Method for extracting gene features in gene data based on sequence search

Abstract

The invention discloses a gene feature extraction method in gene data based on sequence search, and relates to the technical field of biological information. The problems that gene data is wide in source and diversified in format, data processing is extremely complex, data of different formats are difficult to integrate and analyze, and the accuracy and efficiency of feature extraction are affected are solved. According to the method, gene data are comprehensively preprocessed, low-quality and redundant parts are removed, reliable data quality is ensured, data distribution and data conversion are performed at the same time, high-similarity gene feature regions are accurately screened through sequence comparison, importance weights of gene features are determined by using clustering group features, key features are reasonably screened, and the accuracy of gene feature extraction is improved. Data dimensions are effectively reduced, the feature extraction efficiency and accuracy are improved, local and global features are fused to construct a comprehensive feature vector, the feature expression ability is enhanced, a screening result is verified by constructing a test data set, and the accuracy and reliability of the gene feature extraction method are continuously improved.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics, and in particular to a method for extracting gene features from gene data based on sequence search. Background Art

[0002] Accurately extracting valuable features from massive amounts of genetic data has become a key challenge in biological research. Chinese patent application publication number CN107133496A discloses a gene feature extraction method based on a dual manifold learning and closed-loop deep convolutional network model. The method includes the following steps: first, coarse extraction of cancer-associated gene features based on manifold learning; second, fine extraction of gene features based on a closed-loop deep convolutional dual network structure. The process is as follows: A dual network structure combining a forward convolutional neural network and an inverse convolutional neural network is employed to deeply abstract gene expression datasets using the convolutional neural network's feature extraction capabilities, ultimately projecting key features; and then an inverse convolutional neural network is used to inversely project these key features.

[0003] In existing technologies, genetic data comes from a wide range of sources and in diverse formats, making data processing extremely complex. Data in different formats are difficult to integrate and analyze, affecting the accuracy and efficiency of feature extraction. In addition, the diverse features of genetic data make it impossible to fuse the extracted genetic features to form more representative comprehensive features, thus failing to better reflect genetic functions and biological processes. Summary of the Invention

[0004] The purpose of the present invention is to provide a gene feature extraction method in gene data based on sequence search, which ensures the reliability of data quality by comprehensively preprocessing the gene data, removing low-quality and redundant parts, reasonably screening key features, effectively reducing data dimensions, improving feature extraction efficiency and accuracy, fusing local and global features to construct a comprehensive feature vector, and enhancing feature expression capabilities, so as to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] Gene feature extraction methods from gene data based on sequence search, including:

[0007] Obtain gene sequence data from a gene database and preprocess it. Use the preprocessed gene sequence as a query sequence and compare it with the sequences in the gene database based on the sequence type of the query sequence. Based on the comparison results, select gene feature regions with high similarity and identify potential gene features in the gene feature regions.

[0008] Gene feature statistical analysis is performed on each gene feature region, feature correlation is calculated, and gene features are screened. Feature vectors are constructed from the screened gene features, and feature fusion processing is performed on the feature vectors constructed for each gene feature region to form comprehensive features of the gene data.

[0009] Furthermore, gene sequence data is obtained from the gene database and preprocessed, specifically including:

[0010] Data format unification: Build a connection channel with the gene database. Based on the connection channel, convert the gene sequence data in various formats obtained from different gene databases into a standard format to generate a gene sequence dataset in a unified format.

[0011] Data cleaning: Verify each gene sequence in the gene sequence dataset and compare it with the preset threshold ranges of unknown base ratio, sequencing error rate and sequence length. Gene sequences with unknown base ratio exceeding the preset threshold, high test error rate or sequence length deviating from the preset threshold range will be removed from the gene sequence dataset;

[0012] Data conversion: Based on the preset coding mapping rules, the bases in the gene sequence are converted into corresponding digital codes, and a two-dimensional matrix is ​​constructed based on the digital codes. At the same time, the text annotation information associated with the gene sequence is converted into the corresponding classification code.

[0013] Furthermore, the data cleaning further includes:

[0014] Divide the gene sequence data set into multiple data blocks according to the length range of the gene sequence;

[0015] Analyze the variation pattern of the base composition of the gene sequence in each data block over the sequence length to verify the stability and consistency of the sequence characteristics;

[0016] Based on the verification results, the feature differences of each gene sequence at different positions are calculated to obtain the modal feature change value;

[0017] Input each gene sequence into the pre-established feedback model and output the feedback prediction result of the corresponding gene sequence;

[0018] Calculating a sequence feedback coefficient by utilizing the feedback prediction result;

[0019] Calculate the modal feature attenuation index and target hash value based on the modal feature change value and sequence feedback coefficient of each gene sequence;

[0020] Compare the target hash value of each gene sequence with the preset hash value. If the target hash value is greater than or equal to the preset hash value, it is confirmed that the sequence is not redundant data; if the target hash value is less than the preset hash value, it is confirmed that the sequence is redundant data;

[0021] Data sampling is performed on gene sequences identified as redundant data, the number of bases is used as the data volume, and the key features of the gene sequences are extracted as data feature values;

[0022] The data time-varying parameters are obtained according to the data volume and data characteristic values, and the data elimination method is matched according to the data time-varying parameters of each redundant data to eliminate the redundant data from the gene sequence data set.

[0023] Furthermore, obtaining gene sequence data from a gene database and preprocessing the data also includes:

[0024] Perform integrity testing on the cleaned gene sequence dataset and verify the key information in the gene sequence dataset;

[0025] Detect the data distribution of the gene sequence data set, determine the gene sequence feature distribution corresponding to the gene sequence data set, and obtain the detection results;

[0026] The detection results are analyzed by a preset sparse judgment method to generate a random sparse observation matrix that can reflect the characteristics of the data set;

[0027] The sparse observation value vector ratio of the gene sequence dataset is calculated based on the random sparse observation matrix to evaluate the sparsity and feature expression ability of the gene sequence dataset;

[0028] Traversing the sparse observation value vector ratio, according to the sparsity and feature expression ability of the gene sequence data set, screening out gene sequences with higher than the preset sparsity and feature expression ability thresholds in the gene sequence data set;

[0029] The screened gene sequences are clustered and grouped, gene sequences with similar characteristic expression patterns are integrated, a unique label is assigned to each cluster group, and the characteristics of each cluster group are summarized and described.

[0030] Furthermore, based on the comparison results, the gene feature regions with high similarity were screened out, including:

[0031] determining a sequence type of the query sequence based on the cluster group labels of the preprocessed query sequence;

[0032] The query sequence of each cluster group is compared with the sequences in the gene database. During the comparison process, the comparison score, the start and end positions of the comparison, and the matching and mismatching bases of the query sequence and the database sequence are recorded;

[0033] Based on the comparison results, the comparison scores of the gene feature regions are obtained, the gene feature regions with comparison scores higher than the preset comparison threshold are extracted, the consistency between the gene feature regions and the cluster group features is analyzed, and the comparison results are classified and stored according to the cluster group labels;

[0034] Determine the corresponding similarity screening threshold based on the characteristics of each cluster group, traverse the comparison results of each cluster group, and mark the regions with comparison scores higher than the similarity screening threshold as high-similarity gene feature regions;

[0035] Based on the screening results, the similar gene feature regions of each cluster group are analyzed to identify potential gene features, and the identified potential gene features are associated with the cluster group features to obtain gene feature data.

[0036] Furthermore, genetic features are screened, including:

[0037] According to the characteristics of each cluster group, the importance weight of each gene feature is determined by comprehensively considering the average sparse observation value vector ratio of the cluster group, the main base composition characteristics, and the factors involved in the gene function;

[0038] The feature screening threshold is adjusted based on the importance weight of the gene feature, and gene features with importance weights higher than the feature screening threshold are screened out.

[0039] Furthermore, determining the importance weight of each gene feature also includes:

[0040] Obtain direct-gene features that directly interact with the current gene feature, and indirect-gene features that indirectly interact with the current gene feature within the same cluster group;

[0041] By analyzing the degree of synchronization between the expression changes of the current gene signature and each direct-gene signature, the direct mutual-synchronization coefficient is obtained;

[0042] Using all the direct mutual-synchronization coefficients obtained, the direct mutual-reflection matrix of the current gene feature is constructed, and the direct mutual reflection coefficient is obtained through matrix calculation;

[0043] By analyzing the degree of synchronization between the expression changes of the current gene signature and each indirect-gene signature, the indirect mutual-synchronization coefficient is obtained;

[0044] Obtain the number of intermediate mediating genes between the current gene feature and each indirect-gene feature, and calculate the difference between the number of intermediate mediating genes of each indirect-gene feature and the maximum number of intermediate mediating genes;

[0045] The indirect mutual-reflection matrix of the current gene feature is constructed using the corresponding indirect mutual-synchronization coefficients of all indirect-gene features and the difference in the number of intermediate mediating genes, and the indirect mutual-reflection coefficients are obtained through matrix calculation;

[0046] The direct mutual reflection coefficient, indirect mutual reflection coefficient, number of direct-gene features and number of indirect-gene features are combined and calculated to obtain the comprehensive mutual reflection score of the current gene feature;

[0047] For the current gene feature, the mutual influence weight is assigned according to the total number of corresponding gene features with interactions;

[0048] The mutual comprehensive reflection score of the current gene characteristics and the mutual influence weight are combined and calculated to obtain the interaction influence coefficient;

[0049] For all gene features in the current cluster group, the first feature arrangement table is constructed from large to small using the interaction influence coefficient;

[0050] For all gene features in the current cluster group, the weights obtained by comprehensively considering the average sparse observation value vector ratio of the cluster group, the main base composition characteristics, and the factors involved in the gene function will be marked as pending weights;

[0051] For all gene features in the current cluster group, a second feature arrangement table is constructed from large to small using the undetermined weights;

[0052] If the first feature ranking table and the second feature ranking table are consistent, the undetermined weight of each gene feature is output as the corresponding importance weight;

[0053] If the first feature ranking table and the second feature ranking table are inconsistent, the weight value obtained by adjusting the predetermined weight using the interaction influence coefficient will be output as the importance weight of the corresponding gene feature.

[0054] Furthermore, feature fusion processing is performed on the feature vectors constructed for each gene feature region, which also includes:

[0055] Determining statistics of the feature vectors of each gene feature region based on the feature vectors of each gene feature region, and arranging the calculated statistics in order to form a quantitative representation of the local features of the gene feature region;

[0056] Perform statistical analysis on the entire gene data set, calculate the global eigenvalue based on the statistical analysis results, integrate the calculated global eigenvalues ​​to obtain the global feature representation of the gene data set, and fuse the local eigenvectors with the global feature representation.

[0057] Furthermore, it also includes:

[0058] Construct a test data set, and separate the test data set from the original gene data set according to the preset division rules, wherein the real feature information of each gene in the test data set is obtained;

[0059] The overall accuracy of the feature extraction method was determined by calculating the ratio of the number of correctly predicted gene features to the total number of gene features, and the coverage of the real gene features by the feature extraction method was evaluated by calculating the ratio of the number of correctly predicted gene features to the number of actual gene features.

[0060] The performance of the gene feature extraction method was evaluated based on the overall accuracy and recall evaluation indicators, and the gene feature extraction method was optimized based on the evaluation results.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] By comprehensively preprocessing the genetic data, removing low-quality and redundant parts, ensuring the reliability of data quality, and detecting data distribution and conversion data at the same time, a good foundation is laid for subsequent analysis. Through sequence alignment, high-similarity gene feature regions are accurately screened, and the importance weights of gene features are determined by using cluster group features. Key features are reasonably screened, data dimensions are effectively reduced, and feature extraction efficiency and accuracy are improved. Local and global features are integrated to construct a comprehensive feature vector, enhancing feature expression capabilities. The screening results are verified by constructing a test data set. Feature screening parameters, fusion strategies, training data and models are optimized according to evaluation indicators to continuously improve the accuracy and reliability of gene feature extraction methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a flow chart of the gene feature extraction method in gene data based on sequence search of the present invention. DETAILED DESCRIPTION

[0064] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only 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 making creative efforts are within the scope of protection of the present invention.

[0065] To address the technical issues that genetic data comes from a wide range of sources and in various formats, making data processing extremely complex, data in different formats difficult to integrate and analyze, affecting the accuracy and efficiency of feature extraction, and the diverse features of genetic data make it impossible to fuse the extracted genetic features to form a more representative comprehensive feature, thus failing to better reflect gene function and biological processes, please refer to Figure 1 , this embodiment provides the following technical solutions:

[0066] Gene feature extraction methods from gene data based on sequence search, including:

[0067] Obtain gene sequence data from a gene database and preprocess it. Use the preprocessed gene sequence as a query sequence and compare it with the sequences in the gene database based on the sequence type of the query sequence. Based on the comparison results, select gene feature regions with high similarity and identify potential gene features in the gene feature regions.

[0068] Perform statistical analysis of gene features on each gene feature region, calculate feature correlation, and screen gene features. Construct feature vectors from the screened gene features, and perform feature fusion processing on the feature vectors constructed from each gene feature region to form comprehensive features of the gene data.

[0069] Construct a test dataset, separating it from the original gene dataset according to a preset partitioning rule. The test dataset contains the true feature information of each gene in the test dataset. The test dataset is independent of the training dataset used for model training and feature extraction method optimization, and can fully represent the overall feature distribution of the original dataset.

[0070] The overall accuracy of the feature extraction method was determined by calculating the ratio of the number of correctly predicted gene features to the total number of gene features, and the coverage of the real gene features by the feature extraction method was evaluated by calculating the ratio of the number of correctly predicted gene features to the number of actual gene features.

[0071] The performance of the gene feature extraction method is evaluated based on the overall accuracy and recall evaluation indicators, and the gene feature extraction method is optimized based on the evaluation results;

[0072] In this embodiment, the optimization method includes: adjusting feature screening parameters: adjusting the calculation method of the importance weight of gene features based on the evaluation results, such as modifying the weight distribution of cluster group feature factors to more accurately reflect the importance of gene features; optimizing feature fusion strategies: for multimodal feature fusion strategies, evaluating the contribution of different modal features, adjusting the fusion method or fusion parameters of multimodal features to improve the coverage of real gene features; adding or adjusting training data: checking whether the division of training data sets and test data sets is reasonable to ensure that the distribution difference between the two does not have a large impact on model performance, and re-dividing the data sets if necessary; retraining the model: gradually improving the accuracy and reliability of the gene feature extraction method through continuous verification, evaluation and optimization.

[0073] In this embodiment, gene sequence data is obtained from a gene database and preprocessed, specifically including:

[0074] Data format unification: Build a connection channel with the gene database. Based on the connection channel, convert the gene sequence data in various formats obtained from different gene databases into a standard format to generate a gene sequence dataset in a unified format. For example, the sequences and annotations in various formats are integrated into a text file with a specific structure. Each line records a gene sequence and its annotations, and different fields are separated by specific delimiters.

[0075] Data cleaning: Verify each gene sequence in the gene sequence dataset and compare it with the preset threshold ranges of unknown base ratio, sequencing error rate and sequence length. Gene sequences with unknown base ratio exceeding the preset threshold, high test error rate or sequence length deviating from the preset threshold range will be removed from the gene sequence dataset;

[0076] In this example, the threshold for the percentage of unknown bases is set at no more than 10% of the sequence length; the sequencing error rate can be set at no more than 5% based on the average error level of the sequencing technology; and the sequence length is set with reasonable upper and lower limits based on the common length range of gene sequences of the species. Low-quality gene sequences containing a large number of unknown bases, high sequencing error rates, or abnormal sequence lengths are removed from the dataset to improve the overall quality of the data.

[0077] Data conversion: Based on the preset coding mapping rules, the bases in the gene sequence are converted into corresponding digital codes, such as the bases (A, T, C, G), with A corresponding to 1, T corresponding to 2, C corresponding to 3, and G corresponding to 4, to achieve the initial conversion of the gene sequence from character form to numerical form. A two-dimensional matrix is ​​constructed based on the digital code, with each row representing an independent gene sequence and each column corresponding to a specific position in the gene sequence. At the same time, the textual annotation information associated with the gene sequence is converted into the corresponding classification code so that it can be integrated and processed in the same data structure as the gene sequence data.

[0078] In this embodiment, the data cleaning further includes:

[0079] Divide the gene sequence dataset into multiple data blocks according to the length range of the gene sequences to ensure that the gene sequences in each data block have certain similarity and correlation;

[0080] Analyze the variation pattern of the base composition of the gene sequence in each data block over the sequence length to verify the stability and consistency of the sequence characteristics;

[0081] Based on the verification results, the feature differences of each gene sequence at different positions are calculated to obtain the modal feature change value, which reflects the degree of dynamic change of sequence features;

[0082] Input each gene sequence into the pre-established feedback model and output the feedback prediction result of the corresponding gene sequence;

[0083] By using the feedback prediction results, a sequence feedback coefficient is calculated, which reflects the correlation between the sequence and other sequences;

[0084] Calculate the modal feature attenuation index and target hash value based on the modal feature change value and sequence feedback coefficient of each gene sequence;

[0085] Compare the target hash value of each gene sequence with the preset hash value. If the target hash value is greater than or equal to the preset hash value, it is confirmed that the sequence is not redundant data; if the target hash value is less than the preset hash value, it is confirmed that the sequence is redundant data;

[0086] Data sampling is performed on gene sequences identified as redundant data, with the number of bases used as the data volume. At the same time, key features of the gene sequences (such as base composition ratio, frequency of occurrence of specific patterns, etc.) are extracted as data feature values.

[0087] The data time-varying parameter is obtained according to the data volume and data characteristic value. The parameter reflects the changing characteristics of redundant data in time and characteristic dimensions. The data elimination method is matched according to the data time-varying parameter of each redundant data, and the redundant data is eliminated from the gene sequence dataset.

[0088] In this embodiment, the feedback model is obtained by training a neural network. The model used to predict the frequency of occurrence of carbon bases at each position in a gene sequence is obtained by collecting and preprocessing a large amount of gene sequence data, converting the sequence into a format suitable for model input, and extracting sequence data features (including carbon base coding, sequence k-mer frequency, GC content, sequence complexity, etc.) to train the neural network. When constructing the model, an input layer (receiving sequence data), a hidden layer (comprising multiple layers for capturing the complex features of the sequence), and an output layer (predicting the frequency of occurrence of bases at each position) are defined; the feedback prediction result refers to the predicted probability of occurrence of the base in the gene sequence at each position.

[0089] In this embodiment, the calculation formula of the sequence feedback coefficient is as follows:

[0090]

[0091] Where F is the sequence feedback coefficient of the current gene sequence; F1 is the sum of the actual and predicted frequency differences of the carbon bases in the current gene sequence within the range of [1, m-3]; F2 is the sum of the actual and predicted frequency differences of the carbon bases in the current gene sequence within the range of [m-2, m]; y i It represents the predicted frequency of occurrence of the carbon base in the current gene sequence at the i-th position, where i = 1, 2, 3, ..., j + k; m represents the total number of positions; j represents the position index, traversing the position of the current gene sequence; ys,i It represents the actual frequency of occurrence of the carbon base at the i-th position of the current gene sequence; k represents the set window size; i represents the position index within the set window, with position j as the center, extending k positions to the left (m-2≤j≤m) or to the right (1≤j≤m-3).

[0092] The beneficial effects of the above technical solution are: by calculating the sequence feedback coefficient, the correlation between gene sequences can be quantified, and it helps to effectively identify redundant sequences that are highly similar to other sequences and have low information content, and then adopt appropriate data elimination strategies to improve the purity of the data set and analysis efficiency.

[0093] In this example, modal feature variation and sequence feedback coefficients were used to effectively identify and eliminate redundant data, optimizing the dataset structure, data cleaning, and feature extraction processes. This significantly improved data quality and ensured the similarity and relevance of gene sequences. This enhanced the stability and consistency of sequence features, providing a solid foundation for subsequent analysis. Furthermore, the use of time-varying parameters to match data elimination methods improved data processing efficiency.

[0094] In this embodiment, obtaining gene sequence data from a gene database and performing preprocessing also includes:

[0095] Perform integrity testing on the cleaned gene sequence dataset, counting the number of gene sequences in the dataset and comparing it with the expected number; check whether the length of each gene sequence is within the expected range and whether the annotation information is complete; at the same time, verify key information in the gene sequence dataset, such as the uniqueness of gene identifiers and the legitimacy of base sequences;

[0096] Detect the data distribution of the gene sequence data set, determine the gene sequence feature distribution corresponding to the gene sequence data set, including gene sequence length distribution, base composition distribution, gene function classification distribution, etc., and obtain the test results;

[0097] The detection results are analyzed by a preset sparse judgment method to generate a random sparse observation matrix that can reflect the characteristics of the data set;

[0098] The sparse observation value vector ratio of the gene sequence dataset is calculated based on the random sparse observation matrix, the sparsity and feature expression ability of the gene sequence dataset are evaluated, and features that contribute significantly to subsequent gene feature extraction are screened out, thereby reducing data dimensions and improving the efficiency and accuracy of feature extraction.

[0099] Traverse the sparse observation value vector ratio, and according to the sparsity and feature expression ability of the gene sequence dataset, screen out gene sequences with higher than the preset sparsity and feature expression ability thresholds in the gene sequence dataset, indicating that the gene sequence is a feature with strong expression ability and relatively important;

[0100] The screened gene sequences are clustered and grouped, and gene sequences with similar characteristic expression patterns are integrated. A unique label is assigned to each cluster group, and the characteristics of each cluster group are summarized and described, such as the average sparse observation value vector ratio of the group's characteristics, the main base composition characteristics, the gene functions involved, etc.

[0101] In this embodiment, the integrity and quality of the data set are ensured through integrity detection, and analysis bias caused by missing or abnormal data is avoided. Data distribution detection and the generation of sparse observation matrices help to deeply understand the characteristic distribution of gene sequences. By calculating and screening the sparse observation value vector ratio, the data dimension is effectively reduced, and the gene sequences that make important contributions to feature extraction are highlighted, thereby improving the analysis efficiency. Clustering and grouping further integrate gene sequences with similar characteristic expression patterns, and provide a detailed feature description for each cluster group, which not only optimizes the gene data processing process, but also enhances the recognition ability of characteristic expression.

[0102] In this embodiment, based on the comparison results, the gene feature regions with high similarity are screened out, specifically including:

[0103] determining a sequence type of the query sequence based on the cluster group labels of the preprocessed query sequence;

[0104] The query sequence of each cluster group is compared with the sequences in the gene database. During the comparison process, the comparison score, the start and end positions of the comparison, and the matching and mismatching bases of the query sequence and the database sequence are recorded;

[0105] Based on the alignment results, the alignment scores of the gene feature regions are obtained, and the gene feature regions with alignment scores higher than the preset alignment threshold are extracted. The consistency between the gene feature regions and the cluster group characteristics is analyzed. For example, if the cluster group has a specific base composition preference, the alignment region is checked to see if it meets the preference. The alignment results are classified and stored according to the cluster group label to facilitate subsequent screening operations for each cluster group.

[0106] Determine the corresponding similarity screening threshold based on the characteristics of each cluster group. For example, for cluster groups with high feature stability and strong intra-group sequence consistency, set a higher similarity threshold; for cluster groups with relatively complex features and certain differences within the group, appropriately lower the threshold. Traverse the comparison results of each cluster group and mark the regions with comparison scores higher than the similarity screening threshold as high-similarity gene feature regions;

[0107] Based on the screening results, the similar gene feature regions of each cluster group are analyzed to identify potential gene features, and the identified potential gene features are associated with the cluster group features to obtain gene feature data. For example, if a cluster group mainly involves metabolic functions, the correlation between the screened regions and metabolic pathways is analyzed, and the potential gene feature rules contained in the similar gene feature regions of different cluster groups are summarized to provide key information for the subsequent construction of gene feature vectors and feature fusion.

[0108] In this embodiment, screening of gene features specifically includes:

[0109] Based on the characteristics of each cluster group, the importance weight of each gene feature is determined by comprehensively considering the average sparse observation value vector ratio of the cluster group, the main base composition characteristics, and the factors of the gene function involved. Gene features closely related to the core function of the cluster group are given higher importance weights; features that are prevalent and highly stable in the cluster group are also appropriately weighted. For example, if a cluster group is mainly involved in immune response, the weight of gene features related to immunity should be relatively high.

[0110] The feature screening threshold is adjusted in combination with the importance weight of the gene feature. For features with higher importance weight, the screening threshold is lowered to ensure that more valuable features are retained; for features with lower importance weight, the screening threshold is increased to remove possible noise features; and gene features with importance weight higher than the feature screening threshold are screened out.

[0111] In this embodiment, determining the importance weight of each gene feature also includes:

[0112] Obtain direct-gene features that directly interact with the current gene feature, and indirect-gene features that indirectly interact with the current gene feature within the same cluster group;

[0113] By analyzing the degree of synchronization between the expression changes of the current gene signature and each direct-gene signature, the direct mutual-synchronization coefficient is obtained;

[0114] Using all the direct mutual-synchronization coefficients obtained, the direct mutual-reflection matrix of the current gene feature is constructed, and the direct mutual reflection coefficient is obtained through matrix calculation;

[0115] By analyzing the degree of synchronization between the expression changes of the current gene signature and each indirect-gene signature, the indirect mutual-synchronization coefficient is obtained;

[0116] Obtain the number of intermediate mediating genes between the current gene feature and each indirect-gene feature, and calculate the difference between the number of intermediate mediating genes of each indirect-gene feature and the maximum number of intermediate mediating genes;

[0117] The indirect mutual-reflection matrix of the current gene feature is constructed using the corresponding indirect mutual-synchronization coefficients of all indirect-gene features and the difference in the number of intermediate mediating genes, and the indirect mutual-reflection coefficients are obtained through matrix calculation;

[0118] The direct mutual reflection coefficient, indirect mutual reflection coefficient, number of direct-gene features and number of indirect-gene features are combined and calculated to obtain the comprehensive mutual reflection score of the current gene feature;

[0119] For the current gene feature, the mutual influence weight is assigned according to the total number of corresponding gene features with interactions;

[0120] The mutual comprehensive reflection score of the current gene characteristics and the mutual influence weight are combined and calculated to obtain the interaction influence coefficient;

[0121] For all gene features in the current cluster group, the first feature arrangement table is constructed from large to small using the interaction influence coefficient;

[0122] For all gene features in the current cluster group, the weights obtained by comprehensively considering the average sparse observation value vector ratio of the cluster group, the main base composition characteristics, and the factors involved in the gene function will be marked as pending weights;

[0123] For all gene features in the current cluster group, a second feature arrangement table is constructed from large to small using the undetermined weights;

[0124] If the first feature ranking table and the second feature ranking table are consistent, the undetermined weight of each gene feature is output as the corresponding importance weight;

[0125] If the first feature ranking table and the second feature ranking table are inconsistent, the weight value obtained by adjusting the predetermined weight using the interaction influence coefficient will be output as the importance weight of the corresponding gene feature.

[0126] In this embodiment, a direct-gene feature refers to a gene feature in the same cluster group that directly interacts with the current gene feature, that is, there is a direct molecular pathway or regulatory relationship between the two, where the interaction includes physical interaction, genetic interaction, etc.; an indirect-gene feature refers to a gene feature that interacts with the current gene feature through one or more intermediate genes or molecules.

[0127] In this embodiment, the direct mutual-synchronization coefficient is used to measure the degree of synchronization of expression changes between the current gene feature and the direct-gene feature. It is calculated based on the expression data of the current gene feature and the direct-gene feature (referring to the expression level quantification value determined based on the corresponding data of the gene feature in the cluster group, wherein the expression level quantification value is obtained by querying a known gene expression database (such as the GEO database, the ArrayExpress database, the gene expression database is composed of gene sequence data, expression level quantification values ​​and sample information, etc.)) using a preset synchronization degree measurement algorithm (such as the Pearson correlation coefficient).

[0128] In this embodiment, the direct interaction-reflection matrix is ​​a two-dimensional matrix, in which the rows represent different direct-gene features that have direct interactions with the current gene feature, the columns represent the current gene feature, and the elements in the matrix represent the direct interaction-synchronization coefficients between the current gene feature and the different direct-gene features; the direct interaction-reflection coefficients are obtained by calculating the direct interaction matrix using the principal component analysis method, and are used to reflect the overall interaction strength between the current gene feature and the direct-gene feature.

[0129] In this embodiment, the indirect mutual-synchronization coefficient is used to measure the degree of synchronization of expression changes between the current gene feature and the indirect-gene feature. It is calculated based on the expression data of the current gene feature and the indirect-gene feature (referring to the expression level quantification value determined based on the corresponding data of the gene feature in the cluster group, wherein the expression level quantification value is obtained by querying a known gene expression database (such as the GEO database, the ArrayExpress database, the gene expression database consists of gene sequence data, expression level quantification values ​​and sample information, etc.)) using a preset synchronization degree measurement algorithm (such as the Pearson correlation coefficient).

[0130] In this embodiment, the number of intermediate mediating genes refers to the number of one or more intermediate genes or molecules required for the current gene feature to interact with the indirect-gene feature; the maximum number of intermediate mediating genes refers to the maximum number of one or more intermediate genes or molecules required for the current gene feature to interact with each indirect-gene feature in the same cluster group; the difference in the number of intermediate mediating genes refers to the absolute difference between the number of intermediate mediating genes between the current gene feature and each indirect-gene feature and the maximum number of intermediate mediating genes.

[0131] In this embodiment, the indirect interaction-reflection matrix is ​​a two-dimensional matrix, in which rows represent different indirect-gene features that have indirect interactions with the current gene feature, columns represent the current gene feature, and the elements in the matrix represent the indirect interaction-synchronization coefficients between the current gene feature and different indirect-gene features and the difference in the number of intermediate mediating genes; the indirect interaction-reflection coefficient is obtained by calculating the indirect interaction-reflection matrix using the principal component analysis method, and is used to reflect the overall interaction strength between the current gene feature and the indirect-gene feature.

[0132] In this embodiment, the comprehensive mutual reflection score is calculated by combining the direct mutual reflection coefficient, the indirect mutual reflection coefficient, the number of direct-gene features and the number of indirect-gene features, and then introducing the weights assigned to direct mutual reflection and indirect mutual reflection respectively (obtained by solving the matrix constructed after pairwise comparison and scoring using the hierarchical analysis method), which is used to comprehensively reflect the overall interaction of the current gene features in the cluster group.

[0133] In this embodiment, for example, there is a gene feature A including 5 direct gene features, direct mutual reflection coefficient a1, 3 indirect gene features, indirect mutual reflection coefficient a2. At this time, the mutual comprehensive reflection score of gene feature A is Where ω1 represents the contribution weight of the direct interaction degree to the overall interaction of gene features in the cluster group, and its value range is (0, 1); ω2 represents the contribution weight of the indirect interaction degree to the overall interaction of gene features in the cluster group, and its value range is (0, 1).

[0134] In this embodiment, the total number of gene features refers to the sum of the number of direct-gene features (corresponding to the total number of direct-gene features) and the number of indirect-gene features (corresponding to the total number of indirect-gene features) of the gene features; the mutual influence weight is determined based on the total number of gene features, expressed as h1=1-e -c ; Among them, h1 represents the mutual influence weight of the current gene feature; e represents the base of the natural logarithm, which is 2.7; c represents the total number of gene features of the current gene feature; the interaction influence coefficient is calculated by directly multiplying the mutual comprehensive reflection score and the mutual influence weight, which is used to reflect the degree of influence and criticality of the current gene feature on the overall interaction network of the cluster group.

[0135] In this embodiment, the first feature ranking table refers to a list obtained by sorting all gene features in the same cluster group from large to small according to the interaction influence coefficient; the pending weight refers to the weight assigned to each gene feature after comprehensively considering factors such as the average sparse observation value vector ratio of the cluster group, the main base composition characteristics, and the gene functions involved; the second feature ranking table refers to a list obtained by sorting all gene features in the same cluster group from large to small according to the pending weight.

[0136] In this embodiment, for example, if the first feature ranking table l1 and the second feature ranking table l2 are inconsistent, the interaction influence coefficient in the l1 table is used to adjust the undetermined weights of the same gene feature in the l2 table. The formula is expressed as Where U1 represents the weight value after gene feature adjustment; U0 represents the undetermined weight before gene feature adjustment; ΔS z It is expressed as the sum of all interaction coefficients in the l1 table; s is expressed as the interaction coefficient of a single gene feature in the table; s avg Expressed as the coefficient mean of all interaction effect coefficients in the l1 table.

[0137] The working principle of the above technical solution is as follows: first, the direct-gene features and indirect-gene features within the same cluster group of the current gene feature are obtained, and the degree of synchronization of expression changes between the current gene feature and the direct-gene features and indirect-gene features is analyzed to obtain the corresponding synchronization coefficient; then, the obtained synchronization coefficient is used to construct a direct mutual-reflection matrix and an indirect mutual-reflection matrix, and the direct mutual reflection coefficient and the indirect mutual reflection coefficient are obtained through matrix calculation, thereby further quantifying the interaction between the gene features; then, the direct mutual reflection coefficient, the indirect mutual reflection coefficient, the number of direct-gene features and the number of indirect-gene features are combined and calculated to obtain the mutual comprehensive reflection score of the current gene feature to reflect the interaction and importance of the gene feature in the cluster group; then, based on the mutual comprehensive reflection score and the mutual influence weight, the interaction influence coefficient is calculated, and a first feature ranking table is constructed. At the same time, the undetermined weight is assigned according to the overall characteristics of the cluster group, and a second feature ranking table is constructed; finally, the final importance weight is determined by comparing the consistency of the two ranking tables. If the two ranking tables are inconsistent, the undetermined weight is adjusted using the interaction influence coefficient to obtain a more accurate importance weight.

[0138] The beneficial effect of the above technical solution is: by considering the direct and indirect interactions between gene features within the same cluster group, and combining the weights assigned to each gene feature after comprehensively considering factors such as the average sparse observation value vector ratio of the cluster group, the main base composition characteristics, and the gene functions involved, it can effectively achieve a multi-dimensional evaluation of the importance of gene features, which is conducive to improving the accuracy and rationality of the weighting of gene features, and thus helps to ensure the scientific nature of subsequent adjustment of feature screening thresholds and screening of gene features.

[0139] In this embodiment, the feature fusion process of the feature vectors constructed for each gene feature region also includes:

[0140] Based on the eigenvectors of each gene characteristic region, the mean, median, variance and other statistical quantities of the eigenvectors of the gene characteristic region are determined to reflect the central tendency and dispersion degree of the eigenvectors. The calculated statistics are arranged in order to form a quantitative representation of the local characteristics of the gene characteristic region.

[0141] Perform statistical analysis on the entire gene dataset, calculate global eigenvalues ​​based on the statistical analysis results, integrate the calculated global eigenvalues ​​to obtain the global feature representation of the gene dataset, and fuse the local feature vectors with the global feature representation. The fused feature vectors contain both the unique information of the gene feature region and reflect the overall characteristics of the gene dataset, thereby enhancing the expressive power of the comprehensive features.

[0142] In this embodiment, by comprehensively considering the average sparse observation value vector ratio, main base composition characteristics and gene function of the cluster group, an importance weight is assigned to each gene feature, and the feature screening threshold is adjusted based on these weights, which effectively retains valuable features and removes noise, thereby enhancing the accuracy of feature screening. At the same time, by calculating the statistics of the feature vector and fusing local and global features, a fused feature vector is formed that contains both unique information and reflects the overall characteristics, which significantly improves the expressive power of the features. This not only improves the efficiency of data analysis and reduces processing complexity, but also provides a solid foundation for building a more accurate and stable gene analysis model, which helps to deeply explore the biological functions and regulatory mechanisms of genes.

[0143] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A gene feature extraction method in gene data based on sequence search, characterized in that: include: Obtain gene sequence data from a gene database and preprocess it. Use the preprocessed gene sequence as a query sequence and compare it with the sequences in the gene database based on the sequence type of the query sequence. Based on the comparison results, select gene feature regions with high similarity and identify potential gene features in the gene feature regions. Gene feature statistical analysis is performed on each gene feature region, feature correlation is calculated, and gene features are screened. Feature vectors are constructed from the screened gene features, and feature fusion processing is performed on the feature vectors constructed for each gene feature region to form comprehensive features of the gene data.

2. The method for extracting gene features from gene data based on sequence search according to claim 1, wherein: Obtain gene sequence data from the gene database and perform preprocessing, specifically including: Data format unification: Build a connection channel with the gene database. Based on the connection channel, convert the gene sequence data in various formats obtained from different gene databases into a standard format to generate a gene sequence dataset in a unified format. Data cleaning: Verify each gene sequence in the gene sequence dataset and compare it with the preset threshold ranges of unknown base ratio, sequencing error rate and sequence length. Gene sequences with unknown base ratio exceeding the preset threshold, high test error rate or sequence length deviating from the preset threshold range will be removed from the gene sequence dataset; Data conversion: Based on the preset coding mapping rules, the bases in the gene sequence are converted into corresponding digital codes, and a two-dimensional matrix is ​​constructed based on the digital codes. At the same time, the text annotation information associated with the gene sequence is converted into the corresponding classification code.

3. The method for extracting gene features from gene data based on sequence search according to claim 2, wherein: The data cleaning further includes: Divide the gene sequence data set into multiple data blocks according to the length range of the gene sequence; Analyze the variation pattern of the base composition of the gene sequence in each data block over the sequence length to verify the stability and consistency of the sequence characteristics; Based on the verification results, the feature differences of each gene sequence at different positions are calculated to obtain the modal feature change value; Input each gene sequence into the pre-established feedback model and output the feedback prediction result of the corresponding gene sequence; Calculating a sequence feedback coefficient by utilizing the feedback prediction result; Calculate the modal feature attenuation index and target hash value based on the modal feature change value and sequence feedback coefficient of each gene sequence; Compare the target hash value of each gene sequence with the preset hash value. If the target hash value is greater than or equal to the preset hash value, it is confirmed that the sequence is not redundant data; if the target hash value is less than the preset hash value, it is confirmed that the sequence is redundant data; Data sampling is performed on gene sequences identified as redundant data, the number of bases is used as the data volume, and the key features of the gene sequences are extracted as data feature values; The data time-varying parameters are obtained according to the data volume and data characteristic values, and the data elimination method is matched according to the data time-varying parameters of each redundant data to eliminate the redundant data from the gene sequence data set.

4. The method for extracting gene features from gene data based on sequence search according to claim 2, wherein: Obtaining gene sequence data from gene databases and preprocessing it also includes: Perform integrity testing on the cleaned gene sequence dataset and verify the key information in the gene sequence dataset; Detect the data distribution of the gene sequence data set, determine the gene sequence feature distribution corresponding to the gene sequence data set, and obtain the detection results; The detection results are analyzed by a preset sparse judgment method to generate a random sparse observation matrix that can reflect the characteristics of the data set; The sparse observation value vector ratio of the gene sequence dataset is calculated based on the random sparse observation matrix to evaluate the sparsity and feature expression ability of the gene sequence dataset; Traversing the sparse observation value vector ratio, according to the sparsity and feature expression ability of the gene sequence data set, screening out gene sequences with higher than the preset sparsity and feature expression ability thresholds in the gene sequence data set; The screened gene sequences are clustered and grouped, gene sequences with similar characteristic expression patterns are integrated, a unique label is assigned to each cluster group, and the characteristics of each cluster group are summarized and described.

5. The method for extracting gene features from gene data based on sequence search according to claim 1, wherein: Based on the comparison results, high-similarity gene feature regions are screened out, including: determining a sequence type of the query sequence based on the cluster group labels of the preprocessed query sequence; The query sequence of each cluster group is compared with the sequences in the gene database. During the comparison process, the comparison score, the start and end positions of the comparison, and the matching and mismatching bases of the query sequence and the database sequence are recorded; Based on the comparison results, the comparison scores of the gene feature regions are obtained, the gene feature regions with comparison scores higher than the preset comparison threshold are extracted, the consistency between the gene feature regions and the cluster group features is analyzed, and the comparison results are classified and stored according to the cluster group labels; Determine the corresponding similarity screening threshold based on the characteristics of each cluster group, traverse the comparison results of each cluster group, and mark the regions with comparison scores higher than the similarity screening threshold as high-similarity gene feature regions; Based on the screening results, the similar gene feature regions of each cluster group are analyzed to identify potential gene features, and the identified potential gene features are associated with the cluster group features to obtain gene feature data.

6. The method for extracting gene features from gene data based on sequence search according to claim 1, wherein: Screening for genetic characteristics, including: According to the characteristics of each cluster group, the importance weight of each gene feature is determined by comprehensively considering the average sparse observation value vector ratio of the cluster group, the main base composition characteristics, and the factors involved in the gene function; The feature screening threshold is adjusted based on the importance weight of the gene feature, and gene features with importance weights higher than the feature screening threshold are screened out.

7. The method for extracting gene features from gene data based on sequence search according to claim 6, wherein: Determine the importance weight of each gene feature, including: Obtain direct-gene features that directly interact with the current gene feature, and indirect-gene features that indirectly interact with the current gene feature within the same cluster group; By analyzing the degree of synchronization between the expression changes of the current gene signature and each direct-gene signature, the direct mutual-synchronization coefficient is obtained; Using all the direct mutual-synchronization coefficients obtained, the direct mutual-reflection matrix of the current gene feature is constructed, and the direct mutual reflection coefficient is obtained through matrix calculation; By analyzing the degree of synchronization between the expression changes of the current gene signature and each indirect-gene signature, the indirect mutual-synchronization coefficient is obtained; Obtain the number of intermediate mediating genes between the current gene feature and each indirect-gene feature, and calculate the difference between the number of intermediate mediating genes of each indirect-gene feature and the maximum number of intermediate mediating genes; The indirect mutual-reflection matrix of the current gene feature is constructed using the corresponding indirect mutual-synchronization coefficients of all indirect-gene features and the difference in the number of intermediate mediating genes, and the indirect mutual-reflection coefficients are obtained through matrix calculation; The direct mutual reflection coefficient, indirect mutual reflection coefficient, number of direct-gene features and number of indirect-gene features are combined and calculated to obtain the comprehensive mutual reflection score of the current gene feature; For the current gene feature, the mutual influence weight is assigned according to the total number of corresponding gene features with interactions; The mutual comprehensive reflection score of the current gene characteristics and the mutual influence weight are combined and calculated to obtain the interaction influence coefficient; For all gene features in the current cluster group, the first feature arrangement table is constructed from large to small using the interaction influence coefficient; For all gene features in the current cluster group, the weights obtained by comprehensively considering the average sparse observation value vector ratio of the cluster group, the main base composition characteristics, and the factors involved in the gene function will be marked as pending weights; For all gene features in the current cluster group, a second feature arrangement table is constructed from large to small using the undetermined weights; If the first feature ranking table and the second feature ranking table are consistent, the undetermined weight of each gene feature is output as the corresponding importance weight; If the first feature ranking table and the second feature ranking table are inconsistent, the weight value obtained by adjusting the predetermined weight using the interaction influence coefficient will be output as the importance weight of the corresponding gene feature.

8. The method for extracting gene features from gene data based on sequence search according to claim 1, wherein: The feature vectors constructed for each gene feature region are subjected to feature fusion processing, which also includes: Determining statistics of the feature vectors of each gene feature region based on the feature vectors of each gene feature region, and arranging the calculated statistics in order to form a quantitative representation of the local features of the gene feature region; Perform statistical analysis on the entire gene data set, calculate the global eigenvalue based on the statistical analysis results, integrate the calculated global eigenvalues ​​to obtain the global feature representation of the gene data set, and fuse the local eigenvectors with the global feature representation.

9. The method for extracting gene features from gene data based on sequence search according to claim 8, characterized in that: Also includes: Construct a test data set, and separate the test data set from the original gene data set according to the preset division rules, wherein the real feature information of each gene in the test data set is obtained; The overall accuracy of the feature extraction method was determined by calculating the ratio of the number of correctly predicted gene features to the total number of gene features, and the coverage of the real gene features by the feature extraction method was evaluated by calculating the ratio of the number of correctly predicted gene features to the number of actual gene features. The performance of the gene feature extraction method was evaluated based on the overall accuracy and recall evaluation indicators, and the gene feature extraction method was optimized based on the evaluation results.

Images