Multi-label causal feature selection method and system

Abstract

The invention provides a multi-label causal feature selection method and system, and the method comprises the steps: obtaining a multi-label gene expression data set, and carrying out the preprocessing of the multi-label data set; on the basis of the preprocessed multi-tag gene expression data set, searching for approximate Markov blanket features for each tag by using a unary approximate Markov blanket; deleting other tags contained in the approximate Markov blankets of the tags, and recovering the equivalent features to obtain Markov blanket features of the tags; using a symmetry test method to ensure that the selected features belong to the Markov blankets of the labels; and combining the Markov blanket features of all the tags to obtain a gene expression data feature subset, and realizing feature selection of the gene expression data. According to the method, the feature selection method based on the causal structure and the unary approximate Markov blanket is adopted, and the features which are mistakenly deleted due to the equivalent information of the tag are recovered, so that the problem of low accuracy of the existing gene expression data feature selection method is fundamentally solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and in particular relates to a multi-label causal feature selection method and system. Background Art

[0002] At present, multi-label classification learning has been widely used in the biomedical field, and gene expression data is a common data in the biomedical field. However, with the rapid development, the dimensionality and quantity of the collected multi-label gene expression data are growing explosively. If it is not handled properly, it will face the problem of "dimensionality disaster". Therefore, multi-label feature selection of gene expression data has received widespread attention. Many researchers at home and abroad have proposed a large number of multi-label feature selection methods for gene expression data. According to the way they are combined with the learning model, the feature selection methods can be summarized as filtering method, packaging method and embedding method.

[0003] However, most of the existing multi-label feature selection methods for gene expression data only consider the correlation of variables, without considering causality, resulting in a lack of interpretability. Markov blanket-related algorithms have been widely used in single-label tasks, but due to the complexity of multi-label gene expression data, there are currently few Markov algorithms for finding multi-labels. At the same time, when looking for multi-label Markov blankets, existing feature selection methods ignore the omission of features caused by equivalent information. Therefore, the existing multi-label feature selection methods for gene expression data have the problem of inaccurate feature selection.

[0004] Therefore, how to improve the accuracy of feature selection of gene expression data is an urgent problem to be solved in this field. Summary of the invention

[0005] The purpose of the present invention is to provide a multi-label causal feature selection method and system in view of the defects of the prior art. The multi-label causal feature selection method of the present invention uses a univariate approximate Markov blanket to find approximate Markov blanket features for each label based on the preprocessed multi-label gene expression data set; deletes other labels contained in the approximate Markov blanket of the label, and restores equivalent features to obtain the Markov blanket features of the label; uses a symmetry test method to ensure that the selected features belong to the Markov blanket of the label; merges the Markov blanket features of all labels to obtain a subset of gene expression data features, and realizes feature selection of gene expression data. The present invention adopts a feature selection method based on causal structure and univariate approximate Markov blanket, considers three relationships between labels and features, between features and features, and between labels and labels, adds causal relationships on the basis of univariate approximate Markov blanket, so that a complete Markov blanket can be found; and restores features that are mistakenly deleted due to equivalent information of labels, thereby solving the problem of low accuracy of existing gene expression data feature selection methods from the root.

[0006] In order to achieve the above purpose, the present invention adopts the following technical solutions:

[0007] The present invention proposes a multi-label causal feature selection method, which specifically includes:

[0008] S1. Acquire a multi-label dataset and preprocess the multi-label dataset, wherein the multi-label dataset is a multi-label gene expression dataset;

[0009] S2. Based on the preprocessed multi-label gene expression dataset, an approximate Markov blanket is found for each label using a univariate approximate Markov blanket. The specific steps include:

[0010] S21, determining the threshold by calculating the uncertainty between each label and feature; specifically, calculating the symmetric uncertainty between each label and feature, and taking the average of the first t% of all symmetric uncertainties as the threshold α, where t is the set threshold;

[0011] S22, determine whether the feature is related to the label. If the symmetric uncertainty between the label and the feature is greater than α, it means that the label and the feature are related. If it is less than α, it means that the label and the feature are not related, and the feature is deleted from the feature set.

[0012] S23, using a univariate approximate Markov blanket to delete the false MB features in the feature set obtained in step S22;

[0013] S24, using a symmetry check, deleting the false MB features in the feature set obtained in step S23, and obtaining an approximate Markov blanket for each label;

[0014] The unary approximate Markov blanket is specifically:

[0015] In a Bayesian network, given a label T i , two features F i 、F j , if the following formula is satisfied:

[0016]

[0017] SU(F i ; T i )>SU(F j ; T i )&&SU(F i ; F j )>SU(F j ; T i ) (2)

[0018] I(T i ; F j |F i )<I(T i ; F j ) (3)

[0019] Then for F j In terms of i T i The one-dimensional approximation Markov blanket of ;

[0020] Among them, SU represents the symmetric uncertainty, which is normalized by mutual information and represents the correlation between two variables, I(T i ; F i ) represents the size of the mutual information between two variables, and I(T i ; F j |F i ) represents a given F i After the variable, T i With F j The size of the conditional mutual information between F i For T i With F j The degree of influence of the correlation between them; if for F j In terms of i T i The one-dimensional approximate Markov blanket feature, at this time F j It must not be T i MB features, so delete F j ;

[0021] S3, deleting other labels contained in the approximate Markov blanket of labels and restoring equivalent features to obtain the Markov blanket features of labels;

[0022] S4, using a fast symmetry check method to ensure that the features recovered in step S3 belong to the Markov blanket of the label, the features that pass the fast symmetry check method will be retained, and the features that fail will be deleted;

[0023] S5. The Markov blanket features of all labels are merged to obtain a feature subset of gene expression data, thereby realizing feature selection of gene expression data.

[0024] Furthermore, in step S1, the preprocessing includes: data cleaning and discretization encoding operations on different data.

[0025] Furthermore, step S3 specifically includes: deleting other labels contained in the approximate Markov blanket of labels, and restoring features mistakenly deleted due to equivalent information of labels, so as to obtain Markov blanket features of labels.

[0026] Furthermore, step S4 is specifically as follows: if a feature X is part of the Markov blanket of a label T, then T should also be part of the Markov blanket of X; the symmetry ensures the consistency and mutual inclusion between the feature and the label.

[0027] Furthermore, after step S5, the gene classification model is trained based on the feature subset of the gene expression data obtained by the multi-label causal feature selection method to obtain a trained gene classification model; the gene data to be classified is obtained, and based on the feature subset of the gene data to be classified, the gene data to be classified is classified using the trained gene classification model to obtain the label of the gene data to be classified.

[0028] The present invention also proposes a multi-label causal feature selection system, characterized in that the feature selection system executes the multi-label causal feature selection method, including: a data acquisition and preprocessing module, an approximate Markov blanket feature acquisition module, a Markov blanket feature determination module, a feature verification module, and a feature merging module;

[0029] A data acquisition and preprocessing module, which acquires a multi-label data set and preprocesses the multi-label data set, wherein the multi-label data set is a multi-label gene expression data set;

[0030] The approximate Markov blanket feature acquisition module uses a univariate approximate Markov blanket to find the approximate Markov blanket feature for each label based on the preprocessed multi-label gene expression dataset;

[0031] The Markov blanket feature determination module deletes other labels contained in the approximate Markov blanket of the label and restores the equivalent features to obtain the Markov blanket feature of the label;

[0032] The feature verification module uses a fast symmetry check method to ensure that the recovered features belong to the Markov blanket of the label. Features that pass the fast symmetry check method will be retained, and features that fail will be deleted;

[0033] The feature merging module merges the Markov blanket features of all labels to obtain the feature subset of gene expression data and realizes feature selection of gene expression data.

[0034] The present invention further provides a computer device, comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above method when executing the computer program.

[0035] Compared with the prior art, it has the following beneficial effects:

[0036] The multi-label causal feature selection method of the present invention is based on the preprocessed multi-label gene expression data set, and uses a univariate approximate Markov blanket to find approximate Markov blanket features for each label; delete other labels contained in the approximate Markov blanket of the label, and restore equivalent features to obtain the Markov blanket features of the label; use a symmetry test method to ensure that the selected features belong to the Markov blanket of the label; merge the Markov blanket features of all labels to obtain a gene expression data feature subset, and realize feature selection of gene expression data. The present invention adopts a feature selection method based on causal structure and univariate approximate Markov blanket, considers three relationships between labels and features, between features and features, and between labels and labels, adds causal relationships on the basis of univariate approximate Markov blanket, so that a complete Markov blanket can be found; and restores features that are mistakenly deleted due to equivalent information of labels, thereby improving the accuracy of feature selection of gene expression data, thereby improving the accuracy of gene classification models. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0038] Figure 1 Schematic diagram of a multi-label causal feature selection method provided in an embodiment of the present invention.

[0039] Figure 2 A schematic diagram of the structure of an existing Markov blanket provided in an embodiment of the present invention.

[0040] Figure 3 A schematic diagram of equivalent information provided by an embodiment of the present invention.

[0041] Figure 4 Schematic diagram of a multi-label causal feature selection system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0044] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0045] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0046] It should be further understood that the term "and / or" used in the present description and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0047] The present invention proposes a multi-label causal feature selection method. Figure 1 As shown, the multi-label causal feature selection method includes the following steps S1 to S5.

[0048] S1. Obtain a multi-label dataset, and preprocess the multi-label dataset, wherein the multi-label dataset is a multi-label gene expression dataset.

[0049] In step S1, the preprocessing includes: data cleaning and discretization encoding operations on different data.

[0050] In a specific embodiment, different data sources come from different devices and collection platforms. Device failure or insufficient device performance may lead to data loss, recording errors, duplicate data, etc. Therefore, the data cleaning includes the correction of erroneous data and abnormal data, the filling of missing road section data, and the removal of redundant data.

[0051] Since we use mutual information to measure the causality between labels, between features, and between features and labels when performing feature selection later, the use of mutual information requires discretization of the data set first, so discretization encoding operations are performed on different data.

[0052] In one embodiment, the multi-label data set is a VirusGO virus data set, which is used to predict the subcellular location of proteins based on the sequence of proteins. It includes sequences of 207 viral species. GO (gene ontology) features and PseAAC (including 20 amino acids, 20 pseudoamino acids and 400 dipeptide components) are provided. There are 6 subcellular locations (virus capsid, host cell membrane, host endoplasmic reticulum, host cytoplasm, host nucleus and secretion).

[0053] In addition, in another embodiment, the multi-label data set can be the following data set: GpositiveGO data set, which is used to predict the subcellular location of a protein based on its sequence. It contains sequences of 519 Gram-positive species. GO (Gene Ontology) features and PseAAC (including 20 amino acids, 20 pseudoamino acids and 400 dipeptide components) are provided at the same time; there are 4 subcellular locations (cell membrane, cell wall, cytoplasm and extracellular).

[0054] The characteristics of the relevant datasets are listed in the table below, including the number of instances (m), the number of features (d), the number of labels (q), cardinality (Card), density (Dens), diversity (Div), and the average imbalance ratio (avgIR) for each label. Cardinality measures the average number of labels associated with each instance, and density is defined as cardinality divided by the number of labels. Diversity represents the percentage of label sets present in the dataset divided by the number of possible label sets. avgIR measures the average imbalance of all labels. The larger the avgIR, the more imbalanced the dataset is.

[0055]

[0056] S2, based on the preprocessed multi-label gene expression dataset, using a unary approximate Markov blanket to find an approximate Markov blanket for each label;

[0057] This step is mainly to perform preliminary feature selection on the features preprocessed in the previous step. Assume that D = (F, T) represents the d-dimensional instance space, where F = {F1, F2, ..., F m} represents a feature space with m possible features, and T = {T1, T2, …, T q} represents a label space with q possible labels. Our goal is to select fewer features to make the classifier perform better.

[0058] Different from single-label feature selection, multi-label feature selection needs to consider the relationship between labels, between labels and features, and between features. We use the idea of ​​first finding the causal features of each label in the multi-label data and then merging them. Therefore, in order to take into account the various relationships between labels and features, when we first find the causal features of each label in the multi-label data, we will treat the remaining labels as feature variables to determine whether these labels are related to our labels. These connections may lead to some feature misjudgments.

[0059] When looking for the causal features of a single label, we look for the Markov blanket of the label. The structure of the existing Markov blanket is as follows: Figure 2 As shown. Figure 2 As shown in the figure, the Markov blanket refers to the set of parent nodes (A, B), child nodes (C, D), and spouse nodes E around a variable such as X. This set represents all the information that can represent X. Therefore, we need to find the causal features of label X, which is equivalent to finding the Markov blanket features of X.

[0060] However, the traditional Markov blanket method will miss some of the multi-label data sets, so this application proposes to use a univariate approximate Markov blanket to first find the approximate Markov blanket features for each label. Through the univariate approximate Markov blanket, we can find a small number of Markov blanket features we need, thereby improving the accuracy of feature selection and further improving the accuracy and efficiency of the classification model.

[0061] The specific steps of step S2 include:

[0062] S21, determine the threshold by calculating the uncertainty between each label and feature; specifically, calculate the symmetric uncertainty between each label and feature, and take the average of the first t% of all symmetric uncertainties as the threshold α, where t is the set threshold, and the value is [0.01, 0.05];

[0063] Among them, this threshold α represents the correlation between the label and the feature.

[0064] S22, determine whether the feature is related to the label. If the symmetric uncertainty between the label and the feature is greater than α, it means that the label and the feature are related. If it is less than α, it means that the label and the feature are not related, and the feature is deleted from the feature set.

[0065] Specifically, it is determined whether the feature is related to the label. Specifically, considering the influence between multiple labels, when looking for relevant features of a label, the remaining labels are first added to the feature set and regarded as ordinary features; if the symmetric uncertainty between the label and the feature is greater than α, it means that there is a correlation between the label and the feature. If it is less than α, it means that there is no correlation between the label and the feature, and this feature is deleted from the feature set.

[0066] S23, using a univariate approximate Markov blanket to delete the false MB features in the feature set obtained in step S22;

[0067] The unary approximate Markov blanket is used to delete the false MB features, so that the remaining features and labels form the Markov blanket structure we need.

[0068] S24. Using a symmetry check, delete the false MB features in the feature set obtained in step S23 to obtain an approximate Markov blanket for each label.

[0069] The unary approximate Markov blanket is specifically:

[0070] In a Bayesian network, given a label T i , two features F i 、F j , if the following formula is satisfied:

[0071]

[0072] SU(F i ; T i )>SU(F j ; T i )&&SU(F i ; F j )>SU(F j ; T i ) (2)

[0073] I(T i ; F j |F i )<I(T i ; F j ) (3)

[0074] Then for F j In terms of i T iA one-dimensional approximation to a Markov blanket.

[0075] Among them, SU represents the symmetric uncertainty, which is normalized by mutual information and represents the correlation between two variables. i ; F j |F i ) represents the size of the mutual information between two variables, and I(T i ; F j |F i ) represents a given F i After the variable, T i With F j The size of the conditional mutual information between F i For T i With F j The degree of influence of the correlation between them; if for F j In terms of i T i The one-dimensional approximate Markov blanket feature, at this time F j It must not be T i MB features, so delete F j .

[0076] In a faithful Bayesian network, if there exists an F i Features for F j For T i The one-dimensional approximate Markov blanket feature of F j It must not be T i MB features, so delete F j ; We can illustrate this by assuming that for F j In terms of i T i The unary approximate Markov blanket feature and F j Belong to T i MB characteristics, then our formula will be contradictory.

[0077] Case 1: F j T i The parent subset features in the MB features. i 、F j , T i There are three situations in which the relative positions of the three are: 1. i By F j With T i Connected T i -F j -F i , 2, F i By T i With F j Connected Fj -T i -F i , 3, F i -XT i -F j ,F j Through the remaining variables and T i We can find that in cases 1 and 3, due to T i With F i Directly connected, T i With F i Indirectly connected, so T i With F j The correlation between i With F i The correlation between SU(F i ,T i )>SU(F j ,T i ), Similarly, in the second case, due to T i With F j The correlation between them is greater than F i With F j The correlation between them violates the formula SU(F i ,F j )>SU(F j ,T i ). Therefore, for F j In terms of i T i When the one-dimensional approximate Markov blanket feature of j Impossible for T i The parent-child set features in the MB features.

[0078] Case 2: F j T i When the spouse set feature is in the MB feature, according to the definition of the spouse set feature, we can always find an F i Can make F j Satisfy F j →F i ←T i The structure has a characteristic, that is, when F is given i When F j With T i The correlation between them becomes larger, thus violating our formula 3. So F j It can't be us T i Mate set features in MB features.

[0079] In summary, for F j In terms of F iT i The one-dimensional approximate Markov blanket feature of F j For non-MB features, they will be deleted. At the same time, for real gene expression multi-label datasets, the assumption of fidelity may be violated, and F j →F i →T i 、F j →T i This special structure, in which F j Still possible for T i Therefore, we set up a formula 1, that is, only when F i With F j When the correlation between them is greater than the average threshold δ, it means that F i With F j If the redundancy between them is too large, we will perform F j Deletion decision.

[0080] S3. Delete other labels contained in the approximate Markov blanket of labels and restore equivalent features to obtain the Markov blanket features of labels.

[0081] Furthermore, step S3 specifically includes: deleting other labels contained in the approximate Markov blanket of labels, and restoring features mistakenly deleted due to equivalent information of labels, so as to obtain Markov blanket features of labels.

[0082] Step S3 is mainly to delete the Markov blanket of the label in step S2 that contains other labels. At the same time, because real data often violates the assumption of fidelity, equivalent information may appear, which may lead to the mistaken deletion of some true Markov blanket features. Figure 3 In the case of equivalent information shown, T j With {F i ,F j} all contain T i The same information, that is, SU(T i , T j )=SU(T i ,{F i ,F j}), but F i and F j It will be deleted by mistake due to the algorithm in the first stage, and finally when we delete the label T i The labels T in the selected feature set j When F i and F jThis application deletes other labels contained in the feature set of the label, and at the same time restores the features that were mistakenly deleted in the equivalent situation caused by the addition of the label during the deletion process, so that our multi-label feature selection process can obtain the correct Markov blanket feature set.

[0083] S4. Use a fast symmetry check method to ensure that the features recovered in step S3 belong to the Markov blanket of the label. Features that pass the fast symmetry check method will be retained, and features that fail will be deleted.

[0084] Furthermore, step S4 is specifically as follows: if a feature X is part of the Markov blanket of a label T, then T should also be part of the Markov blanket of X; the symmetry ensures the consistency and mutual inclusion between the feature and the label.

[0085] It can be expressed by the following formula: This symmetry ensures the mutual influence between features and labels. Therefore, after recovering the features, we use a symmetry test on the labels and their Markov blanket sets to ensure that the features we select are our true Markov blanket features.

[0086] S5. The Markov blanket features of all labels are merged to obtain a feature subset of gene expression data, thereby realizing feature selection of gene expression data.

[0087] After checking the Markov blanket features of each label, feature selection is not really completed. It is also necessary to merge the features of all labels. The merged features are the features retained by our method.

[0088] The present invention processes multi-label data, maintains the uniformity of data caliber, encodes using a suitable discrete coding method to form different types of features, and then uses a univariate approximate Markov blanket-based method to perform preliminary selection for each label, selects the features and other labels of the approximate Markov blanket belonging to the label, and then screens the features and labels in the approximate Markov blanket, deletes other labels contained in the approximate Markov blanket of the label, and restores the features that were mistakenly deleted due to the equivalent information of the label in this process, and then performs a symmetry check on the features selected for each label to ensure accuracy, and then merges them to obtain the features we need.

[0089] The present invention specifically targets multi-label datasets and implements a feature selection method CFAB based on causal structure and univariate approximate Markov blanket for multi-label feature selection. The method takes into account three relationships between labels and features, between features and between labels, and between labels. It adds causal relationships on the basis of univariate approximate Markov blanket, so that a complete Markov blanket can be found. At the same time, in response to the situation where equivalent information of labels appearing in multi-label datasets leads to mistaken deletion of features, the present invention proposes a method to delete other labels contained in the approximate Markov blanket of labels and restore equivalent features to solve the problem. Compared with traditional multi-label feature selection methods, the multi-label causal feature selection method of the present invention has higher accuracy and is more explanatory.

[0090] After step S5, the gene classification model is trained based on the feature subset of gene expression data obtained by the multi-label causal feature selection method CFAB to obtain a trained gene classification model; the gene data to be classified is obtained, and based on the feature subset of the gene data to be classified, the gene data to be classified is classified using the trained gene classification model to obtain the label of the gene data to be classified.

[0091] Through the multi-label causal feature selection method CFAB, the feature dimension is reduced and the accuracy of feature selection and classification is improved. Since our invention does not rely on a specific classifier, we randomly selected the MLKNN classifier commonly used in multi-label tasks as the gene classification model.

[0092] Specifically, during the model training process, six evaluation indicators were selected: Hamming Loss, Average Precision, Coverage, Ranking Loss, SubsetAccuracy, and F-Micro.

[0093] It was compared with 6 existing methods; at the same time, we ranked each of its indicators, as shown in the following table; from the data in the following table, it can be seen that the method CFAB proposed in the present invention has better feature selection effect on two gene expression data sets.

[0094]

[0095]

[0096] Figure 4 is a multi-label causal feature selection system provided by an embodiment of the present invention. Figure 4 As shown, the multi-label causal feature selection system includes a data acquisition and preprocessing module, an approximate Markov blanket feature acquisition module, a Markov blanket feature determination module, a feature verification module, and a feature merging module;

[0097] A data acquisition and preprocessing module, which acquires a multi-label data set and preprocesses the multi-label data set, wherein the multi-label data set is a multi-label gene expression data set;

[0098] The approximate Markov blanket feature acquisition module uses a univariate approximate Markov blanket to find the approximate Markov blanket feature for each label based on the preprocessed multi-label gene expression dataset;

[0099] The Markov blanket feature determination module deletes other labels contained in the approximate Markov blanket of the label and restores the equivalent features to obtain the Markov blanket feature of the label;

[0100] The feature verification module uses a fast symmetry check method to ensure that the recovered features belong to the Markov blanket of the label. Features that pass the fast symmetry check method will be retained, and features that fail will be deleted;

[0101] The feature merging module merges the Markov blanket features of all labels to obtain the feature subset of gene expression data and realizes feature selection of gene expression data.

[0102] The multi-label causal feature selection system can be implemented in the form of a computer program, which can be run on a computer device.

[0103] The computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0104] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute a multi-label causal feature selection method.

[0105] The processor is used to provide computing and control capabilities to support the operation of the entire computer equipment.

[0106] The internal memory provides an environment for the operation of a computer program in a non-volatile storage medium. When the computer program is executed by a processor, the processor can execute a multi-label causal feature selection method.

[0107] The network interface is used to communicate with other devices over the network. Those skilled in the art will appreciate that the above-mentioned computer device structure is only a partial structure related to the present application solution, and does not constitute a limitation on the computer device to which the present application solution 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.

[0108] The processor is used to run a computer program stored in the memory, which implements the multi-label causal feature selection method described in Example 1.

[0109] It should be understood that in the embodiments of the present application, the processor may be a central processing unit (CPU), and the processor may also be 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. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0110] It can be understood by those skilled in the art that all or part of the processes in the method for implementing the above embodiment can be completed by instructing the relevant hardware through a computer program. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiment of the above method.

[0111] The present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when the computer program is executed by a processor, the processor executes a multi-label causal feature selection method described in Embodiment 1.

[0112] The storage medium may be a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk, or any other computer-readable storage medium that can store program codes.

[0113] 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, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. 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 the present invention.

[0114] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of each unit is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0115] The steps in the method of the embodiment of the present invention can be adjusted in order, combined and deleted according to actual needs. The units in the device of the embodiment of the present invention can be combined, divided and deleted according to actual needs. In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0116] 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 storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, terminal, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0117] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A multi-label causal feature selection method, characterized in that: Includes steps: S1. Acquire a multi-label dataset and preprocess the multi-label dataset, wherein the multi-label dataset is a multi-label gene expression dataset; S2. Based on the preprocessed multi-label gene expression dataset, an approximate Markov blanket is found for each label using a univariate approximate Markov blanket. The specific steps include: S21, determining the threshold by calculating the uncertainty between each label and feature; specifically, calculating the symmetric uncertainty between each label and feature, and taking the average of the first t% of all symmetric uncertainties as the threshold α, where t is the set threshold; S22, determine whether the feature is related to the label. If the symmetric uncertainty between the label and the feature is greater than α, it means that the label and the feature are related. If it is less than α, it means that the label and the feature are not related, and the feature is deleted from the feature set. S23, using a univariate approximate Markov blanket to delete the false MB features in the feature set obtained in step S22; S24, using a symmetry check, deleting the false MB features in the feature set obtained in step S23, and obtaining an approximate Markov blanket for each label; The unary approximate Markov blanket is specifically: In a Bayesian network, given a label T i , two features F i 、F j , if the following formula is satisfied: SU(F i ;T i )>SU(F j ;T i )&&SU(F i ;F j )>SU(F j ;T i ) (2) I(T i ;F j |F i )<I(T i ;F j ) (3) Then for F j In terms of i T i The one-dimensional approximation Markov blanket of ; Among them, SU represents the symmetric uncertainty, which is normalized by mutual information and represents the correlation between two variables. i ; F i ) represents the size of the mutual information between two variables, and I(T i ; F j |F i ) represents a given F i After the variable, T i With F j The size of the conditional mutual information between F i For T i With F j The degree of influence of the correlation between them; if for F j In terms of i T i The one-dimensional approximate Markov blanket feature, at this time F j It must not be T i MB features, so delete F j ; S3, deleting other labels contained in the approximate Markov blanket of labels and restoring equivalent features to obtain the Markov blanket features of labels; S4, using a fast symmetry check method to ensure that the features recovered in step S3 belong to the Markov blanket of the label, the features that pass the fast symmetry check method will be retained, and the features that fail will be deleted; S5. The Markov blanket features of all labels are merged to obtain a feature subset of gene expression data, thereby realizing feature selection of gene expression data.

2. The method according to claim 1, characterized in that In step S1, the preprocessing includes: data cleaning and discretization encoding operations on different data.

3. The method according to claim 1, characterized in that Step S3 specifically includes: deleting other labels contained in the approximate Markov blanket of labels, and restoring features that are mistakenly deleted due to equivalent information of labels, so as to obtain the Markov blanket features of labels.

4. The method according to claim 1, characterized in that Step S4 is specifically: if a feature X is part of the Markov blanket of a label T, then T should also be part of the Markov blanket of X; the symmetry ensures the consistency and mutual inclusion between the feature and the label.

5. The method according to claim 1, characterized in that After step S5, the gene classification model is trained based on the gene expression data feature subset obtained by the multi-label causal feature selection method to obtain a trained gene classification model; Obtain gene data to be classified, and classify the gene data to be classified using a trained gene classification model based on the feature subset of the gene data to be classified to obtain a label for the gene data to be classified.

6. A multi-label causal feature selection system, characterized in that: The feature selection system executes the multi-label causal feature selection method according to claim 1, comprising: a data acquisition and preprocessing module, an approximate Markov blanket feature acquisition module, a Markov blanket feature determination module, a feature verification module, and a feature merging module; A data acquisition and preprocessing module, which acquires a multi-label data set and preprocesses the multi-label data set, wherein the multi-label data set is a multi-label gene expression data set; The approximate Markov blanket feature acquisition module uses a univariate approximate Markov blanket to find the approximate Markov blanket feature for each label based on the preprocessed multi-label gene expression dataset; The Markov blanket feature determination module deletes other labels contained in the approximate Markov blanket of the label and restores the equivalent features to obtain the Markov blanket feature of the label; The feature verification module uses a fast symmetry check method to ensure that the recovered features belong to the Markov blanket of the label. Features that pass the fast symmetry check method will be retained, and features that fail will be deleted; The feature merging module merges the Markov blanket features of all labels to obtain the feature subset of gene expression data and realizes feature selection of gene expression data.

Images