Bronchial asthma phenotype identification index screening method based on exosomics

Abstract

The invention discloses a bronchial asthma phenotype identification index screening method based on exosomics, which comprises the following steps: acquiring bronchial asthma medical record data to construct a medical record data subset, and extracting multi-dimensional information to obtain diagnosis characteristics; generating an original feature set by using common diagnostic features of different bronchial asthma phenotypes; the method comprises the following steps: constructing a knowledge graph according to exosome omics related knowledge and an action mechanism of exosomes from different cell sources on bronchial asthma, screening exosome features, and expanding an original feature set; a feature selection task based on multi-objective optimization is constructed, exosome feature combinations of different bronchial asthma are obtained on the basis of the common diagnostic features, and exosome identification indexes of bronchial asthma phenotypes are determined after the identification accuracy of the exosome feature combinations and the identification accuracy of the original common diagnostic features are compared and verified. According to the invention, exosomic indexes highly related to bronchial asthma diagnosis are screened, biomarkers are obtained for predicting allergy and asthma conditions, and the identification accuracy of bronchial asthma phenotypes is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of feature selection, and more specifically, to a method for screening bronchial asthma phenotype identification indicators based on exosome omics. Background Art

[0002] Asthma is a complex, heterogeneous disease with significant heterogeneity in terms of pathogenesis, clinical features, and response to treatment. Several large multicenter studies have used different clustering techniques to identify asthma phenotypes. Although these studies differ in the disease characteristics used in each cluster, the calculation methods, and the number of clusters studied, the results consistently indicate that there are approximately four major phenotypes of asthma: early-onset mild allergic bronchial asthma, early-onset moderate-to-severe allergic remodeling bronchial asthma, late-onset non-eosinophilic non-allergic bronchial asthma, and late-onset non-allergic eosinophilic bronchial asthma. Phenotypes are observable characteristics based on genetic and environmental relationships that depend on clinical, biochemical, functional, and pharmacological characteristics. Phenotypes may change over time, however, they represent the first criterion for distinguishing patients and are potential predictors of treatment response.

[0003] Exosomes are nanoscale vesicles that are shed from cell membranes or secreted by cells. They have a phospholipid bilayer structure, can mediate intercellular communication, have multiple biological functions, and are involved in inflammation, immune response, angiogenesis, cell death, antigen presentation, neurodegenerative diseases, tumor proliferation and other processes. Studies have found that exosomes are involved in the development of asthma induced by house dust mites. After asthmatic mice were treated with exosome inhibitors, the levels of eosinophils, IgE and inflammatory cytokines were reduced, indicating that inhibiting exosome secretion can alleviate asthma characteristics. At present, the cause and pathogenesis of asthma have not been fully elucidated, and its clinical treatment has certain limitations. It is particularly important to study the pathogenesis of asthma and establish an effective phenotypic identification method. Therefore, how to screen exosome genomic features based on feature selection to accurately determine the phenotype of bronchial asthma is an urgent problem to be solved. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a method for screening bronchial asthma phenotype identification indicators based on exosome omics, so as to solve the problems in the prior art that the bronchial asthma phenotype identification indicators are single and cannot timely capture the changes in airway inflammation phenotype, and provide new ideas for the bronchial asthma phenotype identification.

[0005] The present invention provides a method for screening bronchial asthma phenotype identification indicators based on exosome omics, comprising:

[0006] Obtain a number of bronchial asthma medical record data, construct medical record data subsets based on different diagnostic phenotype categories, extract multidimensional information from different medical record data subsets, and obtain diagnostic features of different bronchial asthma phenotypes;

[0007] Acquire the common diagnostic features and common diagnostic feature distribution of different bronchial asthma phenotypes, and generate an original feature set based on the common diagnostic features;

[0008] Retrieve exosome-related knowledge using big data methods, construct a knowledge graph based on the mechanism of action of exosomes from different cell sources on bronchial asthma, use the knowledge graph to screen exosome features that meet preset requirements, and expand the original feature set;

[0009] A feature selection task based on multi-objective optimization was constructed in the expanded original feature set, and the exosome feature combinations of different bronchial asthma phenotypes were obtained based on the common diagnostic features. The identification accuracy was compared and verified according to the distribution of the common diagnostic features, and the exosome identification indicators of the bronchial asthma phenotype were determined based on the verification results.

[0010] In this scheme, medical record data subsets are constructed based on different diagnostic phenotype categories, and multidimensional information is extracted from different medical record data subsets, specifically:

[0011] Retrieving medical record data using bronchial asthma as a keyword, obtaining bronchial asthma medical record data that meets the search tag constraint, extracting diagnosis information, and obtaining bronchial asthma phenotypes based on the diagnosis information;

[0012] Aggregating the corresponding bronchial asthma medical record data using the bronchial asthma phenotype, constructing medical record data subsets of different bronchial asthma phenotypes, extracting detection indicators in different medical record data subsets, and screening abnormal indicators;

[0013] Determine the diagnosis basis according to the abnormal index combined with the clinical symptoms in the medical record data, obtain low-dimensional features from the diagnosis basis using single-channel convolution, and obtain high-dimensional features from the low-dimensional features based on convolution calculation, normalization processing and activation operation of the main branch;

[0014] At the same time, the low-dimensional features are locally integrated based on the side branches using average pooling processing, the outputs of the two branches are fused to obtain multi-dimensional information, and diagnostic features of different bronchial asthma phenotypes are generated based on the multi-dimensional information.

[0015] In this scheme, the common diagnostic features and common diagnostic feature distribution of different bronchial asthma phenotypes are obtained, and the original feature set is generated based on the common diagnostic features, specifically:

[0016] Acquire diagnostic feature sets of different bronchial asthma phenotypes, perform intersection processing on each diagnostic feature set, extract common diagnostic features to generate an original feature set, and extract corresponding examination data from each medical record data subset according to the common diagnostic features in the original feature set;

[0017] The examination data corresponding to the common diagnostic features are mapped to a low-dimensional space, the exponential distribution of the examination data in the low-dimensional space is obtained to generate a common diagnostic feature distribution, and the identification accuracy is obtained according to the common diagnostic feature distribution of different bronchial asthma phenotypes.

[0018] In this scheme, we retrieved exosome-related knowledge based on big data methods, and constructed a knowledge graph based on the mechanism of action of exosomes from different cell sources on bronchial asthma, specifically:

[0019] Through big data methods, we obtain professional corpus of exosome omics and big data information of exosome omics and bronchial asthma based on search engines. After structured processing, we use the encoder model to obtain the feature vector of exosome omics-related knowledge, and capture deep-level feature information to obtain entities.

[0020] According to the big data information of exosomes and bronchial asthma, semantic features are obtained by embedding, the mechanism of action of exosomes from different cell sources on bronchial asthma is read through the semantic features, the relationship between entities is extracted based on the mechanism of action, and a relationship table is constructed;

[0021] The acquired entities and relationship tables are mapped in triple form to construct a knowledge graph in a graphical way.

[0022] In this scheme, the knowledge graph is used to screen exosome features whose importance meets the preset requirements, and the original feature set is expanded, specifically:

[0023] For each entity in the knowledge graph, obtain the number of entities directly connected to the entity, obtain the entity contribution according to the number of entities, construct an adjacency matrix according to the entity contribution, and represent the knowledge graph in the form of an adjacency matrix;

[0024] Use a graph neural network to perform representation learning on the knowledge graph, aggregate and propagate the features of entities in the knowledge graph through neighbor entities, construct entity embedding vectors, import the entity embedding vectors into the GRU unit to obtain the relationship between entities, and use the Softmax function to obtain the centrality of the entities;

[0025] The importance score of the entity is obtained according to the centrality and the initial score, and the entity is sorted by the importance score, and the exosome features that meet the preset requirements are selected. After reducing feature redundancy by calculating the correlation between the exosome features, the original feature set is expanded.

[0026] In this scheme, a feature selection task based on multi-objective optimization is constructed in the expanded original feature set, specifically:

[0027] Obtain the expanded original feature set, use feature importance and feature redundancy as evaluation indicators, and take maximum feature importance and minimum feature redundancy as optimization goals;

[0028] A bronchial asthma phenotype classification model is constructed based on a convolutional autoencoder and a GUR unit, a data set is constructed based on the expanded original feature set and the corresponding inspection data, the data set is divided into several subsets, each subset is used as a validation set, and the remaining subsets are used as training sample sets to train the bronchial asthma phenotype classification model to obtain several models;

[0029] The convolutional autoencoder part is composed of stacked convolutional layers, pooling layers and batch normalization layers. The convolutional autoencoder parameters are initialized and trained in an unsupervised training manner to obtain the recognition classification features of the output of the encoder part;

[0030] The identification and classification features are introduced into the GRU unit to obtain the time correlation of the features, and the bronchial asthma phenotype classification is realized by using the fully connected layer through deep learning of the temporal features, and the average classification accuracy of several bronchial asthma phenotype classification models is obtained as an importance evaluation indicator;

[0031] The mRMR algorithm was reconstructed using the Pearson correlation coefficient to calculate the correlation between the selected features and the correlation between the features and the bronchial asthma phenotype, and to obtain the redundancy evaluation index of the feature combination.

[0032] In this scheme, the exosome feature combinations of different bronchial asthma phenotypes are obtained based on the common diagnostic features, specifically:

[0033] The optimized Eagle algorithm is used to obtain the optimal exosome feature combination according to the optimization target, the Eagle algorithm parameters are initialized, the expanded original feature set is encoded, the common diagnostic features are used as feature benchmarks, and the initial feature combination is randomly generated on the feature benchmarks as the Eagle individuals;

[0034] Calculate the fitness of the current position of the initialized Skyhawk population, update the global optimal fitness value and optimal position, preset the number of iterations threshold, and expand the search for global optimization when the number of iterations does not exceed the preset number of iterations threshold, otherwise, narrow the search for local search;

[0035] Adaptive control factors are introduced to adjust the flight speed of the individual eagles in global optimization and local search, calculate the fitness value of the new position of the individual eagle, obtain the feature combination with the highest fitness value, use differential calculation to strengthen the recombination of the feature combination, and record the global optimal fitness value and optimal position;

[0036] When the objective function of the feature combination converges, an optimal feature combination is obtained, and the common diagnostic features are removed from the optimal feature combination to obtain the exosome feature combination.

[0037] In this scheme, the identification accuracy comparison and verification are performed based on the common diagnostic feature distribution, and the exosome identification index of bronchial asthma phenotype is determined by the verification results, specifically:

[0038] Based on the inspection data of the exosome feature combination obtained from the medical record data of bronchial asthma, the inspection data of the exosome feature combination is mapped to a low-dimensional space to obtain the exosome feature distribution, and the corresponding identification accuracy is obtained according to the exosome feature distribution of different bronchial asthma phenotypes;

[0039] Obtain the corresponding identification accuracy of the common diagnostic feature distribution of different bronchial asthma phenotypes, and compare the deviation between the corresponding identification accuracy of the exosome feature distribution and the corresponding identification accuracy of the common diagnostic feature distribution;

[0040] If the deviation is greater than the preset deviation threshold, the exosome feature of the current bronchial asthma phenotype is reacquired; if the deviation is not greater than the preset deviation threshold, the exosome feature combination is output.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The present invention can screen exosome omics indicators that are highly correlated with the diagnosis of bronchial asthma, obtain biomarkers for predicting allergic and asthmatic conditions, and provide a reliable auxiliary reference for the diagnosis and treatment of bronchial asthma;

[0043] The present invention improves the accuracy of bronchial asthma phenotype identification and the efficiency of diagnosis and identification through the screened exosome identification indicators, provides a simple and easy means to determine the airway inflammation phenotype, and timely captures the changes in the airway inflammation phenotype, which is of great significance for the treatment and prognosis of asthma. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments or exemplary embodiments of the present invention, the drawings required for use in the embodiments or exemplary descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained according to the drawings without paying creative work.

[0045] Figure 1 A flow chart showing the method for screening bronchial asthma phenotype identification indicators based on exosome omics of the present invention is shown;

[0046] Figure 2A flow chart showing the method of screening exosome features to expand the original feature set according to the present invention is shown;

[0047] Figure 3 A flow chart of obtaining exosome feature combinations of different bronchial asthma phenotypes according to the present invention is shown;

[0048] Figure 4 A block diagram of the bronchial asthma phenotype identification index screening system based on exosome omics of the present invention is shown.

[0049] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0050] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0051] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0052] Figure 1 A flow chart showing the method for screening bronchial asthma phenotype identification indicators based on exosome omics of the present invention is shown;

[0053] like Figure 1 As shown, the first embodiment of the present invention provides a method for screening bronchial asthma phenotype identification indicators based on exosome omics, comprising:

[0054] S102, obtaining a number of bronchial asthma medical record data, constructing medical record data subsets based on different diagnostic phenotype categories, extracting multidimensional information from different medical record data subsets, and obtaining diagnostic features of different bronchial asthma phenotypes;

[0055] S104, obtaining common diagnostic features and common diagnostic feature distributions of different bronchial asthma phenotypes, and generating an original feature set based on the common diagnostic features;

[0056] S106, retrieving exosome-related knowledge according to a big data method, constructing a knowledge graph through the mechanism of action of exosomes from different cell sources on bronchial asthma, using the knowledge graph to screen exosome features whose importance meets preset requirements, and expanding the original feature set;

[0057] S108, constructing a feature selection task based on multi-objective optimization in the expanded original feature set, obtaining exosome feature combinations of different bronchial asthma phenotypes based on the common diagnostic features, and performing comparative verification of identification accuracy based on the distribution of the common diagnostic features, and determining the exosome identification index of the bronchial asthma phenotype through the verification results.

[0058] It should be noted that medical record databases from different sources are accessed as retrieval spaces, and medical record data are retrieved based on bronchial asthma as a keyword to obtain bronchial asthma medical record data that meets the retrieval tag constraints, and diagnostic information is extracted, and the bronchial asthma phenotype is obtained based on the diagnostic information; the corresponding bronchial asthma medical record data are aggregated using the bronchial asthma phenotype to construct medical record data subsets of different bronchial asthma phenotypes, and the detection indicators involved in different medical record data subsets are extracted, and indicators that do not meet the inspection standard range are selected according to the detection indicators and marked as abnormal indicators; the diagnostic basis is determined based on the abnormal indicators combined with clinical symptoms such as wheezing, coughing, and chest tightness in the medical record data, and the diagnostic basis is used to perform targeted filtering on abnormal indicator inspection data and symptom information from different sources using single-channel convolution to obtain low-dimensional features, and to ensure that there is no data omission when all abnormal indicator inspection data and symptom information are extracted for feature extraction. The main branch introduces the CNN structure, and obtains high-dimensional features from the low-dimensional features based on the convolution calculation, normalization processing and activation operation of the main branch; at the same time, the low-dimensional features are locally integrated based on the side branch using average pooling processing, and the outputs of the two branches are fused to obtain multi-dimensional information. The multi-dimensional features retain more features in the data, and the dual-branch feature extraction structure improves the network convergence speed. The diagnostic features of different bronchial asthma phenotypes are generated according to the multi-dimensional information, and the detection data features based on which the bronchial asthma phenotype diagnosis is based are characterized.

[0059] It should be noted that the diagnostic feature sets of different bronchial asthma phenotypes are obtained, the intersection of each diagnostic feature set is processed, the common diagnostic features are extracted to generate the original feature set, and the corresponding examination data are extracted from each medical record data subset according to the common diagnostic features in the original feature set; the examination data corresponding to the common diagnostic features are mapped to the low-dimensional space, and the exponential distribution of the examination data in the low-dimensional space is obtained to generate the common diagnostic feature distribution. The corresponding probability density function f(x) is: x represents the diagnostic feature, θ represents the mean of the identification accuracy, and the corresponding likelihood function L is obtained: y i represents the sample of bronchial asthma phenotype diagnosis results, r represents the number of errors in the sample of bronchial asthma phenotype diagnosis results, n represents the total number of samples, and the maximum likelihood method is used to obtain the identification accuracy according to the common diagnostic characteristics of different bronchial asthma phenotypes.

[0060] It should be noted that the professional corpus of exosome omics and big data information of exosome omics and bronchial asthma were obtained through big data methods based on search engines. After structured processing, the encoder model was used to obtain the feature vector of exosome omics-related knowledge, and the deep-level feature information was captured to obtain entities. Since the names of the same exosome omics entities may have different languages ​​or expressions in different literature data or scientific research materials, the different expressions of the professional corpus of exosome omics were clarified according to big data means, and equivalent concepts were determined according to different expressions of the same related terms. The professional corpus of exosome omics and big data information of exosome omics and bronchial asthma were unified based on equivalent concepts for related terms, and entity resolution was achieved through retrieval matching, and data conflict detection was performed after unification. According to the information of exosomes and bronchial asthma big data, semantic features are obtained through attention embedding and other methods. The mechanism of action of exosomes from different cell sources on bronchial asthma is read through the semantic features. The cell sources are roughly divided into exosomes from immune effector cells, exosomes from lung structural cells, etc. The relationship between entities is extracted based on the mechanism of action, and a relationship table is constructed. For example, there are 12 miRNAs upregulated in mice with allergic airway inflammation, especially miR-233-3p and miR-142a, which are the most abundant, and are upregulated in the sputum of severe asthma patients, indicating that their expression levels are related to the severity of the disease, that is, exosome miR-233-3p and miR-142a have a relationship with bronchial asthma that represents the severity. The acquired entities and the relationship table are mapped in the form of triples to construct a knowledge graph in a graphical way.

[0061] Figure 2 A flow chart of the present invention for screening exosome features to expand the original feature set is shown.

[0062] According to an embodiment of the present invention, the knowledge graph is used to screen exosome features whose importance meets preset requirements, and the original feature set is expanded, specifically:

[0063] S202, for each entity in the knowledge graph, obtaining the number of entities directly connected to the entity, obtaining the entity contribution according to the number of entities, constructing an adjacency matrix according to the entity contribution, and representing the knowledge graph in the form of an adjacency matrix;

[0064] S204, using a graph neural network to perform representation learning on the knowledge graph, constructing an entity embedding vector by aggregating and propagating features of entities in the knowledge graph through neighbor entities, importing the entity embedding vector into a GRU unit to obtain the relationship between entities, and using a Softmax function to obtain the centrality of the entity;

[0065] S206, obtaining the importance score of the entity according to the centrality and the initial score, sorting the entities by the importance score, selecting exosome features that meet the preset requirements, reducing feature redundancy by calculating the correlation between the exosome features, and expanding the original feature set.

[0066] It should be noted that the graph neural network aggregates and propagates the features of entities in the knowledge graph, generates entity embedding vectors, and learns the feature information of triples in the knowledge graph through the GRU unit to obtain the centrality of the entity in the knowledge graph to represent its importance. According to the characteristics of the knowledge graph, the more nodes a node is connected to, the richer the implicit information of the node may be. When calculating the importance score of the entity, the preference of the bronchial asthma phenotype classification for the entity is taken into account. The corresponding relationship of the entity is retrieved from the knowledge graph and the correlation between the relationship and the bronchial asthma phenotype classification is calculated. The initial score of the entity is obtained based on the correlation.

[0067] It should be noted that the expanded original feature set is obtained, and the feature importance and feature redundancy are used as evaluation indicators, with the maximum feature importance and minimum feature redundancy as optimization goals; a bronchial asthma phenotype classification model is constructed based on the convolutional autoencoder and the GRU unit, a data set is constructed based on the expanded original feature set and the corresponding inspection data, the data set is divided into several subsets, each subset is used as a validation set, and the remaining subsets are used as training sample sets to train the bronchial asthma phenotype classification model to obtain several models; the convolutional autoencoder part is composed of stacked convolutional layers, pooling layers and batch normalization layers, the convolutional autoencoder parameters are initialized, and unsupervised training is used for training to obtain the recognition and classification features of the output of the encoder part; the recognition and classification features are introduced into the GRU unit to obtain the time correlation of the features, and the bronchial asthma phenotype classification is realized by deep learning of the time series features and the fully connected layer. The average classification accuracy of several bronchial asthma phenotype classification models is obtained as an importance evaluation indicator, which is expressed as: Among them, n represents the number of classification models, F represents the classification model parameters, D represents the training sample set, and x i Represents training samples.

[0068] The mRMR algorithm was reconstructed using the Pearson correlation coefficient to calculate the correlation between the selected features and the correlation between the features and the bronchial asthma phenotype, and the redundancy evaluation index of the feature combination was obtained, which is expressed as: Among them, R(x i ; y) represents the i-th feature x i Pearson correlation coefficient between the phenotypic classification of bronchial asthma and y, R(x i ;x j ) represents the i-th feature x iWith the jth feature x j Pearson correlation coefficient between them, S represents the selected feature combination, |S| represents the number of features in the feature combination, Ω s Represents the original feature set after expansion.

[0069] Figure 3 A flow chart of the present invention for obtaining exosome feature combinations of different bronchial asthma phenotypes is shown.

[0070] According to an embodiment of the present invention, the exosome feature combinations of different bronchial asthma phenotypes are obtained based on the common diagnostic features, specifically:

[0071] S302, using the optimized Tianying algorithm to obtain the optimal exosome feature combination according to the optimization target, initializing the Tianying algorithm parameters, encoding the expanded original feature set, taking the common diagnostic features as feature benchmarks, and randomly generating an initial feature combination on the feature benchmarks as Tianying individuals;

[0072] S304, calculating the fitness of the current position of the initialized Skyhawk population, updating the global optimal fitness value and the optimal position, and presetting an iteration number threshold. When the iteration number does not exceed the preset iteration number threshold, the search is expanded to perform global optimization, otherwise, the search is narrowed to perform local search;

[0073] S306, introducing an adaptive control factor to adjust the flight speed of the individual eagle in the global optimization and local search, calculating the fitness value of the new position of the individual eagle, obtaining the feature combination with the highest fitness value, using differential calculation to strengthen the reorganization of the feature combination, and recording the global optimal fitness value and optimal position;

[0074] S308, when the objective function of the feature combination converges, an optimal feature combination is obtained, and the common diagnostic features are removed from the optimal feature combination to obtain an exosome feature combination.

[0075] It should be noted that the Eagle algorithm achieves rapid positioning and solving by simulating the behavior of the Eagle catching prey, and its search efficiency is high and the convergence speed is fast. Encoding the expanded original feature set can better express the feature combination. When randomly generating the initial feature combination, the exosome features from different cell sources are grouped, and the initial feature combination is generated based on the grouping, so that the initial feature combination has the local features of exosomes from different cell sources, which increases the diversity of the initial feature combination.

[0076] The Sky Eagle algorithm is improved by introducing an adaptive control factor. In the early stage of iteration, the Sky Eagle individual expands the search range at a higher flight speed to improve the global search capability. In the later stage of iteration, the Sky Eagle individual locks the prey at a lower flight speed to capture it, improving the later optimization performance. Its adaptive control factor λ(i) is expressed as: λ(i) = λc -i(λ c -λ z ) / T max ,λ c , z Distribution represents the range node of the control factor, i represents the number of iterations, T max Represents the total number of iterations. Associate the fitness with the objective function, establish a mapping relationship, use the mapping relationship to optimize the iterative convergence of the feature combination, and introduce the idea of ​​differential mutation in the Sky Eagle algorithm. Select the feature combination with a high fitness value, obtain the difference between other feature combinations and it, and evolve the feature combination based on the obtained difference to achieve information exchange between different feature combinations, so that the feature combination is optimized in the established direction.

[0077] It should be noted that the inspection data of the exosome feature combination obtained based on the bronchial asthma medical record data is mapped to a low-dimensional space to obtain the exosome feature distribution, and the corresponding identification accuracy is obtained according to the exosome feature distribution of different bronchial asthma phenotypes; based on the probability density function of the exponential distribution and the idea of ​​maximum likelihood, the maximum likelihood estimate is calculated to obtain the identification accuracy corresponding to the common diagnostic feature distribution of different bronchial asthma phenotypes, and the deviation between the identification accuracy corresponding to the exosome feature distribution and the identification accuracy corresponding to the common diagnostic feature distribution is compared; if the deviation is greater than the preset deviation threshold, the exosome feature of the current bronchial asthma phenotype is re-obtained; if the deviation is not greater than the preset deviation threshold, the exosome feature combination is output.

[0078] Figure 4 A block diagram of the bronchial asthma phenotype identification index screening system based on exosome omics of the present invention is shown.

[0079] The second embodiment of the present invention further provides a bronchial asthma phenotype identification index screening system 4 based on exosome omics, the system comprising: a memory 41 and a processor 42, wherein the memory comprises a bronchial asthma phenotype identification index screening method program based on exosome omics.

[0080] The third embodiment of the present invention also provides a computer-readable storage medium, which includes a program for screening a method for identifying indicators of a bronchial asthma phenotype based on exosome genomics. When the program for screening a method for identifying indicators of a bronchial asthma phenotype based on exosome genomics is executed by a processor, the steps of the method for screening a method for identifying indicators of a bronchial asthma phenotype based on exosome genomics are implemented.

[0081] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0082] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0083] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0084] Those skilled in the art can understand that: all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, and other media that can store program codes.

[0085] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

Claims

1. A method for screening bronchial asthma phenotype identification indicators based on exosome omics, characterized in that: The following steps are involved: Obtain a number of bronchial asthma medical record data, construct medical record data subsets based on different diagnostic phenotype categories, extract multidimensional information from different medical record data subsets, and obtain diagnostic features of different bronchial asthma phenotypes; Acquire the common diagnostic features and common diagnostic feature distribution of different bronchial asthma phenotypes, and generate an original feature set based on the common diagnostic features; Retrieve exosome-related knowledge using big data methods, construct a knowledge graph based on the mechanism of action of exosomes from different cell sources on bronchial asthma, use the knowledge graph to screen exosome features that meet preset requirements, and expand the original feature set; A feature selection task based on multi-objective optimization was constructed in the expanded original feature set, and the exosome feature combinations of different bronchial asthma phenotypes were obtained based on the common diagnostic features. The identification accuracy was compared and verified according to the distribution of the common diagnostic features, and the exosome identification indicators of the bronchial asthma phenotype were determined based on the verification results.

2. The method for screening bronchial asthma phenotype identification indicators based on exosome omics according to claim 1, characterized in that: Medical record data subsets are constructed based on different diagnostic phenotype categories, and multidimensional information is extracted from different medical record data subsets, specifically: Retrieving medical record data using bronchial asthma as a keyword, obtaining bronchial asthma medical record data that meets the search tag constraint, extracting diagnosis information, and obtaining bronchial asthma phenotypes based on the diagnosis information; Aggregating the corresponding bronchial asthma medical record data using the bronchial asthma phenotype, constructing medical record data subsets of different bronchial asthma phenotypes, extracting detection indicators in different medical record data subsets, and screening abnormal indicators; Determine the diagnosis basis according to the abnormal index combined with the clinical symptoms in the medical record data, obtain low-dimensional features from the diagnosis basis using single-channel convolution, and obtain high-dimensional features from the low-dimensional features based on convolution calculation, normalization processing and activation operation of the main branch; At the same time, the low-dimensional features are locally integrated based on the side branches using average pooling processing, the outputs of the two branches are fused to obtain multi-dimensional information, and diagnostic features of different bronchial asthma phenotypes are generated based on the multi-dimensional information.

3. The method for screening bronchial asthma phenotype identification indicators based on exosome omics according to claim 1, characterized in that: The common diagnostic features and common diagnostic feature distribution of different bronchial asthma phenotypes are obtained, and an original feature set is generated based on the common diagnostic features, specifically: Acquire diagnostic feature sets of different bronchial asthma phenotypes, perform intersection processing on each diagnostic feature set, extract common diagnostic features to generate an original feature set, and extract corresponding examination data from each medical record data subset according to the common diagnostic features in the original feature set; The examination data corresponding to the common diagnostic features are mapped to a low-dimensional space, the exponential distribution of the examination data in the low-dimensional space is obtained to generate a common diagnostic feature distribution, and the identification accuracy is obtained according to the common diagnostic feature distribution of different bronchial asthma phenotypes.

4. The method for screening bronchial asthma phenotype identification indicators based on exosome omics according to claim 1, characterized in that: According to the big data method, the knowledge related to exosome omics was retrieved, and the knowledge graph was constructed through the mechanism of action of exosomes from different cell sources on bronchial asthma, specifically: Through big data methods, we obtain professional corpus of exosome omics and big data information of exosome omics and bronchial asthma based on search engines. After structured processing, we use the encoder model to obtain the feature vector of exosome omics-related knowledge, and capture deep-level feature information to obtain entities. According to the big data information of exosomes and bronchial asthma, semantic features are obtained by embedding, the mechanism of action of exosomes from different cell sources on bronchial asthma is read through the semantic features, the relationship between entities is extracted based on the mechanism of action, and a relationship table is constructed; The acquired entities and relationship tables are mapped in triple form to construct a knowledge graph in a graphical way.

5. The method for screening bronchial asthma phenotype identification indicators based on exosome omics according to claim 1, characterized in that: The knowledge graph is used to screen exosome features whose importance meets the preset requirements, and the original feature set is expanded, specifically: For each entity in the knowledge graph, obtain the number of entities directly connected to the entity, obtain the entity contribution according to the number of entities, construct an adjacency matrix according to the entity contribution, and represent the knowledge graph in the form of an adjacency matrix; Use a graph neural network to perform representation learning on the knowledge graph, aggregate and propagate the features of entities in the knowledge graph through neighbor entities, construct entity embedding vectors, import the entity embedding vectors into the GRU unit to obtain the relationship between entities, and use the Softmax function to obtain the centrality of the entities; The importance score of the entity is obtained according to the centrality and the initial score, and the entity is sorted by the importance score, and the exosome features that meet the preset requirements are selected. After reducing feature redundancy by calculating the correlation between the exosome features, the original feature set is expanded.

6. The method for screening bronchial asthma phenotype identification indicators based on exosome omics according to claim 1, characterized in that: Construct a feature selection task based on multi-objective optimization in the expanded original feature set, specifically: Obtain the expanded original feature set, use feature importance and feature redundancy as evaluation indicators, and take maximum feature importance and minimum feature redundancy as optimization goals; A bronchial asthma phenotype classification model is constructed based on a convolutional autoencoder and a GUR unit, a data set is constructed based on the expanded original feature set and the corresponding inspection data, the data set is divided into several subsets, each subset is used as a validation set, and the remaining subsets are used as training sample sets to train the bronchial asthma phenotype classification model to obtain several models; The convolutional autoencoder part is composed of stacked convolutional layers, pooling layers and batch normalization layers. The convolutional autoencoder parameters are initialized and trained in an unsupervised training manner to obtain the recognition classification features of the output of the encoder part; The identification and classification features are introduced into the GRU unit to obtain the time correlation of the features, and the bronchial asthma phenotype classification is realized by using the fully connected layer through deep learning of the temporal features, and the average classification accuracy of several bronchial asthma phenotype classification models is obtained as an importance evaluation indicator; The mRMR algorithm was reconstructed using the Pearson correlation coefficient to calculate the correlation between the selected features and the correlation between the features and the bronchial asthma phenotype, and to obtain the redundancy evaluation index of the feature combination.

7. The method for screening bronchial asthma phenotype identification indicators based on exosome omics according to claim 1, characterized in that: Based on the common diagnostic features, the exosome feature combinations of different bronchial asthma phenotypes were obtained, specifically: The optimized Eagle algorithm is used to obtain the optimal exosome feature combination according to the optimization target, the Eagle algorithm parameters are initialized, the expanded original feature set is encoded, the common diagnostic features are used as feature benchmarks, and the initial feature combination is randomly generated on the feature benchmarks as the Eagle individuals; Calculate the fitness of the current position of the initialized Skyhawk population, update the global optimal fitness value and optimal position, preset the number of iterations threshold, and expand the search for global optimization when the number of iterations does not exceed the preset number of iterations threshold, otherwise, narrow the search for local search; Adaptive control factors are introduced to adjust the flight speed of the individual eagles in global optimization and local search, calculate the fitness value of the new position of the individual eagle, obtain the feature combination with the highest fitness value, use differential calculation to strengthen the recombination of the feature combination, and record the global optimal fitness value and optimal position; When the objective function of the feature combination converges, an optimal feature combination is obtained, and the common diagnostic features are removed from the optimal feature combination to obtain the exosome feature combination.

8. The method for screening bronchial asthma phenotype identification indicators based on exosome omics according to claim 1, characterized in that: The identification accuracy was compared and verified based on the common diagnostic feature distribution, and the exosome identification index of bronchial asthma phenotype was determined through the verification results, specifically: Based on the inspection data of the exosome feature combination obtained from the medical record data of bronchial asthma, the inspection data of the exosome feature combination is mapped to a low-dimensional space to obtain the exosome feature distribution, and the corresponding identification accuracy is obtained according to the exosome feature distribution of different bronchial asthma phenotypes; Obtain the corresponding identification accuracy of the common diagnostic feature distribution of different bronchial asthma phenotypes, and compare the deviation between the corresponding identification accuracy of the exosome feature distribution and the corresponding identification accuracy of the common diagnostic feature distribution; If the deviation is greater than the preset deviation threshold, the exosome feature of the current bronchial asthma phenotype is reacquired; if the deviation is not greater than the preset deviation threshold, the exosome feature combination is output.

Images