the expressive power of shallow models and may not be able to capture deeper relationships and patterns when dealing with complex gene regulatory networks. Summary of the invention

[0004] In order to solve at least one of the problems mentioned in the above background technology, the present application provides a physiological state category prediction model training method, system, device and medium, which can realize classification without labeled data, saving the time of obtaining sample data sets. At the same time, in this scheme, the mapping matrix learned by the model in the training phase is projected to the attribute space to match the feature vector with its corresponding attribute signature, so that for new biological process categories, the model can accurately classify based only on the attribute description.

[0005] The specific technical solutions provided by the embodiments of this application are as follows:

[0006] In a first aspect, a method for training a physiological state category prediction model is provided, the method comprising:

[0007] Acquire a gene expression matrix of a preset physiological state, and extract a gene expression feature vector in the gene expression matrix;

[0008] Acquire a core feature of a preset physiological state, select one or more characteristic genes associated with the core feature, and construct an attribute signature associated with the preset physiological state by combining the core feature and the characteristic gene;

[0009] Projecting the gene expression feature vector into the attribute signature to obtain a predefined mapping matrix;

[0010] The predefined mapping matrix is ​​input into the prediction model to obtain a trained prediction model.

[0011] In a specific embodiment, a gene expression matrix of a preset physiological state is obtained, and a gene expression feature vector in the gene expression matrix is ​​extracted, specifically including: collecting gene expression data associated with the preset physiological state, combining all the gene expression data to generate a gene expression matrix; and performing convolution and pooling processing on the gene expression matrix through a convolutional neural network to generate the gene expression feature vector.

[0012] In a specific embodiment, the convolutional neural network includes several convolutional layers, pooling layers and fully connected layers. A preset convolutional layer, a preset pooling layer or a preset fully connected layer is selected to extract the gene expression pattern features in the gene expression matrix, and the gene expression pattern features extracted by the preset convolutional layer, the preset pooling layer or the preset fully connected layer are combined to form the gene expression feature vector.

[0013] In a specific embodiment, a preset convolutional layer, a preset pooling layer and a preset fully connected layer are selected to extract the gene expression pattern characteristics in the gene expression matrix, specifically including: using a weight matrix to scan a preset area in the gene expression matrix, and obtaining a feature map obtained by multiplying the gene expression data in the preset area of ​​the weight matrix; compressing the feature map that exceeds the preset size into a normal-size feature map, and configuring the normal-size feature map to have gene characteristics to form the gene expression pattern characteristics.

[0014] In a specific embodiment, the core feature and the characteristic gene are combined to construct an attribute signature associated with a preset physiological state, specifically including: defining each of the characteristic genes to be associated with a state attribute of the preset physiological state, the state attribute being used to describe the manifestation characteristics, molecular mechanism or the core feature of the preset physiological state; and constructing a multidimensional vector in combination with the characteristic gene and all state attributes associated with the characteristic gene to generate the attribute signature.

[0015] In a specific embodiment, at least two features associated with the preset physiological state are obtained, and the features are divided into performance features and principle features; and the performance features and the principle features are sorted according to the degree of association between the core features and the preset physiological state; the features of preset positions in the performance features and the principle features are obtained respectively, and the features of all the preset positions are combined to form a core feature.

[0016] In a specific embodiment, a preset loss function associated with the prediction model is obtained, and the predefined mapping matrix is ​​constrained by minimizing the sum of the preset loss function and the regularization term;

[0017] The predefined mapping matrix is ​​calculated by formula (1):

[0018] (1)

[0019] Where V represents the predefined mapping matrix; X represents the gene expression feature vector; represents the product of the gene expression feature vector; represents the first hyperparameter; Y represents the preset physiological state label; represents the second hyperparameter; I represents the identity matrix; S represents the attribute signature; Represents the self-product of an attribute signature.

[0020] In a specific embodiment, the constrained predefined mapping matrix is ​​obtained, and the constrained predefined mapping matrix is ​​input into the prediction model to obtain a trained prediction model;

[0021] Formula (2) is the trained prediction model:

[0022] (2)

[0023] Among them, P represents the physiological state prediction result; X represents the gene expression feature vector; represents the transpose of the gene expression feature vector; V represents the predefined mapping matrix; S represents the attribute signature; i represents the number of rows of the preset column in the attribute signature.

[0024] In a specific embodiment, the minimization of the sum of the preset loss function and the regularization term is calculated as follows:

[0025] (3)

[0026] Among them, L represents the preset loss function, represents the regularization term; X represents the gene expression feature vector; represents the transpose of the gene expression feature vector; V represents the predefined mapping matrix; S represents the attribute signature; Y represents the preset physiological state label; Indicates that the predefined mapping matrix is ​​a real matrix with d rows and a columns;

[0027] Set the minimized preset loss function to the binary norm of the matrix, and the calculation formula (4) of the regularization term is as follows:

[0028] (4)

[0029] in, represents the regularization term, represents the two-norm of the matrix; represents the first hyperparameter; represents the second hyperparameter; represents the third hyperparameter; V represents the predefined mapping matrix; S represents the attribute signature; Represents the transpose of the gene expression feature vector.

[0030] In a specific embodiment, a similarity matrix is ​​calculated by combining the gene expression feature vector and the attribute signature; a preset number of physiological state categories in the similarity matrix are screened out in descending order of similarity, and it is analyzed whether there are actual physiological state categories in the preset number of physiological state categories; if so, the prediction model is confirmed to be a trained prediction model.

[0031] In a specific embodiment, each row in the similarity matrix represents the similarity between a characteristic gene and all physiological state categories, and the similarity adopts cosine similarity.

[0032] In a specific embodiment, a new sample gene expression matrix is ​​collected, and a new gene expression vector is extracted, and new sample features and new sample feature genes associated with the new sample features are obtained; the new sample features and the new sample feature genes are combined to construct a new sample state attribute signature; the new sample gene expression matrix and the new sample state attribute signature are input into the trained prediction model to obtain a physiological state category prediction result of the new sample.

[0033] In a third aspect, a computer program product is provided, including a computer program, which, when executed by a processor, implements the steps of the method described below:

[0034] Step A: obtaining a gene expression matrix of a preset physiological state, and extracting a gene expression feature vector in the gene expression matrix;

[0035] Step B: obtaining a core feature of a preset physiological state, selecting one or more characteristic genes associated with the core feature, and constructing an attribute signature associated with the preset physiological state by combining the core feature and the characteristic gene;

[0036] Step C: projecting the gene expression feature vector into the attribute signature to obtain a predefined mapping matrix;

[0037] Step D: Input the predefined mapping matrix into the prediction model to obtain a trained prediction model.

[0038] In a fourth aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the following steps are implemented:

[0039] Step A: obtaining a gene expression matrix of a preset physiological state, and extracting a gene expression feature vector in the gene expression matrix;

[0040] Step B: obtaining a core feature of a preset physiological state, selecting one or more characteristic genes associated with the core feature, and constructing an attribute signature associated with the preset physiological state by combining the core feature and the characteristic gene;

[0041] Step C: projecting the gene expression feature vector into the attribute signature to obtain a predefined mapping matrix;

[0042] Step D: Input the predefined mapping matrix into the prediction model to obtain a trained prediction model.

[0043] In a fifth aspect, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0044] Step A: obtaining a gene expression matrix of a preset physiological state, and extracting a gene expression feature vector in the gene expression matrix;

[0045] Step B: obtaining a core feature of a preset physiological state, selecting one or more characteristic genes associated with the core feature, and constructing an attribute signature associated with the preset physiological state by combining the core feature and the characteristic gene;

[0046] Step C: projecting the gene expression feature vector into the attribute signature to obtain a predefined mapping matrix;

[0047] Step D: Input the predefined mapping matrix into the prediction model to obtain a trained prediction model.

[0048] The embodiments of the present application have the following beneficial effects:

[0049] 1. In the embodiments of the present application, the existing physiological state types and corresponding data are first processed, including obtaining a gene expression matrix, extracting gene expression feature vectors in the gene expression matrix, obtaining core features, and selecting feature genes associated with the core features, combining the core features and feature genes to construct an attribute signature, projecting the gene expression feature vectors into the attribute signature, obtaining a predefined mapping matrix, inputting the predefined mapping matrix into the prediction model, and obtaining a trained prediction model, which can realize classification without labeled data, saving time for obtaining sample data sets. At the same time, in this solution, the mapping matrix learned by the model in the training phase is projected into the attribute space to match the feature vector with its corresponding attribute signature, thereby realizing that for new biological process categories, the model can accurately classify based only on the attribute description. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments 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 creative work.

[0051] Figure 1 A schematic diagram of an application environment according to the present application is shown;

[0052] Figure 2 A schematic diagram of a physiological state category prediction model training method according to the present application is shown;

[0053] Figure 3 A schematic diagram of a computer device is shown. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0055] The physiological state category prediction model training method provided in this application can be applied to Figure 1 In the application environment shown. The terminal 102 communicates with the server 104 through the network. The terminal 102 can process the preset physiological state and extract the gene expression feature vector in the gene expression matrix; obtain the core features of the preset physiological state, combine the core features and the characteristic genes to construct an attribute signature associated with the preset physiological state, project the gene expression feature vector into the attribute signature, and obtain a predefined mapping matrix; input the predefined mapping matrix into the prediction model to realize the training of the prediction model. The terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers and portable wearable devices, and the server 104 can be implemented with an independent server or a server cluster consisting of multiple servers.

[0056] In one embodiment, Figure 2 As shown, a physiological state category prediction model training method is provided, and the method is applied to Figure 1 The terminal in is used as an example to illustrate, including the following steps:

[0057] Step 101: Obtain a gene expression matrix of a preset physiological state, and extract a gene expression feature vector in the gene expression matrix.

[0058] In this embodiment, it specifically includes: collecting gene expression data associated with the preset physiological state, combining all the gene expression data to generate a gene expression matrix; performing convolution and pooling processing on the gene expression matrix through a convolutional neural network to generate the gene expression feature vector.

[0059] Furthermore, the convolutional neural network includes several convolutional layers, pooling layers and fully connected layers. Preset convolutional layers, preset pooling layers and preset fully connected layers are selected to extract gene expression pattern features in the gene expression matrix, and the gene expression pattern features extracted by the preset convolutional layer, preset pooling layer or preset fully connected layer are combined to form the gene expression feature vector.

[0060] Among them, selecting a preset convolutional layer, a preset pooling layer and a preset fully connected layer to extract the gene expression pattern characteristics in the gene expression matrix specifically includes: using a weight matrix to scan a preset area in the gene expression matrix, and obtaining a feature map obtained by multiplying the gene expression data in the preset area of ​​the weight matrix; compressing the feature map that exceeds the preset size into a normal-size feature map, and configuring the normal-size feature map to have gene characteristics to form the gene expression pattern characteristics.

[0061] In this embodiment, a physiological state category data set is first obtained, that is, the gene expression data generated by high-throughput sequencing or other technologies are organized into a matrix format. Each column of the matrix represents a sample, and each row represents a gene. In the gene expression matrix, during the data processing process, the gene expression matrix is ​​set as an "image", in which each "pixel" represents the value of gene expression, and this structured data is processed by a convolutional neural network. The convolutional neural network consists of multiple convolutional layers, pooling layers, and fully connected layers. For each layer, the network extracts local gene expression pattern features, thereby gradually constructing a global gene expression relationship.

[0062] The convolutional layer processing process includes the convolutional neural network using a weight matrix to scan the local area of ​​the gene expression matrix one by one. At each scan, the weight matrix is ​​multiplied by the local gene expression data to generate a feature map. Through multiple weight matrices, the convolutional neural network can extract different feature patterns in the gene expression data, and the patterns can be used to express changes in gene expression related to a preset physiological state.

[0063] Among them, the pooling layer processing process includes that the pooling layer usually uses maximum pooling or average pooling to compress a larger feature map into a smaller map while retaining the most important information, which is critical for extracting more abstract expression features.

[0064] After being processed by several convolutional and pooling layers, the gene expression matrix is ​​transformed by the convolutional neural network into a more compressed representation with abstract features. This representation is called a feature vector. The feature vector contains the gene expression pattern of the relevant physiological state. Each feature vector can be regarded as a high-dimensional representation of the gene expression data, representing different features related to different categories.

[0065] For example, the gene expression data of Alzheimer's disease is associated with the high expression of specific genes, among which the characteristic genes include at least "APP" and "PSEN1", and the characteristic genes are closely related to the attributes of neurodegenerative diseases such as "memory loss". Through convolutional layers and pooling layers, the convolutional neural network extracts the features of characteristic gene expression and finally associates them with the attribute signature of the disease, thereby helping the model better identify the physiological state categories in new data in the inference stage.

[0066] Step 102: Acquire a core feature of a preset physiological state, select one or more characteristic genes associated with the core feature, and construct an attribute signature associated with the preset physiological state by combining the core feature and the characteristic gene.

[0067] First, the core biological features are confirmed according to the type of preset physiological state, and the core biological features associated with the disease are extracted from medical literature, gene databases, and existing gene expression studies. In order to improve the prediction accuracy of the prediction model, the accuracy of the training data also needs to be improved during the model training process. Therefore, the core features are used as much as possible, wherein the core features usually include biological features associated with or showing characteristics of different physiological states, such as the expression of specific genes or gene pathways, and features related to abnormal changes at the cell, tissue or system level; specifically, at least two disease features associated with the preset physiological state are obtained, and the features are divided into manifestation features and principle features; and the manifestation features and the principle features are sorted according to the correlation between the core features and the preset physiological state; the features of the preset positions in the manifestation features and the principle features are obtained respectively, and the features of all the preset positions are combined to form the core features. The core features are screened through the above process to ensure the accuracy of model training.

[0068] Furthermore, combining the core features and the characteristic genes to construct an attribute signature associated with a preset physiological state specifically includes: defining each of the characteristic genes to be associated with a state attribute of the preset physiological state, the state attribute being used to describe the manifestation characteristics, molecular mechanisms, pathological characteristics or the core features of the preset physiological state; and constructing a multidimensional vector in combination with the characteristic genes and all state attributes associated with the characteristic genes to generate the attribute signature.

[0069] It should be noted that the characteristic genes of different physiological states may involve multiple genes or gene regulatory networks, so it is necessary to select the most representative genes to reflect different aspects of different physiological states; for example, for Alzheimer's disease, the selected characteristic genes include "APP, PSEN1 and PSEN2", which are closely related to the metabolic process of Aβ protein.

[0070] Furthermore, the corresponding state attributes are defined for the selected characteristic genes for association, and the attributes describe the main symptoms, progression or cellular mechanism of the disease. This step further refines the attribute signature by mapping the expression pattern of the gene to a certain state attribute; illustratively, the state attributes include "symptoms" or "dysfunctions" used to describe the physiological state, such as "memory loss" or "neuronal death". In Alzheimer's disease, the abnormal expression of the "APP gene" may be related to the "memory loss" attribute, while the PSEN1 gene may be related to "cognitive dysfunction".

[0071] Based on the previously selected genes and their corresponding state attributes, attribute signature vectors of different physiological states are constructed. Attribute signature vectors can be understood as a set of multidimensional vectors describing different physiological states, where each dimension corresponds to a specific physiological state category. The attribute signature vector reflects the overall characteristics, helping the model to match gene expression data with physiological state categories during training and inference.

[0072] Exemplarily, the Alzheimer's disease attribute signature vector may be defined as: [memory loss, cognitive dysfunction, Aβ protein deposition, neuronal degeneration].

[0073] Step 103: Projecting the gene expression feature vector into the attribute signature to obtain a predefined mapping matrix;

[0074] Step 104: input the predefined mapping matrix into the prediction model to obtain a trained prediction model.

[0075] During the model training process, the gene expression feature vector is obtained through convolutional neural network processing and matched with the attribute signature. During the training, a mapping matrix is ​​defined to project the gene expression feature vector into the attribute signature space. By optimizing the matrix, the gene expression feature vector and its corresponding attribute signature are matched as much as possible, thereby improving the classification accuracy of the prediction model.

[0076] The predefined mapping matrix is ​​calculated by formula (1):

[0077] (1)

[0078] Where V represents the predefined mapping matrix; X represents the gene expression feature vector; represents the product of the gene expression feature vector; represents the first hyperparameter; Y represents the preset physiological state label; represents the second hyperparameter; I represents the identity matrix; S represents the attribute signature; Represents the self-product of an attribute signature.

[0079] It should be noted that the constrained predefined mapping matrix is ​​obtained, and the constrained predefined mapping matrix is ​​input into the prediction model to obtain a trained prediction model;

[0080] Formula (2) is the trained prediction model:

[0081] (2)

[0082] Among them, P represents the physiological state prediction result; X represents the gene expression feature vector; represents the transpose of the gene expression feature vector; V represents the predefined mapping matrix; S represents the attribute signature; i represents the number of rows of the preset column in the attribute signature.

[0083] By obtaining the minimized preset loss function and setting the minimized preset loss function to the Frobenius norm, the sum calculation formula (3) for minimizing the preset loss function and the regularization term is obtained as follows:

[0084] (3)

[0085] Among them, L represents the preset loss function, represents the regularization term; X represents the gene expression feature vector; represents the transpose of the gene expression feature vector; V represents the predefined mapping matrix; S represents the attribute signature; Y represents the preset physiological state label.

[0086] Set the minimized preset loss function to the binary norm of the matrix, and the calculation formula (4) of the regularization term is as follows:

[0087] (4)

[0088] in, represents the regularization term, represents the two-norm of the matrix; represents the first hyperparameter; represents the second hyperparameter; represents the third hyperparameter; V represents the predefined mapping matrix; S represents the attribute signature; Represents the transpose of the gene expression feature vector.

[0089] It should be noted that the loss function L in this embodiment is used to measure the gap between the prediction results of the prediction model and the actual label. Its function is to minimize the error between the predicted value and the true value to guide the prediction model to learn better feature representation; the regularization term is used to impose constraints on the mapping matrix during the optimization process to prevent the prediction model from overfitting and ensure the generalization ability of the model. The regularizer includes controlling the Euclidean norm of the attribute signature in the feature space, limiting the variance of the instance representation in the attribute space, and the standard weight decay term.

[0090] It should also be noted that γ, λ, and β in formula (4) are hyperparameters, which control the Euclidean norm of the attribute signature, the variance of the instance representation, and the size of the model parameters, respectively, to ensure the stability and generalization ability of the model and prevent overfitting. γ ensures that the representation of the attribute signature in the feature space has a controlled norm to prevent the representation from being too large. λ limits the variance of the instance representation in the attribute space to ensure that the representation of the instance in the attribute space does not fluctuate too much. β is used to prevent the prediction model from overfitting by imposing a penalty on the size of the model parameters to prevent the model from overfitting on the training data.

[0091] In a specific embodiment, the present implementation also provides an exploration of the domain adaptability of the present model, wherein the sum of the minimization of the preset loss function and the regularization term (3) can be expanded into the following expression (5):

[0092] (5)

[0093] Wherein, V represents the predefined mapping matrix; A gene expression feature vector representing a preset physiological state; a signature of attributes representing a preset physiological state; A label indicating whether the preset physiological state i belongs to category t; represents the loss function; represents the regularization term.

[0094] Further, the following conversion process is performed to vectorize the mapping matrix V into a vector v, which is done by using the vectorization operator This is done by stacking the columns of the matrix into a single vector and by and the property signature Combine into a new vector , the new vector Defined as ; After vectorization, the inner product Equivalent to ; then use the new vector and the vectorized parameter v to obtain the sum expression (6) that minimizes the preset loss function and the regularization term;

[0095] (6)

[0096] Where v represents the vectorized parameter; represents the new vector; A label indicating whether the preset physiological state i belongs to category t; represents the loss function; represents the regularization term.

[0097] The domain adaptation characteristics of the training method in this embodiment are verified through the above conversion process, and then the rationality and adaptability of this prediction model are explained. The domain adaptation technology and prediction model in this embodiment realize accurate classification and prediction of new categories in the absence of labeled data. By applying this method, researchers can deeply explore the gene expression patterns in the physiological states of different organisms and reveal the regulatory relationships and functional networks between genes. This analysis is not only helpful for basic biological research, but can also be applied to the study of the mechanisms of specific diseases.

[0098] Furthermore, the present implementation also includes optimizing the combination of minimizing the source risk and the difference measure between the source domain and the target domain, specifically, the standard limit of the target risk and It can be expressed as expression (7):

[0099] (7)

[0100] in, H divergence represents the difference between the source domain and the target domain distribution. represents a term that takes into account the complexity of the hypothesis class H and the best achievable joint error on the two domains.

[0101] In extreme cases, when the attribute labels of the training class and the attribute labels of the test class are the same, in this case, the distributions of the training class and the test class are the same, resulting in zero inter-domain difference, then the performance of the model on the test data will be directly related to its performance on the training data, without any additional inter-domain difference penalty. When the two attribute signatures are completely different, resulting in the largest inter-domain difference. Orthogonality means that the training class does not provide any useful information for predicting the test class. The risk boundary becomes invalid, which means that it does not provide a useful boundary for the target error. The inter-domain difference term dominates the boundary, indicating that the generalization error may be very large. If the attribute signatures of the training class and the test class are orthogonal, then the model may perform poorly on the test data because the information learned from the training data cannot be applied to the test data. Understanding these extreme cases helps design zero-shot learning models, emphasizing that when selecting or designing attribute signatures, there should be a certain overlap or similarity between the training class and the test class. In practical applications, ensuring that the attribute signatures of the training class and the test class are not completely separated can reduce the inter-domain difference and improve the generalization performance of the model. The reliability of the model in this embodiment is verified through the above process.

[0102] In a specific embodiment, the prediction model verification process is also included, specifically including combining the gene expression feature vector and the attribute signature to calculate a similarity matrix; filtering out a preset number of physiological state categories in the similarity matrix in descending order of similarity, and analyzing whether there are actual physiological state categories in the preset number of physiological state categories; if there are, confirming that the prediction model is a trained prediction model. Wherein, each row in the similarity matrix represents the similarity between a characteristic gene and all physiological state categories, and the similarity adopts cosine similarity.

[0103] Exemplarily, first, by calculating the similarity matrix between the gene expression feature vector and the state attribute signature, the model selects the top k most similar physiological state categories in its prediction results; then, the top k categories predicted by the model are compared with the actual physiological state categories to determine whether the correct physiological state category is included in the prediction results. If so, the prediction model is confirmed to be a trained prediction model. This reflects whether the model fully utilizes the gene expression features in the physiological state classification task and whether it can correctly match the corresponding state attribute signature, thereby evaluating the model's ability to recognize unknown categories.

[0104] In a specific embodiment, after the prediction model is trained, the trained prediction model is used to predict the physiological state category of the new sample, the new sample gene expression matrix is ​​collected, and the new gene expression vector is extracted, and the new sample features and the new sample feature genes associated with the new sample features are obtained; the new sample features and the new sample feature genes are combined to construct a new sample state attribute signature; the new sample gene expression matrix and the new sample state attribute signature are input into the trained prediction model to obtain the physiological state category prediction result of the new sample.

[0105] It should be understood that although Figure 2 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 2 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0106] In one embodiment, a computer program product is provided, including a computer program, which, when executed by a processor, implements the steps of the method described below:

[0107] Step 201, obtaining a gene expression matrix of a preset physiological state, and extracting a gene expression feature vector in the gene expression matrix;

[0108] Step 202: Acquire a core feature of a preset physiological state, select one or more characteristic genes associated with the core feature, and construct an attribute signature associated with the preset physiological state by combining the core feature and the characteristic gene;

[0109] Step 203: Project the gene expression feature vector into the attribute signature to obtain a predefined mapping matrix;

[0110] Step 204: input the predefined mapping matrix into the prediction model to obtain a trained prediction model.

[0111] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0112] Obtaining a gene expression matrix of a preset physiological state and extracting a gene expression feature vector in the gene expression matrix specifically includes: collecting gene expression data associated with the preset physiological state, combining all the gene expression data to generate a gene expression matrix; performing convolution and pooling processing on the gene expression matrix through a convolutional neural network to generate the gene expression feature vector.

[0113] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0114] The convolutional neural network includes several convolutional layers, pooling layers and fully connected layers. A preset convolutional layer, a preset pooling layer or a preset fully connected layer is selected to extract the gene expression pattern features in the gene expression matrix, and the gene expression pattern features extracted by the preset convolutional layer, the preset pooling layer or the preset fully connected layer are combined to form the gene expression feature vector.

[0115] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0116] Selecting a preset convolutional layer, a preset pooling layer, and a preset fully connected layer to extract the gene expression pattern features in the gene expression matrix specifically includes: using a weight matrix to scan a preset area in the gene expression matrix, and obtaining a feature map obtained by multiplying the gene expression data in the preset area of ​​the weight matrix; compressing a feature map that exceeds a preset size into a normal-size feature map, and configuring the normal-size feature map to have gene features to form the gene expression pattern features.

[0117] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0118] Define that each of the characteristic genes is associated with a state attribute of the preset physiological state, and the state attribute is used to describe the manifestation characteristics, molecular mechanisms, pathological characteristics or the core characteristics of the preset physiological state; and construct a multidimensional vector by combining the characteristic gene and all the state attributes associated with the characteristic gene to generate the attribute signature.

[0119] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0120] Acquire at least two features associated with the preset physiological state, and divide the features into performance features and principle features; and sort the performance features and the principle features respectively according to the degree of association between the core features and the preset physiological state; acquire the features of preset positions in the performance features and the principle features respectively, and combine the features of all the preset positions to form a core feature.

[0121] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0122] Obtaining a preset loss function associated with the prediction model, and constraining the predefined mapping matrix by minimizing the sum of the preset loss function and a regularization term;

[0123] The predefined mapping matrix is ​​calculated by formula (1):

[0124] (1)

[0125] Where V represents the predefined mapping matrix; X represents the gene expression feature vector; represents the product of the gene expression feature vector; represents the first hyperparameter; Y represents the preset physiological state label; represents the second hyperparameter; I represents the identity matrix; S represents the attribute signature; Represents the self-product of an attribute signature.

[0126] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0127] Obtaining the constrained predefined mapping matrix, and inputting the constrained predefined mapping matrix into the prediction model to obtain a trained prediction model;

[0128] Formula (2) is the trained prediction model:

[0129] (2)

[0130] Among them, P represents the physiological state prediction result; X represents the gene expression feature vector; represents the transpose of the gene expression feature vector; V represents the predefined mapping matrix; S represents the attribute signature; i represents the number of rows of the preset column in the attribute signature.

[0131] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0132] The minimization of the preset loss function and the regularization term is calculated by the following equation (3):

[0133] (3)

[0134] Among them, L represents the preset loss function, represents the regularization term; X represents the gene expression feature vector; represents the transpose of the gene expression feature vector; V represents the predefined mapping matrix; S represents the attribute signature; Y represents the preset physiological state label;

[0135] Set the minimized preset loss function to the binary norm of the matrix, and the calculation formula (4) of the regularization term is as follows:

[0136] (4)

[0137] in, represents the regularization term, represents the two-norm of the matrix; represents the first hyperparameter; represents the second hyperparameter; represents the third hyperparameter; V represents the predefined mapping matrix; S represents the attribute signature; Represents the transpose of the gene expression feature vector.

[0138] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0139] A similarity matrix is ​​calculated by combining the gene expression feature vector and the attribute signature; a preset number of physiological state categories in the similarity matrix are screened out in descending order of similarity, and it is analyzed whether there are actual physiological state categories in the preset number of physiological state categories; if so, it is confirmed that the prediction model is a trained prediction model.

[0140] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0141] Each row in the similarity matrix represents the similarity between a characteristic gene and all physiological state categories, and the similarity adopts cosine similarity.

[0142] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0143] Collect the new sample gene expression matrix, extract the new gene expression vector, obtain the new sample feature and the new sample feature gene associated with the new sample feature; construct a new sample state attribute signature by combining the new sample feature and the new sample feature gene; input the new sample gene expression matrix and the new sample state attribute signature into the trained prediction model to obtain the physiological state category prediction result of the new sample.

[0144] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 3As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the associated features and characteristic gene data of the preset physiological state. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a physiological state category prediction model training method is implemented.

[0145] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0146] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0147] Step 301, obtaining a gene expression matrix of a preset physiological state, and extracting a gene expression feature vector in the gene expression matrix;

[0148] Step 302: Acquire a core feature of a preset physiological state, select one or more characteristic genes associated with the core feature, and construct an attribute signature associated with the preset physiological state by combining the core feature and the characteristic gene;

[0149] Step 303: Project the gene expression feature vector into the attribute signature to obtain a predefined mapping matrix;

[0150] Step 304: input the predefined mapping matrix into the prediction model to obtain a trained prediction model.

[0151] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0152] The gene expression data associated with the preset physiological state are collected, and a gene expression matrix is ​​generated by combining all the gene expression data; the gene expression matrix is ​​convolved and pooled by a convolutional neural network to generate the gene expression feature vector.

[0153] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0154] The convolutional neural network includes several convolutional layers, pooling layers and fully connected layers. Preset convolutional layers, preset pooling layers and preset fully connected layers are selected to extract gene expression pattern features in the gene expression matrix, and the gene expression pattern features extracted by the preset convolutional layers, preset pooling layers or preset fully connected layers are combined to form the gene expression feature vector.

[0155] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0156] A weight matrix is ​​used to scan a preset area in the gene expression matrix, and a feature map obtained by multiplying the gene expression data in the preset area of ​​the weight matrix is ​​obtained; a feature map exceeding a preset size is compressed into a normal-size feature map, and the normal-size feature map is configured to have gene features to form the gene expression pattern feature.

[0157] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0158] Define that each of the characteristic genes is associated with a state attribute of the preset physiological state, and the state attribute is used to describe the performance characteristics, molecular mechanism or core characteristics of the preset physiological state; construct a multidimensional vector by combining the characteristic gene and all the state attributes associated with the characteristic gene to generate the attribute signature.

[0159] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0160] Acquire at least two features associated with the preset physiological state, and divide the features into performance features and principle features; and sort the performance features and the principle features respectively according to the degree of association between the core features and the preset physiological state; acquire the features of preset positions in the performance features and the principle features respectively, and combine the features of all the preset positions to form a core feature.

[0161] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0162] Obtaining a preset loss function associated with the prediction model, and constraining the predefined mapping matrix by minimizing the sum of the preset loss function and a regularization term;

[0163] The predefined mapping matrix is ​​calculated by formula (1):

[0164] (1)

[0165] Where V represents the predefined mapping matrix; X represents the gene expression feature vector; represents the product of the gene expression feature vector; represents the first hyperparameter; Y represents the preset physiological state label; represents the second hyperparameter; I represents the identity matrix; S represents the attribute signature; Represents the self-product of an attribute signature.

[0166] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0167] Obtaining the constrained predefined mapping matrix, and inputting the constrained predefined mapping matrix into the prediction model to obtain a trained prediction model;

[0168] Formula (2) is the trained prediction model:

[0169] (2)

[0170] Among them, P represents the physiological state prediction result; X represents the gene expression feature vector; represents the transpose of the gene expression feature vector; V represents the predefined mapping matrix; S represents the attribute signature; i represents the number of rows of the preset column in the attribute signature.

[0171] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0172] The minimization of the preset loss function and the regularization term is calculated by the following equation (3):

[0173] (3)

[0174] Among them, L represents the preset loss function, represents the regularization term; X represents the gene expression feature vector; represents the transpose of the gene expression feature vector; V represents the predefined mapping matrix; S represents the attribute signature; Y represents the preset physiological state label;

[0175] Set the minimized preset loss function to the binary norm of the matrix, and the calculation formula (4) of the regularization term is as follows:

[0176] (4)

[0177] in, represents the regularization term, represents the two-norm of the matrix; represents the first hyperparameter; represents the second hyperparameter; represents the third hyperparameter; V represents the predefined mapping matrix; S represents the attribute signature; Represents the transpose of the gene expression feature vector.

[0178] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0179] A similarity matrix is ​​calculated by combining the gene expression feature vector and the attribute signature; a preset number of physiological state categories in the similarity matrix are screened out in descending order of similarity, and it is analyzed whether there are actual physiological state categories in the preset number of physiological state categories; if so, it is confirmed that the prediction model is a trained prediction model.

[0180] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0181] Each row in the similarity matrix represents the similarity between a characteristic gene and all physiological state categories, and the similarity adopts cosine similarity.

[0182] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0183] Collect the new sample gene expression matrix, extract the new gene expression vector, obtain the new sample feature and the new sample feature gene associated with the new sample feature; construct a new sample state attribute signature by combining the new sample feature and the new sample feature gene; input the new sample gene expression matrix and the new sample state attribute signature into the trained prediction model to obtain the physiological state category prediction result of the new sample.

[0184] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0185] Step 401, obtaining a gene expression matrix of a preset physiological state, and extracting a gene expression feature vector in the gene expression matrix;

[0186] Step 402: Acquire a core feature of a preset physiological state, select one or more characteristic genes associated with the core feature, and construct an attribute signature associated with the preset physiological state by combining the core feature and the characteristic gene;

[0187] Step 403: Project the gene expression feature vector into the attribute signature to obtain a predefined mapping matrix;

[0188] Step 404: input the predefined mapping matrix into the prediction model to obtain a trained prediction model.

[0189] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0190] The gene expression data associated with the preset physiological state are collected, and a gene expression matrix is ​​generated by combining all the gene expression data; the gene expression matrix is ​​convolved and pooled by a convolutional neural network to generate the gene expression feature vector.

[0191] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0192] The convolutional neural network includes several convolutional layers, pooling layers and fully connected layers. Preset convolutional layers, preset pooling layers and preset fully connected layers are selected to extract gene expression pattern features in the gene expression matrix, and the gene expression pattern features extracted by the preset convolutional layers, preset pooling layers or preset fully connected layers are combined to form the gene expression feature vector.

[0193] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0194] A weight matrix is ​​used to scan a preset area in the gene expression matrix, and a feature map obtained by multiplying the gene expression data in the preset area of ​​the weight matrix is ​​obtained; a feature map exceeding a preset size is compressed into a normal-size feature map, and the normal-size feature map is configured to have gene features to form the gene expression pattern feature.

[0195] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0196] Define that each of the characteristic genes is associated with a state attribute of the preset physiological state, and the state attribute is used to describe the performance characteristics, molecular mechanism or core characteristics of the preset physiological state; construct a multidimensional vector by combining the characteristic gene and all the state attributes associated with the characteristic gene to generate the attribute signature.

[0197] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0198] Acquire at least two features associated with the preset physiological state, and divide the features into performance features and principle features; and sort the performance features and the principle features respectively according to the degree of association between the core features and the preset physiological state; acquire the features of preset positions in the performance features and the principle features respectively, and combine the features of all the preset positions to form a core feature.

[0199] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0200] Obtaining a preset loss function associated with the prediction model, and constraining the predefined mapping matrix by minimizing the sum of the preset loss function and a regularization term;

[0201] The predefined mapping matrix is ​​calculated by formula (1):

[0202] (1)

[0203] Where V represents the predefined mapping matrix; X represents the gene expression feature vector; represents the product of the gene expression feature vector; represents the first hyperparameter; Y represents the preset physiological state label; represents the second hyperparameter; I represents the identity matrix; S represents the attribute signature; Represents the self-product of an attribute signature.

[0204] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0205] Obtaining the constrained predefined mapping matrix, and inputting the constrained predefined mapping matrix into the prediction model to obtain a trained prediction model;

[0206] Formula (2) is the trained prediction model:

[0207] (2)

[0208] Among them, P represents the physiological state prediction result; X represents the gene expression feature vector; represents the transpose of the gene expression feature vector; V represents the predefined mapping matrix; S represents the attribute signature; i represents the number of rows of the preset column in the attribute signature.

[0209] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0210] The minimization of the preset loss function and the regularization term is calculated by the following equation (3):

[0211] (3)

[0212] Among them, L represents the preset loss function, represents the regularization term; X represents the gene expression feature vector; represents the transpose of the gene expression feature vector; V represents the predefined mapping matrix; S represents the attribute signature; Y represents the preset physiological state label;

[0213] Set the minimized preset loss function to the binary norm of the matrix, and the calculation formula (4) of the regularization term is as follows:

[0214] (4)

[0215] in, represents the regularization term, represents the two-norm of the matrix; represents the first hyperparameter; represents the second hyperparameter; represents the third hyperparameter; V represents the predefined mapping matrix; S represents the attribute signature; Represents the transpose of the gene expression feature vector.

[0216] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0217] A similarity matrix is ​​calculated by combining the gene expression feature vector and the attribute signature; a preset number of physiological state categories in the similarity matrix are screened out in descending order of similarity, and it is analyzed whether there are actual physiological state categories in the preset number of physiological state categories; if so, it is confirmed that the prediction model is a trained prediction model.

[0218] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0219] Each row in the similarity matrix represents the similarity between a characteristic gene and all physiological state categories, and the similarity adopts cosine similarity.

[0220] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0221] Collect the new sample gene expression matrix, extract the new gene expression vector, obtain the new sample feature and the new sample feature gene associated with the new sample feature; construct a new sample state attribute signature by combining the new sample feature and the new sample feature gene; input the new sample gene expression matrix and the new sample state attribute signature into the trained prediction model to obtain the physiological state category prediction result of the new sample.

[0222] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0223] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0224] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A physiological state category prediction model training method, characterized in that: The method comprises: Acquire a gene expression matrix of a preset physiological state, and extract a gene expression feature vector in the gene expression matrix; Obtaining a core feature of a preset physiological state, selecting one or more characteristic genes associated with the core feature, and combining the core feature and the characteristic gene to construct an attribute signature associated with the preset physiological state; wherein the core feature includes the expression of a specific gene or gene pathway, and features related to abnormal changes at the cell, tissue or system level; Projecting the gene expression feature vector into the attribute signature to obtain a predefined mapping matrix; inputting the predefined mapping matrix into the prediction model to obtain a trained prediction model; Obtaining a preset loss function associated with the prediction model, and constraining the predefined mapping matrix by minimizing the sum of the preset loss function and a regularization term; Obtaining the constrained predefined mapping matrix, and inputting the constrained predefined mapping matrix into the prediction model to obtain a trained prediction model; Collect the new sample gene expression matrix, extract the new gene expression vector, obtain the new sample feature and the new sample feature gene associated with the new sample feature; construct a new sample state attribute signature by combining the new sample feature and the new sample feature gene; input the new sample gene expression matrix and the new sample state attribute signature into the trained prediction model to obtain the physiological state category prediction result of the new sample.

2. The physiological state category prediction model training method according to claim 1, characterized in that: Obtaining a gene expression matrix of a preset physiological state and extracting a gene expression feature vector in the gene expression matrix specifically includes: Collecting gene expression data associated with the preset physiological state, and combining all the gene expression data to generate a gene expression matrix; The gene expression matrix is ​​convolved and pooled by a convolutional neural network to generate the gene expression feature vector.

3. The physiological state category prediction model training method according to claim 2, characterized in that: The method comprises: The convolutional neural network includes a plurality of convolutional layers, pooling layers and fully connected layers, and a preset convolutional layer, a preset pooling layer and a preset fully connected layer are selected to extract gene expression pattern features in the gene expression matrix. The gene expression feature vector is formed by combining the gene expression pattern features extracted by the preset convolutional layer, the preset pooling layer or the preset fully connected layer.

4. The physiological state category prediction model training method according to claim 3, characterized in that: Selecting a preset convolutional layer, a preset pooling layer, and a preset fully connected layer to extract gene expression pattern features in the gene expression matrix specifically includes: Scanning a preset region in the gene expression matrix using a weight matrix, and obtaining a feature map obtained by multiplying the gene expression data in the preset region of the weight matrix; The feature map compressed beyond a preset size is converted into a normal-size feature map, and the normal-size feature map is configured to have a gene feature to form the gene expression pattern feature.

5. The physiological state category prediction model training method according to any one of claims 1 to 4, characterized in that: Combining the core features and the characteristic genes to construct a property signature associated with a preset physiological state specifically includes: Define that each of the characteristic genes is associated with a state attribute of the preset physiological state, where the state attribute is used to describe the manifestation characteristics, molecular mechanism or core characteristics of the preset physiological state; A multidimensional vector is constructed by combining the characteristic gene and all state attributes associated with the characteristic gene to generate the attribute signature.

6. The physiological state category prediction model training method according to claim 5, characterized in that: The method further comprises: Acquire at least two features associated with the preset physiological state, and divide the features into performance features and principle features; and respectively sorting the performance characteristics and the principle characteristics according to the correlation between the core characteristics and the preset physiological state; The features of the preset positions in the performance features and the principle features are respectively obtained, and the features of all the preset positions are combined to form a core feature.

7. The physiological state category prediction model training method according to any one of claims 1 to 4, characterized in that: The method further comprises: The predefined mapping matrix is ​​calculated by formula (1): ； Where V represents the predefined mapping matrix; X represents the gene expression feature vector; represents the product of the gene expression feature vector; represents the first hyperparameter; Y represents the preset physiological state label; represents the second hyperparameter; I represents the identity matrix; S represents the attribute signature; Represents the self-product of an attribute signature.

8. The physiological state category prediction model training method according to claim 7, characterized in that: The method further comprises: Formula (2) is the trained prediction model: ； Among them, P represents the physiological state prediction result; X represents the gene expression feature vector; represents the transpose of the gene expression feature vector; V represents the predefined mapping matrix; S represents the attribute signature; i represents the number of rows of the preset column in the attribute signature.

9. The physiological state category prediction model training method according to claim 7, characterized in that: The method further comprises: The minimization of the preset loss function and the regularization term is calculated by the following equation (3): ； Among them, L represents the preset loss function, represents the regularization term; X represents the gene expression feature vector; represents the transpose of the gene expression feature vector; V represents the predefined mapping matrix; S represents the attribute signature; Y represents the preset physiological state label; Set the minimized preset loss function to the binary norm of the matrix, and the calculation formula (4) of the regularization term is as follows: ； in, represents the regularization term, represents the two-norm of the matrix; represents the first hyperparameter; represents the second hyperparameter; represents the third hyperparameter; V represents the predefined mapping matrix; S represents the attribute signature; Represents the transpose of the gene expression feature vector.

10. The physiological state category prediction model training method according to any one of claims 1 to 4, characterized in that: The method further comprises: Calculating a similarity matrix by combining the gene expression feature vector and the attribute signature; Filtering out a preset number of physiological state categories in the similarity matrix according to descending ranking of similarity, and analyzing whether there is an actual physiological state category in the preset number of physiological state categories; If so, it is confirmed that the prediction model is a trained prediction model.

11. The physiological state category prediction model training method according to claim 10, characterized in that: The method comprises: Each row in the similarity matrix represents the similarity between a characteristic gene and all physiological state categories, and the similarity adopts cosine similarity.

12. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method described in any one of claims 1 to 11 are implemented.

13. A computer 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 steps of the method described in any one of claims 1 to 11 are implemented.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 11 are implemented.

Images