Pathogenic gene identification method and system based on multi-scale modular kernel

Abstract

In the disclosed embodiment, a method and system for identifying pathogenic genes based on a multi-scale module core are provided, which belongs to the field of bioinformatics technology, and specifically includes: extracting network multi-scale modules from a comprehensive protein interaction group through an exponentially sampled multi-scale module optimization algorithm to obtain a standardized multi-scale module spectrum; constructing a multi-scale module core matrix based on the standardized multi-scale module spectrum and performing core sparseness; extracting two prediction lists; fusing the two prediction lists to obtain a gene probability list, and predicting the pathogenic genes of the disease based on the gene probability list. Through the scheme disclosed in the present invention, the multi-scale module spectrum is extracted and standardized, and the network module structure information of different levels and specificities is effectively fused and then sparsely processed, providing a new gene-gene association, and then the ranking aggregation of the prediction lists from the original heterogeneous network and the heterogeneous network combined with the multi-scale module core effectively fuses the complementary prediction information, improving the prediction efficiency and prediction accuracy.

Description

Technical Field

[0001] The disclosed embodiments relate to the technical field of bioinformatics, and in particular to a method and system for identifying pathogenic genes based on a multi-scale modular kernel. Background Art

[0002] At present, the identification of disease-related genes is very important for the study of complex human diseases, which is the basis for understanding the molecular mechanism of the disease. Traditional methods have problems such as large candidate space and expensive and time-consuming experimental identification of candidate genes. Therefore, the development of disease-causing gene prediction methods has become an important topic in computational biology, which plays an important role in promoting the study of complex diseases.

[0003] Many computational methods have been proposed to predict disease-associated genes, and network-based methods are one of the most popular strategies for disease gene prediction. It is a natural and effective strategy to predict disease-associated genes through direct interactions, network distances, or proximity between candidate genes and seed genes (i.e., genes known to be associated with specific diseases). Network propagation methods such as random walks with restarts can more effectively explore the neighborhood topology around seeds in biological networks, and therefore have been widely used in disease gene prediction and related problems, and have also been extended to more complex network models, such as multi-graph models, multi-network models, and multiple network models in homogeneous or heterogeneous networks. These models can integrate multiple types of biological networks. However, due to problems such as data noise in biological networks, these methods still have room for further improvement.

[0004] Analysis of modular structures in biomolecular networks can provide useful insights into the study of complex human diseases. Multiscale modular structures (MMS) are widely present in biomolecular networks. These biomolecular networks are composed of sub-networks, which can be gradually decomposed into smaller sub-networks in a deep hierarchy. For example, some complexes may be composed of multiple sub-complexes, and the hierarchical structure of gene function annotations also means that there are functional modules at multiple scales. MMS can provide richer information from local to global for disease research, but how to mine the valuable information hidden in MMS remains a challenging topic for disease gene prediction.

[0005] It can be seen that there is an urgent need for an efficient and accurate method for identifying pathogenic genes based on multi-scale modular kernels. Summary of the invention

[0006] In view of this, the embodiments of the present disclosure provide a method and system for identifying pathogenic genes based on a multi-scale modular kernel, which at least partially solves the problem of poor pathogenic gene identification efficiency and recognition accuracy in the prior art.

[0007] In a first aspect, the present disclosure provides a method for identifying pathogenic genes based on a multi-scale modular kernel, comprising:

[0008] The network multi-scale modules were extracted from the comprehensive protein interaction group by using the multi-scale modularity optimization algorithm with exponential sampling, and the initial multi-scale module spectrum containing local and global information of the network was constructed. The initial multi-scale module spectrum was weighted by using the characteristic information method to obtain the standardized multi-scale module spectrum.

[0009] Constructing a multi-scale module kernel matrix according to the standardized multi-scale module spectrum and performing kernel sparseness to obtain a sparse matrix;

[0010] constructing two probability matrices according to the protein interactome, the disease-disease association and the disease-gene association, and according to the sparse matrix, the disease-disease association and the disease-gene association, respectively, and extracting two prediction lists from the two probability matrices through a walk dynamics process;

[0011] The two prediction lists are fused through a probability model based on node ranking aggregation to obtain a gene probability list, and the pathogenic gene of the disease is predicted based on the gene probability list.

[0012] According to a specific implementation of the embodiment of the present disclosure, the formula of the network multi-scale module is: Among them, e ss represents the edge ratio within module s, a s It represents the edge ratio in module s in the random model, γ is the resolution parameter, and the γ resolution parameter values ​​corresponding to different scales are obtained by exponential sampling;

[0013] The standardized multi-scale module spectrum: B = (B (1) ,B (2) ,…,B (h) ,…), where B (h) is the module partition matrix for each scale, the elements of which are Indicates whether gene i belongs to module s;

[0014] The formula for the weighting process is: Among them, the diagonal matrix D IC The diagonal elements (D IC ) ss =-log(∑0I(B is ) / N), I(x) represents the counting function, and N represents the number of nodes.

[0015] According to a specific implementation of the embodiment of the present disclosure, the step of constructing a multi-scale module kernel matrix according to the standardized multi-scale module spectrum and performing kernel sparseness to obtain a sparse matrix includes:

[0016] The standardized multi-scale module spectrum is used as a node feature spectrum and the multi-scale module kernel matrix is ​​obtained by calculating the inner product of the node feature spectrum vector;

[0017] The elements in the multi-scale module core matrix are standardized by a node degree-related standardization method so that the element values ​​are constrained to be between 0 and 1, thereby obtaining a standardized module core matrix;

[0018] A preset number of similar neighbor nodes corresponding to each node in the standardized module core matrix is ​​retained, and the standardized module core matrix is ​​subjected to core sparseness to obtain the sparse matrix.

[0019] According to a specific implementation of an embodiment of the present disclosure, the step of retaining a preset number of similar neighbor nodes corresponding to each node in the standardized module core matrix, performing core sparseness on the standardized module core matrix, and obtaining the sparse matrix includes:

[0020] Calculating the descending ranking values ​​of the similarity scores between each of the nodes and its neighboring nodes in the standardized module core matrix to obtain a ranking vector for each of the nodes;

[0021] Combining the ranking vectors of all the nodes to form a ranking matrix;

[0022] Calculate a new gene network according to the ranking matrix;

[0023] The sparse matrix is ​​obtained according to the gene network.

[0024] According to a specific implementation of the embodiment of the present disclosure, the steps of constructing two probability matrices respectively according to the protein interaction group, the disease-disease association and the disease-gene association, and according to the sparse matrix, the disease-disease association and the disease-gene association, and extracting two prediction lists from the two probability matrices through a walk dynamics process include:

[0025] establishing a first probability matrix based on the protein interactome, the disease-disease associations, and the disease-gene associations, and establishing a second probability matrix based on the sparse matrix, the disease-disease associations, and the disease-gene associations;

[0026] Normalizing the first probability matrix to obtain a first probability transfer matrix, and normalizing the second probability matrix to obtain a second probability transfer matrix;

[0027] The prediction scores corresponding to the first probability transfer matrix and the second probability transfer matrix are respectively generated according to the walk dynamics process to form two prediction lists.

[0028] According to a specific implementation of the embodiment of the present disclosure, the step of fusing the two prediction lists through a probability model based on node ranking aggregation to obtain a gene probability list, and predicting the pathogenic gene of the disease according to the gene probability list includes:

[0029] Arrange the two prediction lists in descending order to obtain two ranking lists;

[0030] Merging the two ranking lists through the probability model to obtain the gene probability list;

[0031] The gene probability list is sorted in descending order and the causative gene of the disease is predicted based on the sorting result.

[0032] In a second aspect, the present disclosure provides a pathogenic gene identification system based on a multi-scale module core, including:

[0033] An extraction module is used to extract network multiscale modules from the comprehensive protein interaction group through an exponential sampling multiscale modularity optimization algorithm, construct an initial multiscale module spectrum containing local and global information of the network, and perform weighted processing on it using a characteristic information method to obtain a standardized multiscale module spectrum;

[0034] A construction module is used to construct a multi-scale module kernel matrix according to the standardized multi-scale module spectrum and perform kernel sparseness to obtain a sparse matrix;

[0035] an association module, for constructing two probability matrices according to the protein interactome, the disease-disease association and the disease-gene association, and according to the sparse matrix, the disease-disease association and the disease-gene association, respectively, and extracting two prediction lists from the two probability matrices through a walk dynamics process;

[0036] The prediction module is used to fuse the two prediction lists through a probability model based on node ranking aggregation to obtain a gene probability list, and predict the pathogenic gene of the disease based on the gene probability list.

[0037] The pathogenic gene identification scheme based on multi-scale module kernel in the embodiment of the present disclosure includes: extracting network multi-scale modules from a comprehensive protein interaction group through an exponentially sampled multi-scale modularity optimization algorithm, constructing an initial multi-scale module spectrum containing local and global information of the network, and weighting it using a characteristic information method to obtain a standardized multi-scale module spectrum; constructing a multi-scale module kernel matrix according to the standardized multi-scale module spectrum and performing kernel sparseness to obtain a sparse matrix; constructing two probability matrices according to the protein interaction group, disease-disease association and disease-gene association, and according to the sparse matrix, the disease-disease association and the disease-gene association, respectively, and extracting two prediction lists from the two probability matrices through a walk dynamics process; fusing the two prediction lists through a probability model based on node ranking aggregation to obtain a gene probability list, and predicting the pathogenic gene of the disease according to the gene probability list.

[0038] The beneficial effects of the embodiments of the present disclosure are as follows: through the scheme of the present disclosure, the multi-scale module spectrum is extracted and standardized, and the multi-scale module core is constructed and sparsely processed after the network module structure information of different levels and specificities is effectively integrated, providing new gene-gene association information, and then the ranking aggregation of the prediction lists from the original heterogeneous network and the heterogeneous network combined with the multi-scale module core effectively integrates the complementary prediction information, thereby improving the prediction efficiency and prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0040] Figure 1 A schematic diagram of a process for identifying pathogenic genes based on a multi-scale modular kernel provided in an embodiment of the present disclosure;

[0041] Figure 2 A schematic diagram of a processing flow of a pathogenic gene identification method based on a multi-scale module kernel provided in an embodiment of the present disclosure;

[0042] Figure 3 A schematic diagram of the effect of the sparseness process of the multi-scale modular kernel provided in the embodiment of the present disclosure on the prediction performance of the pathogenic gene identification method based on the multi-scale modular kernel;

[0043] Figure 4 A schematic diagram of the effect of the sparseness process of the multi-scale modular kernel provided in the embodiment of the present disclosure on the time and space requirements of the pathogenic gene identification method based on the multi-scale modular kernel;

[0044] Figure 5 A comparison chart of the prediction performance of the pathogenic gene identification method based on the multi-scale modular kernel provided in the embodiment of the present disclosure;

[0045] Figure 6 A comparison chart of local prediction performance of the pathogenic gene identification method based on the multi-scale module kernel provided in the embodiment of the present disclosure;

[0046] Figure 7 A schematic diagram of the structure of a pathogenic gene identification system based on a multi-scale modular core provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0047] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0048] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.

[0049] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.

[0050] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The drawings only show components related to the present disclosure rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0051] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0052] The embodiments of the present disclosure provide a method for identifying pathogenic genes based on a multi-scale modular kernel, which can be applied to the process of identifying pathogenic genes in medical and health scenarios.

[0053] See also Figure 1 , is a flow chart of a method for identifying pathogenic genes based on a multi-scale module core provided by an embodiment of the present disclosure. Figure 1 As shown, the method mainly comprises the following steps:

[0054] S101, extracting network multiscale modules from the comprehensive protein interaction group through the multiscale modularity optimization algorithm of exponential sampling, constructing an initial multiscale module spectrum containing local and global information of the network, and weighting it using the characteristic information method to obtain a standardized multiscale module spectrum;

[0055] In the specific implementation, considering that the comprehensive protein interaction group contains information at multiple scales and levels, we can first extract the network multi-scale modules from the comprehensive protein interaction group through the exponential sampling multi-scale modularity optimization algorithm, and construct an initial multi-scale module spectrum containing local and global network information. The characteristic information method is then used to weight it and obtain a standardized multi-scale module spectrum.

[0056] Optionally, the formula of the network multi-scale module is: Among them, e ss represents the edge ratio within module s, a s It represents the edge ratio in module s in the random model, γ is the resolution parameter, and the γ resolution parameter values ​​corresponding to different scales are obtained by exponential sampling;

[0057] The standardized multi-scale module spectrum: B = (B (1) ,B (2) ,…,B (h) ,…), where B (h) is the module partition matrix for each scale, the elements of which are Indicates whether gene i belongs to module s;

[0058] The formula for the weighting process is: Among them, the diagonal matrix D IC The diagonal elements (D IC ) ss =-log(∑ i I(B is) / N), I(x) represents the counting function, and N represents the number of nodes.

[0059] For example, the multiscale modules of the protein interactome are extracted by the exponential sampling multiscale module optimization method, where the multiscale modularity function is defined as: e ss represents the edge ratio within module s, a s represents the edge ratio within module s in the random model, and γ is the resolution parameter; in the resolution parameter space, a discrete set of resolution parameters is obtained by exponential sampling; for each reasonable resolution parameter value, a set of network module partitions will be generated by optimizing the Q(γ) function (module partitioning represents the result of module segmentation of the network at this resolution), and the module partition matrix B is constructed (h) To describe the module division results of each scale, its elements Indicates whether gene i belongs to module s; integrate the module partition matrix to construct a multi-scale module spectrum containing local and global information of the network: B = (B (1) ,B (2) ,…,B (h) ,…), where B (h) is the module partition matrix for each scale.

[0060] Then, the multi-scale module spectrum is weighted and preprocessed by the feature relative information method to obtain a standardized multi-scale module spectrum. The calculation method is: The diagonal matrix D IC The diagonal elements (D IC ) ss =-log(∑ i I(B is ) / N), I(x) represents the counting function, N represents the number of nodes;

[0061] In the above steps, the exponential sampling multi-scale module optimization can better cover the network module structure of each scale. The multi-scale module spectrum and its standardized processing process can not only effectively integrate the network structure information at different levels but also take into account the specificity of modules at different scales, thereby providing an effective solution for the representation of multi-scale module information and providing an important foundation for the subsequent disease-causing gene identification capability.

[0062] S102, constructing a multi-scale module kernel matrix according to the standardized multi-scale module spectrum and performing kernel sparseness to obtain a sparse matrix;

[0063] In specific implementation, after obtaining the standardized multi-scale module spectrum, a multi-scale module kernel matrix can be constructed according to the standardized multi-scale module spectrum, which further condenses the information in the multi-scale module spectrum and performs kernel sparsification to obtain a sparse matrix, which provides a new gene-gene association type for subsequent disease-causing gene prediction and improves computational efficiency.

[0064] S103, constructing two probability matrices according to the protein interactome, the disease-disease association and the disease-gene association, and according to the sparse matrix, the disease-disease association and the disease-gene association, respectively, and extracting two prediction lists from the two probability matrices through a walk dynamics process;

[0065] In specific implementation, the protein interaction group, the disease-disease association and the disease-gene association can construct an original heterogeneous network, and a new heterogeneous network can be constructed based on the sparse matrix, the disease-disease association and the disease-gene association, and then two probability matrices are obtained based on two different heterogeneous networks, and two prediction lists are extracted from the two probability matrices through the walker dynamics process, and the two prediction lists provide complementary information for the effective prediction of pathogenic genes.

[0066] S104, fusing the two prediction lists through a probability model based on node ranking aggregation to obtain a gene probability list, and predicting the causative gene of the disease according to the gene probability list.

[0067] In specific implementation, the two prediction lists can be fused through a probability model based on node ranking aggregation to obtain a gene probability list, thereby comprehensively utilizing the complementary information of multiple prediction score lists about candidate genes, especially combining the prediction list information of the multi-scale module core, thereby effectively improving the prediction ability of disease-causing genes, and predicting the disease-causing genes based on the gene probability list.

[0068] The pathogenic gene identification method based on the multi-scale module core provided in this embodiment extracts and standardizes the multi-scale module spectrum, effectively integrates the network module structure information of different levels and specificities, constructs the multi-scale module core and performs sparseness, provides new gene-gene association information, and then the ranking aggregation of the prediction lists from the original heterogeneous network and the heterogeneous network combined with the multi-scale module core effectively integrates the complementary prediction information, thereby improving the prediction efficiency and prediction accuracy.

[0069] Furthermore, as described in step S102, constructing a multi-scale module kernel matrix according to the standardized multi-scale module spectrum and performing kernel sparseness to obtain a sparse matrix includes:

[0070] The standardized multi-scale module spectrum is used as a node feature spectrum and the multi-scale module kernel matrix is ​​obtained by calculating the inner product of the node feature spectrum vector;

[0071] The elements in the multi-scale module core matrix are standardized by a node degree-related standardization method so that the element values ​​are constrained to be between 0 and 1, thereby obtaining a standardized module core matrix;

[0072] A preset number of similar neighbor nodes corresponding to each node in the standardized module core matrix is ​​retained, and the standardized module core matrix is ​​subjected to core sparseness to obtain the sparse matrix.

[0073] For example, the standardized multi-scale module spectrum is used as the node feature spectrum The multi-scale module kernel matrix is ​​obtained by calculating the inner product of the node feature spectrum vector, and the calculation formula is:

[0074] The element values ​​of the multi-scale module kernel matrix are constrained to be between 0 and 1 through a node degree-related normalization method, and the calculation formula is: Where D NM is a diagonal matrix whose diagonal elements are defined as Then according to the standardized multi-scale module kernel The multi-scale module kernel is sparsely processed by retaining the most similar k neighbor nodes of each node to obtain the sparse matrix.

[0075] Furthermore, the step of retaining a preset number of similar neighbor nodes corresponding to each node in the standardized module core matrix, performing core sparseness on the standardized module core matrix, and obtaining the sparse matrix includes:

[0076] Calculating the descending ranking values ​​of the similarity scores between each of the nodes and its neighboring nodes in the standardized module core matrix to obtain a ranking vector for each of the nodes;

[0077] Combining the ranking vectors of all the nodes to form a ranking matrix;

[0078] Calculate a new gene network according to the ranking matrix;

[0079] The sparse matrix is ​​obtained according to the gene network.

[0080] Specifically, for each node, the descending ranking value of the similarity score between it and its neighboring nodes is calculated to obtain the ranking vector where r ij Represents the ranking value of the neighbor j of gene i (let gene i take the maximum ranking value for itself), and then combine the ranking vectors of all genes to get the ranking matrix Define a new gene network G = (V, E, W), where V, E and W represent the gene set, edge set and corresponding edge weight set respectively, and E = {(i, j) | R ij ≤korR ji ≤k}, and Then, according to the new gene network G = (V, E, W), the sparse matrix representation corresponding to the multi-scale module kernel is obtained: M MSMK , the element values ​​greater than zero in this matrix represent the edges and edge weights present in G.

[0081] The construction of the multi-scale module kernel and its standardized processing further condenses the information in the multi-scale module spectrum, providing a new gene-gene association type for subsequent disease-causing gene prediction. The sparse processing process can effectively reduce the time and space requirements of the multi-scale module kernel in subsequent calculations and improve computational efficiency.

[0082] Further, in step S103, according to the protein interaction group, the disease-disease association and the disease-gene association, and according to the sparse matrix, the disease-disease association and the disease-gene association, two probability matrices are respectively constructed, and two prediction lists are extracted from the two probability matrices through a walk dynamics process, including:

[0083] establishing a first probability matrix based on the protein interactome, the disease-disease associations, and the disease-gene associations, and establishing a second probability matrix based on the sparse matrix, the disease-disease associations, and the disease-gene associations;

[0084] Normalizing the first probability matrix to obtain a first probability transfer matrix, and normalizing the second probability matrix to obtain a second probability transfer matrix;

[0085] The prediction scores corresponding to the first probability transfer matrix and the second probability transfer matrix are respectively generated according to the walk dynamics process to form two prediction lists.

[0086] In a specific implementation, a first probability matrix is ​​established based on the protein interaction group, disease-disease association and disease-gene association. And, establishing a second probability matrix based on the sparse matrix, the disease-disease association and the disease-gene association Among them, T G is the column normalized matrix (i.e., probability transfer matrix) corresponding to the protein interaction group, T MSMK is the probability transfer matrix corresponding to the multi-scale module kernel sparse matrix representation, T D is the probability transfer matrix corresponding to the disease-disease association matrix, T GDis the probability transfer matrix corresponding to the gene-disease association matrix, T DG is the probability transfer matrix corresponding to the disease-gene association matrix, and β is the inter-layer jump probability.

[0087] Then the matrix M HNA and M HNB Further column normalization is performed to obtain the two final heterogeneous network probability transfer matrices T HNA and T HNB , and then generate two prediction score lists for disease candidate genes through the heterogeneous network random walk dynamics process. The calculation formula is: Where T H represents the heterogeneous network probability transfer matrix, and represents the initial probability vector of genes and diseases. It should be noted that The elements corresponding to known disease genes in are greater than zero, The element value corresponding to the disease being studied is equal to 1, and the other elements are 0. G and p D Represent the steady-state probability vector of the gene and the steady-state probability vector of the disease respectively. Using the above formula, in T H Equal to T HNA and T HNB From the two heterogeneous networks, two prediction lists of candidate genes are obtained.

[0088] Furthermore, as described in step S104, the two prediction lists are fused through a probability model based on node ranking aggregation to obtain a gene probability list, and the pathogenic gene of the disease is predicted according to the gene probability list, including:

[0089] Arrange the two prediction lists in descending order to obtain two ranking lists;

[0090] Merging the two ranking lists through the probability model to obtain the gene probability list;

[0091] The gene probability list is sorted in descending order and the causative gene of the disease is predicted based on the sorting result.

[0092] In specific implementation, after obtaining the prediction list, the two prediction lists can be arranged in descending order to obtain two ranking lists of candidate genes. A and f B , then through the probability model P(x i |f A ,f B )=∏ h∈{A,B} P(x i |f h) The above two ranking lists are combined to obtain the final gene probability list, where the probability function is defined as: P(x i |f A )=1 / f Ai , and f Ai Indicates that gene i is in f A Then the gene probability list is sorted in descending order and the disease-causing gene is predicted based on the sorting result.

[0093] The method for identifying pathogenic genes based on a multi-scale modular kernel disclosed in the present invention will be described below in conjunction with a specific embodiment. In order to verify the effectiveness of the method, the known pathogenic gene association data is used as a test platform to comprehensively evaluate the performance of the method for identifying pathogenic genes based on a multi-scale modular kernel, HyMSMK. The process is as follows: Figure 2 As shown:

[0094] (1) Test data: Disease-gene association data comes from databases such as OMIM and GWAS; gene-gene association data comes from comprehensive protein-protein interaction studies; disease-disease association data comes from disease-symptom association studies using text mining;

[0095] (2) Evaluation method: The performance of this method was tested and evaluated by five-fold cross validation. The five-fold cross validation randomly split the known pathogenic gene set of each disease into five parts, each of which was used as a positive test set, and the rest were used as training sets. The prediction accuracy (Precision or Prec), recall (Recall), and area under the Precision-Recall curve (AUPRC) of the top-k prediction list were used to quantitatively evaluate the performance of this method.

[0096] (3) Evaluation results

[0097] Figure 3 In the figure, (a) to (f) show that the prediction performance of the HyMSMK method, a pathogenic gene identification method based on a multi-scale modular kernel, gradually improves as the sparsity of the multi-scale kernel decreases. When the number of neighbors in the sparsification operation exceeds 1000, it has a relatively stable prediction performance. Among them, (a)-(b) is the recall rate Recall, (c)-(d) is the precision Prec, and (e)-(f) is the AUPRC performance; the first row corresponds to the linkage interval control set condition, and the second row corresponds to the whole genome control set condition. Figure 4As shown in the figure, (a) to (f) show that as the sparsity of the multi-scale kernel decreases, the storage space and time requirements of the HyMSMK method increase approximately linearly, and even exceed the space and time requirements of the full matrix mode. sMRM represents the sparse matrix mode, fMRM represents the full matrix mode, and pNNZE represents the proportion of non-zero elements. (a)-(c) are tests on multi-scale kernels; (d)-(f) are tests on heterogeneous networks combined with multi-scale kernels. However, when k=1000, the time and space requirements of the HyMSMK method are significantly less than those of the full matrix mode. Therefore, for HyMSMK, k=1000 is a good sparsity parameter.

[0098] like Figure 5 As shown, (a)-(b) are AUPRC performance, (c)-(d) are recall rates, and (e)-(f) are precision rates. The first row corresponds to the linkage interval control set condition, and the second row corresponds to the whole genome control set condition, showing the performance indicators of the HyMSMK method and other advanced network-based prediction methods. Figure 6 In the figure, (a)-(b) represent the top-k recall and precision under the linkage interval control set; (c)-(d) represent the top-k recall and precision under the whole genome control set. (a) to (d) further demonstrate the local prediction performance (top-1 to top-10) of HyMSMK and other methods. The results show that the prediction performance of the HyMSMK method is better than these advanced comparison methods, showing the advantage of the HyMSMK method in prediction performance.

[0099] It can be seen that this method HyMSMK effectively describes the multi-scale module structure information of the network by constructing a multi-scale module spectrum, and uses this multi-scale structure information to infer potential new gene-gene associations by constructing a multi-scale module kernel, providing an effective solution for predicting disease-causing genes using the multi-scale module structure of the network, thereby improving the prediction ability of disease-causing genes.

[0100] Corresponding to the above method embodiment, see Figure 7 The embodiment of the present disclosure also provides a pathogenic gene identification system 70 based on a multi-scale module core, comprising:

[0101] Extraction module 701, used to extract network multi-scale modules from the comprehensive protein interaction group through the multi-scale modularity optimization algorithm of exponential sampling, construct an initial multi-scale module spectrum containing local information and global information of the network, and perform weighting processing on it using the characteristic information method to obtain a standardized multi-scale module spectrum;

[0102] A construction module 702 is used to construct a multi-scale module kernel matrix according to the standardized multi-scale module spectrum and perform kernel sparseness to obtain a sparse matrix;

[0103] an association module 703, for constructing two probability matrices according to the protein interactome, the disease-disease association and the disease-gene association, and according to the sparse matrix, the disease-disease association and the disease-gene association, respectively, and extracting two prediction lists from the two probability matrices through a walk dynamics process;

[0104] The prediction module 704 is used to fuse the two prediction lists through a probability model based on node ranking aggregation to obtain a gene probability list, and predict the disease-causing gene of the disease based on the gene probability list.

[0105] Figure 7 The system shown can correspondingly execute the contents of the above method embodiment. For the parts not described in detail in this embodiment, refer to the contents recorded in the above method embodiment and will not be repeated here.

[0106] The above is only a specific implementation of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present disclosure should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims

1. A method for identifying pathogenic genes based on a multi-scale modular kernel, characterized in that: include: The network multiscale modules were extracted from the comprehensive protein interaction group by the multiscale modularity optimization algorithm with exponential sampling, and the initial multiscale module spectrum containing local and global information of the network was constructed. The characteristic information method was used to weight the initial multiscale module spectrum to obtain the standardized multiscale module spectrum. The formula of the network multiscale module is: Among them, e ss represents the edge ratio within module s, a s It represents the edge ratio in module s in the random model, γ is the resolution parameter, and the γ resolution parameter values ​​corresponding to different scales are obtained by exponential sampling; The standardized multi-scale module spectrum: B = (B (1) ,B (2) ,…,B (h) ,…), where B (h) is the module partition matrix for each scale, the elements of which are Indicates whether gene i belongs to module s; The formula for the weighting process is: Among them, the diagonal matrix D IC The diagonal elements (D IC ) ss =-log(∑ i I(B is ) / N), I(x) represents the counting function, N represents the number of nodes; Constructing a multi-scale module kernel matrix according to the standardized multi-scale module spectrum and performing kernel sparseness to obtain a sparse matrix; constructing two probability matrices according to the protein interactome, the disease-disease association and the disease-gene association, and according to the sparse matrix, the disease-disease association and the disease-gene association, respectively, and extracting two prediction lists from the two probability matrices through a walk dynamics process; The two prediction lists are fused through a probability model based on node ranking aggregation to obtain a gene probability list, and the pathogenic gene of the disease is predicted based on the gene probability list.

2. The method according to claim 1, characterized in that The step of constructing a multi-scale module kernel matrix according to the standardized multi-scale module spectrum and performing kernel sparseness to obtain a sparse matrix includes: The standardized multi-scale module spectrum is used as a node feature spectrum and the multi-scale module kernel matrix is ​​obtained by calculating the inner product of the node feature spectrum vector; The elements in the multi-scale module core matrix are standardized by a node degree-related standardization method so that the element values ​​are constrained to be between 0 and 1, thereby obtaining a standardized module core matrix; A preset number of similar neighbor nodes corresponding to each node in the standardized module core matrix is ​​retained, and the standardized module core matrix is ​​subjected to core sparseness to obtain the sparse matrix.

3. The method according to claim 2, characterized in that The step of retaining a preset number of similar neighbor nodes corresponding to each node in the standardized module core matrix, performing core sparseness on the standardized module core matrix, and obtaining the sparse matrix includes: Calculating the descending ranking values ​​of the similarity scores between each of the nodes and its neighboring nodes in the standardized module core matrix to obtain a ranking vector for each of the nodes; Combining the ranking vectors of all the nodes to form a ranking matrix; Calculate a new gene network according to the ranking matrix; The sparse matrix is ​​obtained according to the gene network.

4. The method according to claim 1, characterized in that The steps of constructing two probability matrices according to the protein interactome, the disease-disease association and the disease-gene association, and according to the sparse matrix, the disease-disease association and the disease-gene association, respectively, and extracting two prediction lists from the two probability matrices through a walk dynamics process include: establishing a first probability matrix based on the protein interactome, the disease-disease associations, and the disease-gene associations, and establishing a second probability matrix based on the sparse matrix, the disease-disease associations, and the disease-gene associations; Normalizing the first probability matrix to obtain a first probability transfer matrix, and normalizing the second probability matrix to obtain a second probability transfer matrix; The prediction scores corresponding to the first probability transfer matrix and the second probability transfer matrix are respectively generated according to the walk dynamics process to form two prediction lists.

5. The method according to claim 1, characterized in that The step of fusing the two prediction lists through a probability model based on node ranking aggregation to obtain a gene probability list, and predicting the pathogenic gene of the disease according to the gene probability list includes: Arrange the two prediction lists in descending order to obtain two ranking lists; Merging the two ranking lists through the probability model to obtain the gene probability list; The gene probability list is sorted in descending order and the causative gene of the disease is predicted based on the sorting result.

6. A pathogenic gene identification system based on a multi-scale modular core, characterized in that: include: The extraction module is used to extract network multiscale modules from the comprehensive protein interaction group through the multiscale modularity optimization algorithm of exponential sampling, construct an initial multiscale module spectrum containing local and global information of the network, and perform weighted processing on it using the characteristic information method to obtain a standardized multiscale module spectrum, wherein the formula of the network multiscale module is Among them, e ss represents the edge ratio within module s, a s It represents the edge ratio in module s in the random model, γ is the resolution parameter, and the γ resolution parameter values ​​corresponding to different scales are obtained by exponential sampling; The standardized multi-scale module spectrum: B = (B (1) ,B (2) ,…,B (h) ,…), where B (h) is the module partition matrix for each scale, the elements of which are Indicates whether gene i belongs to module s; The formula for the weighting process is: Among them, the diagonal matrix D IC The diagonal elements (D IC ) ss =-log(∑ i I(B is ) / N), I(x) represents the counting function, N represents the number of nodes; A construction module is used to construct a multi-scale module kernel matrix according to the standardized multi-scale module spectrum and perform kernel sparseness to obtain a sparse matrix; an association module, for constructing two probability matrices according to the protein interactome, the disease-disease association and the disease-gene association, and according to the sparse matrix, the disease-disease association and the disease-gene association, respectively, and extracting two prediction lists from the two probability matrices through a walk dynamics process; The prediction module is used to fuse the two prediction lists through a probability model based on node ranking aggregation to obtain a gene probability list, and predict the pathogenic gene of the disease based on the gene probability list.

Images