Potential associated gene mining system based on medical knowledge base analysis and processing method thereof

Abstract

The invention provides a potential gene mining system based on medical knowledge base analysis and a processing method thereof, and belongs to the field of artificial intelligence bioinformatics. The associated gene mining system comprises a medical knowledge base construction module, a gene-text clustering engine module, a gene network construction module and a potential genome mining module. The associated gene mining system can perform unsupervised text clustering on genes through medical text literature data, then calculates the similarity of gene sequences according to a clustered result and preprocessed data so as to obtain weights of edges between the genes, constructs a gene network according to the weights, mines genomes highly related to diseases, and improves the disease tolerance of the genes. The mining process is efficient and high in correlation.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence bioinformatics, and is a potential associated gene mining system based on medical knowledge base analysis and a processing method thereof. Background Art

[0002] Gene correlation mining is an important research direction in the field of bioinformatics. Systemic lupus erythematosus (SLE) and antiphospholipid antibody syndrome (APS) are complex multi-gene autoimmune diseases. This paper uses the above two autoimmune diseases as examples to mine the potential correlation and interaction between pathogenic genes, providing important clues for the discovery of disease treatment targets. Medical text literature data, gene expression data and gene sequence data are used as data sources for processing, gene networks are constructed and dense subgraph mining is performed to find potential related gene sets.

[0003] At present, gene correlation mining methods are mainly divided into two categories. One is the correlation analysis method based on statistical methods. Its core is to calculate the correlation between genes and screen out gene pairs with significant correlation by statistical processing of genomic data, so as to mine genes related to diseases. The other is based on gene network analysis methods. According to the correlation between genes and other biological associations, the network topology and pattern characteristics are analyzed to identify central nodes with biological significance, so as to mine genes related to diseases.

[0004] Using genomic data to conduct statistical correlation analysis, the limitations of the above gene correlation mining are: when the genomic data has high dimensions, the computational complexity increases, making analysis and mining more difficult; statistical methods can usually only capture the linear correlation between genes, but cannot accurately reflect the complex nonlinear relationship between genes. When dealing with large-scale genomic data, the computational complexity of gene network analysis is high; the incompleteness of genomic data leads to inaccurate edges in the network, affecting the quality of gene correlation mining. Therefore, it is difficult for existing gene correlation mining methods to achieve ideal results. Summary of the invention

[0005] The present invention aims to solve at least one of the problems of the prior art. To this end, the present invention proposes a potential associated gene mining system based on medical knowledge base analysis and a processing method thereof. By using medical text document data to perform unsupervised text clustering on genes, based on the clustering results and pre-processed data, the weights between genes are calculated, and a gene network is constructed, thereby mining the gene set related to the disease.

[0006] On the one hand, the present invention proposes a potential associated gene mining system based on medical knowledge base analysis. According to an embodiment of the present application, the system includes: a medical knowledge base construction module, which is used to collect raw data and perform preprocessing; a text clustering engine module, which constructs a gene and text library based on the preprocessed data and performs unsupervised text clustering; a network construction module, which calculates the weight of the edge between genes based on gene text clustering and preprocessed data to complete the gene network construction module; a potential associated genome mining module, which performs subgraph mining based on the graph data after the gene network is constructed, calculates the average degree of the mined subgraph, and finds the potential associated genomes related to disease I and disease H.

[0007] According to an embodiment of the present invention, the aforementioned system may further include at least one of the following technical features:

[0008] According to an embodiment of the present invention, the aforementioned medical knowledge base construction module includes a data acquisition device and a data preprocessing device.

[0009] The data acquisition device is used to acquire raw data, including: (1) medical text data A, including disease I medical literature text A1 and disease H medical literature text A2, where A = {A1, A2}; (2) gene name data GE, including disease I gene name data GE1 and disease H gene name data GE2, where GE = {GE1, GE2}; (3) gene sequence data GE sequence , including disease I gene sequence data GE sequence1 , disease H gene sequence data GE sequence2 , GE sequence = {GE sequence1 , GE sequence2}.

[0010] The data preprocessing device is used to preprocess the medical text data, including: (1) dividing the medical literature text data A into an article set a according to the article, where the document in a is a word sequence <w1,w2,...,w i , w i+1 , ..., w n >, where w i Represents the i-th word; Use the gene name data GE to map the gene data of document a, delete the words that are not in the gene name data, and reconstruct the gene word sequence in the document as <gw1,gw2,...,gw i , gw i+1 , ..., gw n >, where gwi represents the i-th gene word; (2) construct a set LCOG, which contains the different gene name words obtained after all document mapping.

[0011] According to an embodiment of the present invention, the gene-text clustering engine module includes: (1) a topic modeling device, which sets the number of cluster topics C, that is, the number of topics expected to be obtained, adjusts the number according to the clustering effect, trains the model, and obtains the "document-topic" distribution and the "topic-gene word" distribution; (2) a document-topic probability calculation device, which is used to calculate for each document a, for different cluster topics c i , calculate the probability that document a belongs to the topic, O ci Indicates that it is assigned to topic c i The number of documents, O total represents the total number of documents in the training data, P(c i ) indicates the subject c i The prior probability of P(c i |a) means that the given document belongs to topic c under condition a i The probability of the document appearance is: P(a) represents the document edge probability. Given a document set A, the probability of the document appearance is i To sum, use the formula Calculate the document-topic probability distribution using the formula (3) A subject-gene word probability calculation device, which is used to calculate the probability of subject c i and the gene word gw i , in topic c i The probability in is, using the formula Count(gw i , c i ) is the subject c i The frequency of the i-th gene word in the document is used to obtain the topic probability corresponding to the document and the gene word probability corresponding to the topic. The document is classified into topics and the correlation analysis of gene words is performed to complete unsupervised text clustering.

[0012] According to an embodiment of the present invention, the gene network construction module includes: (1) based on the clustering result of the unsupervised text clustering module, converting the clustering result into gene network graph data, where N represents the node set, the connection between genes is represented by the edge set E in the graph, and CS ij represents the comprehensive score of the edge between node i and node j, and CS represents the set of comprehensive scores of all edges; (2) Based on the gene sequence data GE sequence , calculate SS ij Similarity score, Q represents the length of gene sequence S1, S1 = {s 1，1 ,s 1，2 , ..., s 1，Q}; M represents the length of gene sequence S2, S2 = {s 1，1 ,s 1，2 , ..., s1，M}, L ij represents the matching score, D represents the mismatching score, G open represents the open gap score, G extend Represents the extended gap score, initializes the alignment score matrix F and direction matrix D, F ij Indicated by s 1，i and 2，j is the local comparison score at the end; D ij Indicates the result when the gap is opened at the current position, recursive alignment score matrix and direction matrix, F ij =max(0, F i-1，j-1 +L ij , F i-1，j +Gextend,F i，j-1 +G extend , D ij );D ij =max(0, F i-1，j +G open , F i，j-1 +G open ), during the recursive process, if the score of a position is less than zero, set it to zero, find the maximum value maxScore in the comparison score matrix and its location (i max , j max ), the similarity score is the maximum score maxScore divided by the length of the longer gene sequence, using the formula (3) HS ij represents the fusion score of the edge between node i and node j, and HS represents all HS ij The calculated value is based on the formula HS ij =CS ij +SS ij , mean represents the average fusion score of all nodes in the clustering result, std represents the standard deviation of the fusion score of all nodes in the clustering result, θ can be adjusted according to the actual situation, and the threshold ρ is defined as mean-θ*std, θ=1, HS ij >ρ Calculate the fusion score of all clustering results that meet the threshold setting and construct the gene network graph model G = (N, E, HS).

[0013] According to an embodiment of the present invention, the potential genome mining module includes:

[0014] Recursively search for all maximum subgraphs: (1) According to the gene network construction module, G represents the input original graph data, which is stored in an adjacency list; N is the node in the graph, representing the gene; T represents the current candidate subgraph set, which contains all the nodes in the graph when initialized; R represents the subgraph set currently being constructed, which is initially empty; X represents the node set that cannot be expanded to the maximum subgraph, which is initially empty; Maxgraphs is used to store the list of found maximum subgraphs; (2) For all nodes in the graph, sort them in descending order of degree, optimize the subgraph construction process, and recursively construct subgraphs: iterate the sorted nodes N, add N to the currently constructed subgraph R, and remove it from the candidate set T, check whether the current R constitutes the maximum subgraph, the maximum subgraph judgment condition, partial adjacent node check, only check whether the newly added node is adjacent to some of the nodes in the existing subgraph, when the ratio of the newly added node to the existing subgraph node is greater than or equal to 0.8, that is, the condition is met, if it is the maximum subgraph, add R to Maxgraph. graphs In the above example, for node N in T, remove N from T and add it to X, calculate the new T, R, and X, recurse the previous steps, remove N from X after the recursion, add N back to T, and remove N from R after the recursion ends; (3) The recursion ends when the candidate subgraph set T is empty. Finally, Max graphs contains a list of all maximal subgraphs found.

[0015] Related potential gene subgraph edge density calculation device: (1) n represents Max graphs The number of nodes in the subgraph, calculate the number of edges num in all subgraphs edges , using the formula Actual number of edges edges ,edge i，j represents the edge formed by the i-th and j-th nodes in the subgraph. The delta function indicates that if this edge is in the edge set of the graph, the formula is used. The value of the indicator function is 1, otherwise it is 0; (2) subgraph edge density density edges , using the formula When density edges >0.8, the more potential correlation between the subgraph genes and the disease, the potential gene set associated with the disease is finally output.

[0016] According to an embodiment of the present invention, the aforementioned disease I and disease H are independently selected from SLE and / or APS.

[0017] On the other hand, the present invention proposes a method for processing potential associated genes based on medical knowledge base analysis, which is characterized by including: collecting original data and preprocessing it; constructing a gene and text library based on the preprocessed data, and performing unsupervised text clustering; calculating the weights of the edges between genes based on gene text clustering and the preprocessed data, and completing the gene network construction module; performing subgraph mining based on the graph data after the gene network is constructed, calculating the average degree of the mined subgraph, and finding the potential associated gene set related to disease I and disease H.

[0018] It should be understood that the system embodiment and the method embodiment can correspond to each other, and similar descriptions can refer to the system embodiment. To avoid repetition, it will not be repeated here. Specifically, the above system can execute the embodiment of the potential associated gene processing method based on medical knowledge base analysis, and the description of each module in the system corresponds to the method embodiment, which will not be repeated here for the sake of brevity.

[0019] The potential gene mining system based on medical knowledge base analysis and the processing method thereof of the present invention have the following advantages:

[0020] (1) The potential gene mining system and processing method based on medical knowledge base analysis of the present invention mainly performs cluster analysis based on public medical text documents and gene data. It does not require prior knowledge of gene classification information, reduces prior information bias, clusters similar genes together, and can reveal potential patterns and associations between genes.

[0021] (2) The present invention utilizes the calculation of gene sequence similarity and combines the comprehensive scores between genes to design a reasonable threshold, which has high flexibility. The weights between genes are complete and do not contain missing values, overcoming the inaccuracy of network edges and providing a good and accurate data form for the construction of gene networks.

[0022] (3) The present invention constructs a gene network through the results of gene clustering analysis and gene sequence similarity calculation, and uses graph theory algorithms to mine dense subgraphs therein, which can discover functionally related gene modules and identify gene groups with similar functions, which helps to reveal the functional patterns between genes.

[0023] (4) In the system of the present invention, the subgraph edge density is calculated for the dense subgraph mined by the graph theory algorithm, and the genome with the strongest potential correlation with the disease in the group of subgraphs is screened and output.

[0024] (5) The method is scientific and reasonable, and has strong applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram of the process of the potential associated gene mining system based on medical knowledge base analysis of the present invention;

[0026] Figure 2 This is a structural block diagram of a potential associated gene mining system based on medical knowledge base analysis of the present invention;

[0027] Figure 3 The present invention is a flowchart of a potential associated gene mining system based on medical knowledge base analysis. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail by specific embodiments in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0029] In a specific embodiment of the present invention, a potential associated gene mining system based on medical knowledge base analysis is provided to mine potential associated gene sets related to autoimmune system diseases SLE and APS. Hereinafter, disease I and disease H are used to refer to autoimmune system diseases SLE and APS. Figure 1 and Figure 2 As shown, the potential associated gene mining system based on medical knowledge base analysis includes: a medical knowledge base construction module, a gene text clustering engine module, a gene network construction module, and a potential genome mining module.

[0030] In Example 1, the medical knowledge base construction module is used to collect medical literature text data, gene name data, gene sequence data, and preprocess the collected data. The module includes a data collection device and a data preprocessing device.

[0031] The data collected by the data collection device:

[0032] Medical text data A includes disease I medical literature text A1 and disease H medical literature text A2, where A = {A1, A2}.

[0033] The gene name data GE include disease I gene name data GE1 and disease H gene name data GE2, GE={GE1, GE2}.

[0034] Gene sequence data sequence , including disease I gene sequence data GE sequence1 , Disease H gene sequence data GE sequence2 , GE sequence = {GE sequence1 , GE sequence2}.

[0035] Define the attributes of the collected data and create an attribute list 1 according to the data source type.

[0036] Property List 1

[0037] factor property Disease I Medical Texts <![CDATA[A1]]> Disease H Medical Text Literature <![CDATA[A2]]> Disease I gene name data <![CDATA[GE1]]> Disease H gene name data <![CDATA[GE2]]> Disease I Gene Sequence Data <![CDATA[GE sequence1 <!-- 4 -->]]> Disease H gene sequence data <![CDATA[GE sequence2 ]]>

[0038] The data preprocessing device in this module preprocesses the medical literature text and gene name data. Specifically:

[0039] (1) The medical literature text data A is divided into article set a by article, where each document in a is a word sequence <w1,w2,...,w i , w i+1 , ..., w n >, where w i Represents the i-th word.

[0040] (2) Use the gene name data GE to map the gene data of each document a, delete the words that are not in the gene name data, and reconstruct the gene word sequence in each document as follows: <gw1,gw2,...,gw i , gw i+1 , ..., gw n >.

[0041] (3) Construct a set LCOG, which contains the different gene name words obtained after all document mapping.

[0042] The main function of the gene text clustering engine module is to map gene name data with document data based on preprocessed data and reconstruct document gene word sequence. The gene text clustering engine is constructed, including: a topic modeling device, a document-topic probability calculation device and a topic-gene word probability calculation device.

[0043] The topic modeling device is expected to obtain the number of topics after clustering, specifically: set the number of cluster topics C, and adjust it according to the clustering effect. Train the model to obtain the "document-topic" distribution and the "topic-gene word" distribution.

[0044] The document-topic probability calculation device is used to calculate for each document a, for different cluster topics c i , calculate the probability that document a belongs to the topic, specifically:

[0045] (1)O ci Indicates that it is assigned to topic c i The number of documents, O total Represents the total number of documents in the training data. P(c i ) indicates the subject c i The prior probability of

[0046] (2)P(c i|a) means that the given document belongs to topic c under condition a i The probability of P(a) is the document edge probability. Given a document set A, the probability of any document appearing for all topics c i To sum, use the formula

[0047] (3) Calculate the document-topic probability distribution using the formula

[0048] The subject-gene word probability calculation device is used to calculate for each subject c i and each gene word gw i , in topic c i The probability in is:

[0049] The probability of the topic gene word is calculated using the formula Count(gw i , c i ) is the subject c i The frequency of the i-th gene word in . Using the topic probability corresponding to the document and the gene word probability corresponding to the topic, the document is classified into topics and the correlation analysis of gene words is performed, and the gene text clustering result is output.

[0050] The main function of the gene network construction module is to convert the clustering results output by the gene text clustering engine module into gene network graph data, conduct comprehensive analysis and calculation of gene edge weights, and construct a gene network graph model. Specifically:

[0051] (1) Based on the clustering results of the unsupervised text clustering module, the clustering results are converted into gene network graph data. N represents the node set, and the connection between genes is represented by the edge set E in the graph, CS ij It represents the comprehensive score of the edge between node i and node j, and CS represents the set of comprehensive scores of all edges.

[0052] (2) Based on gene sequence data GE sequence , calculate SS ij Similarity score. Q represents the length of gene sequence S1, S1 = {s 1，1 ,s 1，2 , ..., s 1，Q}; M represents the length of gene sequence S2, S2 = {s 1，1 ,s 1，2 , ..., s 1，M}. ij represents the matching score, D represents the mismatching score, G open represents the open gap score, G extendRepresents the expansion gap score. Initialize the alignment score matrix F and direction matrix D, F ij Indicated by s 1，i and 2，j is the local comparison score at the end; D ij Represents the score when the gap is opened at the current position. Recursive alignment score matrix and direction matrix, F ij =max(0, F i-1，j-1 +L ij , F i-1，j +G extend , F i，j-1 +G extend , D ij );D ij =max(0, F i-1，j +G open , F i，j-1 +G open ), during the recursive process, if the score of a position is less than zero, set it to zero. Find the maximum value maxScore and its position in the comparison score matrix (i max , j max ). The similarity score is the maximum score maxScore divided by the length of the longer gene sequence, using the formula

[0053] (3) HS ij Represents the fusion score of the edge between node i and node j, using the formula HS ij =CS ij +SS ij Mean represents the average fusion score of all nodes in the clustering result, std represents the standard deviation of the fusion score of all nodes in the clustering result, and the threshold value ρ = mean-θ*std, θ = 1, HS ij >ρ. θ can be adjusted according to the actual situation. Calculate the fusion score of all clustering results that meet the threshold setting. Construct the gene network graph model G = (N, E, HS).

[0054] The main function of the potential associated genome mining module is to perform graph theory calculations based on preprocessed gene data, gene clustering results and gene network graph data, and output potential genomes related to the disease, including a recursive search for all maximum subgraphs and a related potential gene subgraph edge density calculation device.

[0055] The recursive searching device for all the largest subgraphs finds a list of all the largest subgraphs based on the gene network graph data. Specifically:

[0056] (1) According to the gene network construction module, G represents the input raw graph data, which is stored in an adjacency list; N is the node in the graph, representing the gene; T represents the current candidate subgraph set, which contains all the nodes in the graph when initialized; R represents the subgraph set currently being constructed, which is initially empty; X represents the node set that cannot be expanded to the maximum subgraph, which is initially empty; Maxgraphs is used to store the list of found maximum subgraphs.

[0057] (2) Sort all nodes in the graph in descending order of degree and optimize the subgraph construction process. Recursively construct the subgraph: Iterate each sorted node N. Add N to the currently constructed subgraph R and remove it from the candidate set T. Check whether the current R constitutes the largest subgraph. The maximum subgraph judgment condition, partial adjacent node check, only check whether the newly added node is adjacent to some nodes in the existing subgraph. When the ratio of the newly added node to the existing subgraph node is greater than or equal to 0.8, the condition is met. If it is the largest subgraph, add R to the Max. graphs For each node N in T, remove N from T and add it to X, calculate the new T, R, and X, and recurse the previous step. After the recursion, remove N from X and add N back to T. After the recursion ends, remove N from R. The recursion ends when the candidate subgraph set T is empty. Finally, Max graphs contains a list of all maximal subgraphs found.

[0058] The related potential gene subgraph density calculation device calculates the subgraph edge density based on the recursively found list of all maximum subgraphs, and outputs the potential gene group related to the disease. Specifically:

[0059] (1) n represents Max graphs The number of nodes in the subgraph, calculate the number of edges in all subgraphs num edges , using the formula Actual number of edges edges , using the formula edge i，j Represents the edge formed by the i-th and j-th nodes in the subgraph. The delta function indicates that if this edge is in the edge set of the graph, the value of the indicator function is 1, otherwise it is 0.

[0060] (2) Subgraph edge density edges , using the formula When density edges >0.8, the more potential correlation between the subgraph genes and the disease. The final output is the potential gene set associated with the disease.

[0061] The above describes the system of the embodiment of the present application from the perspective of functional modules in conjunction with the accompanying drawings. It should be understood that the functional modules can be implemented in hardware form, can also be implemented by software instructions, and can also be implemented by a combination of hardware and software modules.

[0062] In another embodiment of the present invention, a method for mining potential gene correlation based on SLE and APS is provided, referring to Figure 3 As shown in the figure, disease I and disease H are used to refer to SLE and APS. The specific method is as follows:

[0063] Step 1: Classify the collected data. Medical text data A, including disease I medical literature text A1, disease H medical literature text A2, A = {A1, A2}. Gene name data GE, including disease I gene name data GE1, disease H gene name data GE2, GE = {GE1, GE2}. Gene sequence data GE sequence , including disease I gene sequence data GE sequence1 , Disease H gene sequence data GE sequence2 , GE sequence = {GE sequence1 , GE sequence2}.

[0064] Step 2: Preprocess different data. Divide the medical text data A into article sets a by article, where each document a is a word sequence <w1,w2,...,w i , w i+1 , ..., w n >, where w i Represents the i-th word. Use the gene name data GE to map the gene data of each document a, delete the words that are not in the gene name data, and reconstruct the gene word sequence in each document as <gw1,gw2,...,gw i , gw i+ 1, ..., gw n >. Construct a set LCOG, which contains the different gene name words obtained after all documents are mapped.

[0065] Step 3: Perform unsupervised text clustering. Topic modeling, set the number of cluster topics, and adjust according to the clustering effect. Calculate the document-topic probability, calculate for each document a, for different cluster topics c i , calculate the probability that document a belongs to the topic. ci Indicates that it is assigned to topic c i Number of documents in Otota l Represents the total number of documents in the training data. P(c i ) indicates the subject c i The prior probability of P(c i |a) means that the given document belongs to topic c under condition a i The probability of the document edge probability is: given the document set A, the probability of any document appearing for the topic c i To sum, use the formula Calculate the document-topic probability distribution using the formula Calculate the topic-gene word probability, and calculate for each topic c i and each gene word gw i , in topic c i The probability in is, using the formula Count(gw i , c i ) is the subject c i The frequency of the i-th gene word in . Using the topic probability corresponding to the document and the gene word probability corresponding to the topic, the document is classified by topic and the correlation analysis of gene words is performed, and the clustered genome data is output.

[0066] Step 4: Gene network construction. Based on the clustering results, the clustering results are converted into gene network data. N represents the node set, and the connection between genes is represented by the edge set E in the graph, CS ij represents the comprehensive score of the edge between node i and node j, and CS represents the set of comprehensive scores of all edges. sequence , calculate SS ij Similarity score. Q represents the length of gene sequence S1, S1 = {s 1，1 ,s 1，2 , ..., s 1，Q}; M represents the length of gene sequence S2, S2 = {s 1，1 ,s 1，2 , ..., s 1，M}. ij represents the matching score, D represents the mismatching score, G open represents the open gap score, G extend Represents the expansion gap score. Initialize the alignment score matrix F and direction matrix D, F ij Indicated by s 1，i and 2，j is the local comparison score at the end; D ij Represents the score when the gap is opened at the current position. Recursive alignment score matrix and direction matrix, F ij =max(0, F i-1，j-1 +L ij , F i-1，j +G extend , F i，j-1 +Gextend , D ij );D ij =max(0, F i-1，j +G open , F i，j -1+G open ), during the recursive process, if the score of a position is less than zero, set it to zero. Find the maximum value maxScore and its position in the comparison score matrix (i max , j max ). The similarity score is the maximum score maxScore divided by the length of the longer gene sequence, using the formula HSij represents the fusion score of the edge between node i and node j, using the formula HS ij =CS ij +SS ij Mean represents the average fusion score of all nodes in the clustering result, std represents the standard deviation of the fusion score of all nodes in the clustering result, and the threshold value ρ = mean-θ*std, θ = 1, HS ij >ρ. θ can be adjusted according to the actual situation. Calculate the fusion score of all clustering results that meet the threshold setting. Construct the gene network graph model G = (N, E, HS).

[0067] Step 5: Potentially associated genome mining. Recursively search for all maximum subgraphs and build modules according to the gene network, where G represents the input original graph data, stored in an adjacency list; N is the node in the graph, representing the gene; T represents the current candidate subgraph set, which contains all the nodes in the graph when initialized; R represents the subgraph set currently being built, which is initially empty; X represents the node set that cannot be expanded to the maximum subgraph, which is initially empty; Max graphs Used to store the list of the largest subgraph found. Sort all nodes in the graph in descending order of degree to optimize the subgraph construction process. Recursively construct subgraphs: iterate each sorted node N. Add N to the currently constructed subgraph R and remove it from the candidate set T to check whether the current R constitutes the largest subgraph. Maximum subgraph judgment condition, partial adjacent node check, only check whether the newly added node is adjacent to some nodes in the existing subgraph. When the ratio of the newly added node to the existing subgraph node is greater than or equal to 0.8, the condition is met. If it is the largest subgraph, add R to Max graphs For each node N in T, remove N from T and add it to X, calculate the new T, R, and X, and recurse the previous step. After the recursion, remove N from X and add N back to T. After the recursion ends, remove N from R. The recursion ends when the candidate subgraph set T is empty. Finally, Max graphs contains a list of all maximal subgraphs found.

[0068] Step 6: Calculate the gene subgraph density. n represents Max graphs The number of nodes in the subgraph, calculate the number of edges in all subgraphs num edges , using the formula Actual number of edges edges , using the formula edge i，j Indicates the edge formed by the i-th and j-th nodes in the subgraph. The delta function indicates that if this edge is in the edge set of the graph, the value of the indicator function is 1, otherwise it is 0. Subgraph edge density density edges , using the formula

[0069] Step 7: Output potential gene sets associated with the disease. edges >0.8, the more potential correlation between the subgraph genes and the disease. The final output is the potential gene set associated with the disease.

[0070] In summary, the system of the present invention can realize the mining of potential gene correlations of autoimmune diseases SLE and APS, which helps to reveal the potential correlations between diseases and genes and provide important clues for the discovery of therapeutic targets for diseases.

[0071] The above embodiments are merely examples for the purpose of clarifying the present invention, and are not intended to limit the embodiments. For those skilled in the art, other different forms of changes or modifications may be made based on the above description. It is not necessary and impossible to list all the embodiments here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. A potential associated gene mining system based on medical knowledge base analysis, characterized in that: The system comprises: Medical knowledge base construction module, used to collect raw data and perform preprocessing; The gene-text clustering engine module builds gene and text libraries based on preprocessed data and performs unsupervised text clustering; Gene network construction module, based on gene text clustering and preprocessed data, calculates the weights of the edges between genes and completes the gene network construction module; The potential associated genome mining module performs subgraph mining based on the graph data after the gene network is constructed, calculates the average degree of the mined subgraph, and finds the potential associated genomes related to disease I and disease H.

2. The potential associated gene mining system according to claim 1, characterized in that: The medical knowledge base construction module includes a data acquisition device and a data preprocessing device; Wherein, the data acquisition device is used to collect raw data, including: (1) Medical text data A, including disease I medical literature text A1, disease H medical literature text A2, A = {A1, A2}; (2) Gene name data GE, including disease I gene name data GE1, disease H gene name data GE2, GE = {GE1, GE2}; (3) Gene sequence data GE sequence , including disease I gene sequence data GE sequence1 , Disease H gene sequence data GE sequence2 , GE sequence = {GE sequence1 , GE sequence2 }; The data preprocessing device is used to preprocess the medical text data, including: (1) The medical literature text data A is divided into article set a by article. The document in a is a word sequence. <w1,w2,...,w i , w i+1 , ..., w n >, where w i Represents the i-th word; Use the gene name data GE to map the gene data of document a, delete the words that are not in the gene name data, and reconstruct the gene word sequence in the document as <gw1,gw2,...,gw i , gw i+1 , ..., gw n >, where gwi represents the i-th gene word; (2) Construct a set LCOG, which contains the different gene name words obtained after all document mapping.

3. The potential associated gene mining system according to claim 1, characterized in that: The gene-text clustering engine module includes: (1) Topic modeling device, set the number of cluster topics C, that is, the number of topics expected to be obtained, adjust it according to the clustering effect, train the model, and obtain the "document-topic" distribution and "topic-gene word" distribution; (2) A document-topic probability calculation device, which is used to calculate for each document a, for different cluster topics c i , calculate the probability that document a belongs to the topic, O ci Indicates that it is assigned to topic c i The number of documents, O total represents the total number of documents in the training data, P(c i ) indicates the subject c i The prior probability of P(c i |a) means that the given document belongs to topic c under condition a i The probability of the document appearance is: P(a) represents the document edge probability. Given a document set A, the probability of the document appearance is i To sum, use the formula Calculate the document-topic probability distribution using the formula (3) A subject-gene word probability calculation device, which is used to calculate the probability of subject c i and the gene word gw i , in topic c i The probability in is, using the formula Count(gw i ,c i ) is the subject c i The frequency of the i-th gene word in the document is used to obtain the topic probability corresponding to the document and the gene word probability corresponding to the topic. The document is classified into topics and the correlation analysis of gene words is performed to complete unsupervised text clustering.

4. The potential associated gene mining system according to claim 1, characterized in that: The gene network building module includes: (1) Based on the clustering results of the unsupervised text clustering module, the clustering results are converted into gene network graph data, where N represents the node set, and the connection between genes is represented by the edge set E in the graph, CS ij represents the comprehensive score of the edge between node i and node j, and CS represents the set of comprehensive scores of all edges; (2) Based on gene sequence data GE sequence , calculate SS ij Similarity score, Q represents the length of gene sequence S1, S1 = {s 1,1 ,s 1,2 , ..., s 1,Q }; M represents the length of gene sequence S2, S2 = {s 1,1 ,s 1,2 , ..., s 1,M }, L ij represents the matching score, D represents the mismatching score, G open represents the open gap score, G extend Represents the extended gap score, initializes the alignment score matrix F and direction matrix D, F ij Indicated by s 1,i and 2,j is the local comparison score at the end; D ij Indicates the result when the gap is opened at the current position, recursive alignment score matrix and direction matrix, F ij =max(0,F i-1,j-1 +L ij ,F i-1,j +G extend ,F i,j-1 +G extend ,D ij );D ij =max(0,F i-1,j +G open ,F i,n-1 +G open ), during the recursive process, if the score of a position is less than zero, set it to zero, find the maximum value maxScore in the comparison score matrix and its location (i max , j max ), the similarity score is the maximum score maxScore divided by the length of the longer gene sequence, using the formula (3) HS ij represents the fusion score of the edge between node i and node j, and HS represents all HS ij The calculated value is based on the formula HS ij =CS ij +SS ij , mean represents the average fusion score of all nodes in the clustering result, std represents the standard deviation of the fusion score of all nodes in the clustering result, θ can be adjusted according to the actual situation, and the threshold ρ is defined as mean-θ*std, θ=1, HS ij >ρCalculate the fusion score of all clustering results that meet the threshold setting and construct the gene network graph model G=(N,E,HS).

5. The potential associated gene mining system according to claim 1, characterized in that: The potential genome mining module includes: Recursively find all maximal subgraph fixtures: (1) According to the gene network construction module, G represents the input raw graph data, which is stored in an adjacency list; N is the node in the graph, representing the gene; T represents the set of current candidate subgraphs, which contains all the nodes in the graph when initialized; R represents the subgraph set currently being constructed, which is initially empty; X represents the set of nodes that cannot be expanded to the maximum subgraph, which is initially empty; Max graphs A list used to store the largest subgraphs found; (2) For all nodes in the graph, sort them in descending order of degree, optimize the subgraph construction process, and recursively construct the subgraph: iterate the sorted nodes N, add N to the currently constructed subgraph R, and remove it from the candidate set T, check whether the current R constitutes the largest subgraph, and the maximum subgraph judgment condition, partial adjacent node check, only check whether the newly added node is adjacent to some nodes in the existing subgraph. When the ratio of the newly added node to the existing subgraph node is greater than or equal to 0.8, the condition is met. If it is the largest subgraph, add R to the Max. graphs In the above example, for node N in T, remove N from T and add it to X, calculate the new T, R, and X, recurse the previous step, after the recursion, remove N from X, add N back to T, and after the recursion ends, remove N from R; (3) Recursive end condition: When the candidate subgraph set T is empty, the recursion ends and Max graphs contains a list of all maximal subgraphs found; Related potential gene subgraph edge density calculation device: (1) n represents Max graphs The number of nodes in the subgraph, calculate the number of edges num in all subgraphs edges , using the formula Actual number of edges edges ,edge i,j represents the edge formed by the i-th and j-th nodes in the subgraph. The delta function indicates that if this edge is in the edge set of the graph, the formula is used. The value of the indicator function is 1, otherwise it is 0; (2) Subgraph edge density edges , using the formula When density edges >0.8, the potential correlation between the subgraph genes and the disease is stronger, and the potential gene set associated with the disease is finally output.

6. A method for processing potential associated genes based on medical knowledge base analysis, characterized in that: include: Collect raw data and pre-process them; Build gene and text libraries based on preprocessed data and perform unsupervised text clustering; Based on gene text clustering and preprocessed data, the weights of the edges between genes are calculated to complete the gene network construction module; Based on the graph data after the gene network is constructed, subgraph mining is performed, the average degree of the mined subgraph is calculated, and the potential associated gene groups related to disease I and disease H are found.

Images