Disease gene identification method and system based on heterogeneous network pulse dynamics

Abstract

The invention discloses a disease gene identification method and system based on heterogeneous network pulse dynamics, and the method comprises the following steps: obtaining a GGA (Gene Associated Network), a DDA (Disease Associated Network) and a DGA (Disease Gene Associated Network), function similarity scores between gene nodes and function similarity scores between disease nodes are respectively calculated through a GGA network, a DDA network and a DGA network to enhance the GGA network and the DDA network, the enhanced GGA network and the enhanced DDA network are both combined with the DGA network, and an enhanced double-layer heterogeneous network comprising a disease layer and a gene layer is constructed; applying a pulse signal to an excitation node of the enhanced double-layer heterogeneous network, and tracking by taking a dynamic response of a node to be predicted to pulse excitation as a function of time to obtain a heterogeneous network dynamic pulse model; and according to the enhanced heterogeneous network dynamic pulse model, acquiring the maximum pulse response of the node as an effective pulse dynamic feature to identify a disease gene.

Description

Technical Field

[0001] The present invention relates to the technical field of bioinformatics analysis, and in particular to a disease gene identification method based on heterogeneous network impulse dynamics and a system thereof. Background Art

[0002] Identifying disease-associated genes (DAGs) is of great significance for the study of complex diseases such as cancer, because the occurrence and development of diseases are closely related to mutations or dysfunctions of related genes. The knowledge of DAGs is constantly developing, but its practical application is still very limited due to the time-consuming and high cost of biological experiments. Therefore, many computational methods combining various biomedical data have been used to promote the study of DAGs.

[0003] Genes and their products usually play biological functions through synergistic effects in organisms, forming complex biomolecular networks, such as protein-protein interaction (PPI) networks. The onset of disease is usually considered to be the result of local abnormalities in biomolecular network systems, and biomolecular networks can serve as a basic organizational framework for studying diseases. In addition, many related diseases with similar phenotypes or symptoms are often related at the genetic level. Therefore, the biomolecular network-based approach is a powerful and concise strategy to infer DAGs and related problems.

[0004] Network propagation can successfully capture some intrinsic biological characteristics and mechanisms of disease onset from a network perspective, which helps explore unknown associations between biomolecules and diseases. Therefore, network propagation is one of the most widely studied network-based methods and the most advanced representative of genetic association studies. However, traditional network propagation-based methods are usually based only on steady-state solutions of propagation dynamics, which will lead to the neglect of some potentially valuable information in the dynamic process of network propagation. Summary of the invention

[0005] In view of this, the present invention provides a disease gene identification method and system based on heterogeneous network pulse dynamics, which are used to at least solve the problems in the prior art such as how to effectively integrate and mine multi-source heterogeneous networks such as gene-gene associations, disease-gene associations, and disease-disease associations, and how to more effectively extract disease gene-related feature information from heterogeneous networks, so as to more effectively calculate and identify disease-related genes.

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

[0007] A disease gene identification method based on heterogeneous network impulse dynamics comprises the following steps:

[0008] S1. Obtain data; obtain gene-gene association matrix A respectively GG , disease-disease association matrix ADD and the disease-gene association matrix A DG , to characterize the gene association network GGA, disease association network DDA and disease gene association network DGA;

[0009] S2. Construct an enhanced double-layer heterogeneous network; calculate the functional similarity scores between gene nodes through the DDA network and the DGA network, calculate the functional similarity scores between disease nodes through the GGA network and the DGA network, and then enhance the GGA network and the DDA network through the functional similarity scores between gene nodes and disease nodes, combine the enhanced GGA network and the enhanced DDA network with the DGA network, and construct an enhanced double-layer heterogeneous network including the disease layer and the gene layer, and connect the enhanced double-layer heterogeneous network through its adjacent matrix A. H express;

[0010] S3. Construct a dynamic pulse model of heterogeneous networks; in an enhanced two-layer heterogeneous network, the disease node to be identified and the corresponding known gene node are used as excitation nodes, a pulse signal is applied to the excitation node, and the dynamic response of the node to be predicted to the pulse signal as the pulse signal propagates is recorded, and it is tracked as a function of time to obtain a dynamic pulse model of heterogeneous networks;

[0011] S4. Based on the heterogeneous network dynamic impulse model, the maximum impulse response is obtained as an effective impulse dynamic feature to identify disease genes.

[0012] Preferably, in S1, The element value indicates the strength of association between genes; The element value represents the strength of association between diseases. The element value indicates whether there is an association between the disease and the gene, R is a real number matrix, N G is the total number of genes, N D The total number of diseases.

[0013] Preferably, in S2, the functional similarity scores between gene nodes are calculated by the DDA network and the DGA network, the functional similarity scores between disease nodes are calculated by the GGA network and the DGA network, and the specific contents of enhancing the GGA network and the DDA network by the functional similarity scores between gene nodes and disease nodes include:

[0014] The functional similarity scores between gene nodes are calculated from the perspectives of DDA network and DGA network, and the functional similarity matrix S is constructed. GG , according to S GG The GGA network enhanced with the GGA network is expressed as

[0015] The functional similarity scores between disease nodes are calculated from the perspectives of GGA network and DGA network, and the functional similarity matrix S is constructed. DD , according to S DD The DDA network enhanced with the GGA network is expressed as

[0016] Preferably, the specific contents of constructing the enhanced double-layer heterogeneous network include:

[0017] Functional similarity matrix S GG The calculation method of the element value in is:

[0018]

[0019] Among them, the subscripts i and j represent the numbers of the two gene nodes in their respective networks, i∈{1,2,3,…,N G}, j∈{1,2,3,…,N G},N G represents the number of gene nodes, (S GG ) ij represents the functional similarity score between gene i and gene j; subscripts h and k represent the numbers of the two disease nodes in their respective networks, h∈{1,2,3,…,N D}, k∈{1,2,3,…,N D},N D represents the number of disease nodes, (A DD ) hk represents the original similarity score between disease h and disease k; (A GD ) jk Represents the association between gene i and disease j. If there is a known association, its value is equal to 1, otherwise it is equal to 0;

[0020] The enhanced GGA network The element values ​​in are:

[0021]

[0022] Functional similarity matrix S DD The element values ​​in are:

[0023]

[0024] The enhanced DDA network The element values ​​in are:

[0025]

[0026] Based on the enhanced GGA and enhanced DDA networks, by integrating and A DG , construct an enhanced two-layer heterogeneous network, the adjacent matrix A H for:

[0027]

[0028] In the formula, R is a real matrix, N G is the total number of genes, N D The total number of diseases.

[0029] Preferably, the dynamic pulse model of heterogeneous network is:

[0030] According to the classical dynamics model, the dynamics equation of the enhanced double-layer heterogeneous network is written as:

[0031]

[0032] In the formula, x it is the state of the i-th gene node in the gene layer at time t; y ht represents the state of the h-th disease node in the disease layer at time t; k m represents the average of the inter- and intra-layer diffusion contributions, and Among them, mean() is the mean function, 1 is the vector of all 1s, is the matrix considering the contribution of inter-layer and intra-layer diffusion; ω is the relative weight of intra-layer and inter-layer diffusion; b Gi Indicates whether the i-th gene node is a control node stimulated by the pulse signal; b Dh Indicates whether the hth disease node is a control node stimulated by the pulse signal; u t =δ(tt σ ) is an indicator of a periodic pulse signal, where δ is a Dirac pulse function and the pulse time constant t σ =5σ,σ=0,1,2,...;

[0033] Define the matrix that takes into account the inter-layer and intra-layer diffusion contributions

[0034]

[0035] Then the dynamic equation of the enhanced double-layer heterogeneous network is rewritten as:

[0036]

[0037]

[0038]

[0039]

[0040]

[0041]

[0042] Where b G Indicates b Gi The vector composed of D Indicates b Dh is a vector composed of , and η is the weight of the exciting node.

[0043] Preferably, in S4, the maximum response of the node to be predicted to the impulse signal on the dynamic impulse model of the heterogeneous network is calculated, that is, as effective pulse dynamic signatures to identify genes associated with specific diseases.

[0044] A disease gene identification system based on heterogeneous network impulse dynamics, at least one processor; and at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute a disease gene identification method based on heterogeneous network impulse dynamics.

[0045] A non-transitory computer-readable storage medium stores computer instructions, which enable the computer to execute a disease gene identification method based on heterogeneous network impulse dynamics.

[0046] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a disease gene identification method and system based on heterogeneous network impulse dynamics, which has the following beneficial effects:

[0047] The present invention is to effectively integrate multivariate heterogeneous network data, jointly construct the original disease-gene heterogeneous network, and then construct an enhanced gene-gene network and an enhanced disease-disease network by preliminarily mining multiple networks of gene-gene association, disease-gene association and disease-disease association, and then construct an enhanced disease-gene double-layer heterogeneous network, thereby alleviating the data noise problem existing in the existing network and promoting the improvement of the prediction model performance. Unlike the traditional method, the present invention designs an impulse dynamics process for the above-mentioned jointly constructed enhanced disease-gene double-layer heterogeneous network to realize a heterogeneous network impulse dynamics model for disease gene prediction, because the model can utilize disease-disease association information and utilize excitation nodes (including disease nodes and their known disease genes) to stimulate heterogeneous network impulse dynamics processes to obtain effective impulse dynamic characteristics of gene nodes in the process of responding to stimulus responses, thereby improving the disease gene prediction performance. Through a series of experimental evaluations such as cross-validation experiments and new disease experiments, the method of the present invention has been proven to be superior to the classic network-based baseline algorithm in performance indicators such as AUPRC, AURecall, AUPrec and AUROC. The case study of cancer further proves the effectiveness of the method. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 A flowchart of a disease gene identification method based on heterogeneous network impulse dynamics provided by the present invention, wherein Figure 1 (a) is a schematic diagram of the GGA network, DDA network and DGA network. Figure 1 (b) is the enhanced DDA network. Figure 1 (c) is the enhanced GGA network. Figure 1 (d) is an enhanced two-layer heterogeneous network. Figure 1 (e) is the dynamic response function after applying a pulse signal to the excitation node;

[0050] Figure 2 A line graph showing the effect of the model parameters provided in the embodiment of the present invention on the prediction performance under different control set conditions, wherein Figure 2 (a)-(d) are the experimental results of the HNID method of the present invention in terms of AUPRC, AURecall, AUPrec and AUROC under the ALI control set. Figure 2(e)-(h) are the experimental results of the HNID method of the present invention in terms of AUPRC, AURecall, AURrec and AUROC in the RC control set;

[0051] Figure 3 The statistical diagram of the impact of prediction performance under different control set conditions in the cross-validation experiment provided by the embodiment of the present invention, wherein Figure 3 (a)-(d) are the experimental results of the HNID method of the present invention and various classic network algorithms in terms of AUPRC, AURecall, AUPrec and AUROC under the ALI control set. Figure 3 (e)-(h) are the experimental results of the HNID method of the present invention and various classic network algorithms in terms of AUPRC, AURecall, AURrec and AUROC in the RC control set;

[0052] Figure 4 The statistical diagram of the impact of different control set conditions on the prediction performance in the experiment of a new disease provided by the embodiment of the present invention, wherein Figure 4 (a)-(d) are the experimental results of the HNID method of the present invention and various classic network algorithms in terms of AUPRC, AURecall, AUPrec and AUROC under the ALI control set. Figure 4 (e)-(h) are the experimental results of the HNID method of the present invention and various classic network algorithms in terms of AUPRC, AURecall, AURrec and AUROC in the RC control set. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are 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 creative work are within the scope of protection of the present invention.

[0054] The present invention provides a disease gene identification method based on heterogeneous network impulse dynamics, comprising the following steps:

[0055] S1. Obtain data; obtain gene-gene association matrix A respectively GG , disease disease association matrix A DD and the disease-gene association matrix A DG , to characterize the gene association network GGA, disease association network DDA and disease gene association network DGA;

[0056] S2. Construct an enhanced double-layer heterogeneous network; calculate the functional similarity scores between gene nodes through the DDA network and the DGA network, calculate the functional similarity scores between disease nodes through the GGA network and the DGA network, and then enhance the GGA network and the DDA network through the functional similarity scores between gene nodes and disease nodes, combine the enhanced GGA network and the enhanced DDA network with the DGA network, and construct an enhanced double-layer heterogeneous network including the disease layer and the gene layer, and connect the enhanced double-layer heterogeneous network through its adjacent matrix A. H express;

[0057] S3. Construct a dynamic pulse model of heterogeneous networks; in an enhanced two-layer heterogeneous network, the disease node to be identified and the corresponding known gene node are used as excitation nodes, a pulse signal is applied to the excitation node, and the dynamic response of the node to be predicted to the pulse signal as the pulse signal propagates is recorded, and it is tracked as a function of time to obtain a dynamic pulse model of heterogeneous networks;

[0058] S4. Based on the heterogeneous network dynamic impulse model, the maximum impulse response is obtained as an effective impulse dynamic feature to identify disease genes.

[0059] It should be noted that:

[0060] DGA network: In this embodiment, a large number of known related genes of diseases are obtained from the authoritative human gene and disease database OMIM. The gene-related annotation files are directly downloaded, and the gene-disease relationship pairs are extracted and saved as a list file. Then, the DGA network is extracted. The list file can be represented by a matrix in the computer, that is, the disease-gene association matrix A DG , to facilitate indexing and matrix operations;

[0061] GGA network: The protein interaction network PPI can reflect the intrinsic relationship between genes. Therefore, in this example, comprehensive protein interaction network data was obtained from existing literature. Its form is also a list file of protein-protein interaction pairs. The gene-gene association matrix A GG express;

[0062] DDA network: Obtain the phenotypic annotation files of human diseases and the ontology annotation files associated with phenotypes from the authoritative Human Phenotype Ontology database, use the relationship between the phenotypic annotations of diseases and phenotypes organized as directed acyclic phenotype graphs to calculate the phenotype-based disease similarity score, and then express it as a disease-disease association matrix A DD , to characterize the disease association network DDA.

[0063] There are known related genes for the disease to be identified. Therefore, in S3, the disease node to be identified and its known related genes are used as excitation nodes, and a pulse signal is applied to the excitation node. As the pulse signal propagates, the node to be predicted will be affected by the pulse excitation to varying degrees. According to the dynamic response, a dynamic pulse model of a heterogeneous network is obtained. According to the dynamic response model, an effective pulse dynamic feature is obtained, and the association between the pulse and the above-mentioned disease to be identified is judged by the pulse dynamic feature.

[0064] In order to further implement the above technical solution, in S1, The element value indicates the strength of association between genes; The element value represents the strength of association between diseases. The element value indicates whether there is an association between the disease and the gene, R is a real number matrix, N G is the total number of genes, N D The total number of diseases.

[0065] In order to further implement the above technical solution, S2 calculates the functional similarity scores between gene nodes through the DDA network and the DGA network, calculates the functional similarity scores between disease nodes through the GGA network and the DGA network, and then enhances the GGA network and the DDA network through the functional similarity scores between gene nodes and disease nodes. The specific contents include:

[0066] The functional similarity scores between gene nodes are calculated from the perspectives of DDA network and DGA network, and the functional similarity matrix S is constructed. GG , according to S GG The GGA network enhanced with the GGA network is expressed as

[0067] The functional similarity scores between disease nodes are calculated from the perspectives of GGA network and DGA network, and the functional similarity matrix S is constructed. DD , according to S DD The DDA network enhanced with the GGA network is expressed as

[0068] In order to further implement the above technical solution, the specific contents of constructing an enhanced two-layer heterogeneous network include:

[0069] Functional similarity matrix S GG The calculation method of the element value in is:

[0070]

[0071] Among them, the subscripts i and j represent the numbers of the two gene nodes in their respective networks, i∈{1,2,3,…,NG}, j∈{1,2,3,…,N G},N G represents the number of gene nodes, (S GG ) ij represents the functional similarity score between gene i and gene j; subscripts h and k represent the numbers of the two disease nodes in their respective networks, h∈{1,2,3,…,N D}, k∈{1,2,3,…,N D},N D represents the number of disease nodes, (A DD ) hk represents the original similarity score between disease h and disease k; (A GD ) jk Represents the association between gene i and disease j. If there is a known association, its value is equal to 1, otherwise it is equal to 0;

[0072] The enhanced GGA network The element values ​​in are:

[0073]

[0074] Functional similarity matrix S DD The element values ​​in are:

[0075]

[0076] The enhanced DDA network The element values ​​in are:

[0077]

[0078] Based on the enhanced GGA and DDA networks, by integrating and A DG , construct an enhanced two-layer heterogeneous network, the adjacent matrix A H for:

[0079]

[0080] In the formula, R is a real matrix, N G is the total number of genes, N D The total number of diseases.

[0081] It should be noted that:

[0082] In order to jointly utilize DGA, DDA and GGA networks for disease gene inference, the present invention naturally constructs a disease gene heterogeneous network by integrating these networks. However, data noise is prevalent in the network, and the GGA network is incomplete, especially there are a large number of false negatives. Therefore, the present invention strengthens the GGA and DDA networks by using the functional similarity scores between nodes obtained by the DGA, GGA and DDA networks.

[0083] In order to further implement the above technical solution, the dynamic pulse model of heterogeneous network is:

[0084] According to the classical dynamics model, the dynamics equation of the enhanced double-layer heterogeneous network is written as:

[0085]

[0086] In the formula, x it is the state of the i-th gene node in the gene layer at time t; y ht represents the state of the h-th disease node in the disease layer at time t; k m represents the average of the inter- and intra-layer diffusion contributions, and Among them, mean() is the mean function, 1 is the vector of all 1s, is the matrix considering the contribution of inter-layer and intra-layer diffusion; ω is the relative weight of intra-layer and inter-layer diffusion; b Gi Indicates whether the i-th gene node is a control node stimulated by the pulse signal; b Dh Indicates whether the hth disease node is a control node stimulated by the pulse signal; u t =δ(tt σ ) is an indicator of a periodic pulse signal, where δ is a Dirac pulse function and the pulse time constant t σ =5σ,σ=0,1,2,...;

[0087] Define the matrix that takes into account the inter-layer and intra-layer diffusion contributions

[0088]

[0089] Then the dynamic equation of the enhanced double-layer heterogeneous network is rewritten as:

[0090]

[0091]

[0092]

[0093]

[0094]

[0095]

[0096] Where b G Indicates b Gi The vector composed of D Indicates b Dh is a vector composed of , and η is the weight of the exciting node.

[0097] It should be noted that:

[0098] The general kinetic equation of the classical kinetic model is:

[0099]

[0100] where x it represents the state of node i (i = 1, 2, 3, ...) at time t; f(x it )=-x it is a continuous differentiable function that describes the self-attenuation of the signal on a node without being affected by other nodes; A ij It is the element of the adjacent matrix A in the network, which determines the diffusion coefficient between nodes; k m =mean(A1), where 1 is the full vector; b i Indicates whether node i is a control node excited by a pulse signal; u t =δ(tt σ ) is an indicator of a periodic pulse signal, where δ is a Dirac (pulse) function and the pulse time constant t σ =5σ,σ=0,1,2,…。

[0101] According to the above dynamic equation, the pulse signal applied to the excitation node will propagate along the network. The dynamic response of a node to the pulse excitation implies the association information between the node and the excited node. The above dynamic model is extended to a two-layer heterogeneous network including the gene network layer and the disease network layer, such as the adjacent matrix A of the above heterogeneous network H For the nodes of the gene network layer and the disease network layer, the dynamic equations can be written as the dynamic equations of the enhanced two-layer heterogeneous network: and

[0102] and The two sets of dynamic equations can describe the dynamic pulse process of genes and diseases in heterogeneous networks, which are jointly affected by the gene and disease network layers and the interactions between layers. By recording the dynamic response of the predicted node to the pulse signal applied to the excitation node, the hidden information in the heterogeneous network is mined to reveal the intrinsic relationship between the research node and the excitation node. For simplicity, a matrix considering the inter-layer and intra-layer diffusion contribution is defined in this invention:

[0103] The final vector equation shows that the dynamic pulse model of heterogeneous networks includes three parts: signal self-attenuation, signal diffusion in non-uniform networks, and pulse excitation. In order to mine useful information for disease gene inference, the present invention stimulates the excitation node to start the dynamic pulse process in the heterogeneous network according to the above model, and then tracks the response of the node to be predicted (such as gene) to the pulse excitation as a function of time after the pulse dynamic process is stimulated.

[0104] Figure 1 e shows an example of the impulse response of a node, indicating that the impulse response of a node will rise first, reach a peak, and then fall. The stronger the response of a node to the impulse excitation, the more likely it is to be associated with the exciting node(s). The present invention uses the maximum response of the node to the impulse excitation, i.e. As an effective pulse dynamic feature to infer genes that may be associated with specific diseases.

[0105] In order to further implement the above technical solution, in S4, the maximum response of the node to be predicted to the pulse signal on the dynamic pulse model of the heterogeneous network is calculated, that is, as effective pulse dynamic signatures to identify genes associated with specific diseases.

[0106] A disease gene identification system based on heterogeneous network impulse dynamics, at least one processor; and at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute a disease gene identification method based on heterogeneous network impulse dynamics.

[0107] A non-transitory computer-readable storage medium stores computer instructions, which enable the computer to execute a disease gene identification method based on heterogeneous network impulse dynamics.

[0108] The present invention will be verified by specific experiments below:

[0109] In order to evaluate the prediction performance of the present invention and infer potential disease-related genes, high-quality networks are required. In this embodiment, disease and gene networks are obtained / derived through several authoritative databases / datasets.

[0110] Disease-gene associations. Disease-gene association data were collected from the OMIM (Online Mendelian Inheritance in Man) database (https: / / omim.org / ), a comprehensive and authoritative online catalog of human genes and genetic disorders / diseases.

[0111] Gene-gene association: There are complex interactions or correlations between proteins, forming a complex protein-protein interaction (PPI) network, which reflects the intrinsic association between proteins / genes and is an important basis for studying the molecular mechanisms of complex diseases. Because of the widespread existence of incomplete data, a high-quality protein interaction dataset is integrated - a comprehensive protein interaction group constructed by integrating PPIs from multiple sources, such as yeast two-hybrid high-throughput experiments and low-throughput interaction experiments, protein complexes and signaling interactions. The interaction group only considers physical protein interactions with experimental support.

[0112] Disease correlation. Disease-disease association (DDA) can provide useful information for the inference of disease-related genes. Here, the phenotypic or symptom annotations of the disease are used to construct the DDA network, because the knowledge of DDA at the phenotype and symptom levels can usually provide useful insights for disease associations at the gene level. This embodiment derives a DDA network from the human phenotype ontology database of authoritative human phenotypes. The phenotypic similarities between diseases are first calculated by using the phenotypic annotations of the disease and the relationships between phenotypes constructed as directed acyclic graphs in the human phenotype ontology database, and then a DDA network is constructed based on the phenotypic similarities between diseases.

[0113] In this embodiment, the algorithm model disclosed in the present invention is called the heterogeneous network dynamic impulse model (HNID). In order to verify its performance, this embodiment uses known disease gene association data as a benchmark and applies two evaluation strategies:

[0114] Traditional 5-fold cross validation (5FCV) and novel disease detection (TND). 5FCV randomly divides known disease-associated genes into 5 subsets. Each of these subsets is used in turn as a (positive) test set (TND). d ), and the rest are used as a training set. TND will mask all known disease-associated genes, that is, it uses all known disease-associated genes as the (positive) test set (T d ), while there are no known disease-related genes in the training set.

[0115] In this embodiment, a test gene set is integrated with one of the two control gene sets to construct two types of candidate gene sets (C d): Artificial Linkage Interval (ALI) control set and random (RC) control set. The ALI strategy selects 99 control genes from the closest genes to the test gene on the same chromosome, and the RC strategy randomly selects 99 control genes for each test gene from the entire genome.

[0116] Evaluation indicators: Based on the scores of candidate genes, the prediction performance of the method is quantitatively evaluated. Given a set of top r candidate genes for a disease, these are predicted positive genes, TP and FP are used to represent the true positive genes and false positive genes in the set, and then several standard evaluation indicators are calculated: Precision Prec(r) = |TP| / r, Recall Recall(r) = |TP| / |T d |, false positive rate FPR(r) = 1-|FP| / |C d \T d |.

[0117] According to the above evaluation indicators, we can get the area under the operating characteristic curve AUROC and the area under the precision-recall curve AUPRC, which have been widely used in the comprehensive performance evaluation of algorithms. In order to facilitate a comprehensive and quantitative comparison of the local performance of each method, we further calculate the area under the precision curve AUPrec and the area under the recall curve AURecall:

[0118]

[0119]

[0120] In the formula, Precision is the accuracy and Recall is the recall rate.

[0121] Overall performance in cross-validation experiments:

[0122] 1) Influence of model parameters:

[0123] Firstly, the influence of model parameters on the prediction performance is studied through cross-validation experiments using two local performance curves and four comprehensive performance indicators. Figure 2 It shows that the weight of inter-layer diffusion (ω) has a relatively small impact on the performance (AUPRC, AURecall, AUPrec, and AUROC); for a given value of ω, the performance of the model under both control sets increases slowly with the increase of the weight of the disease network layer excitation node (η), but when the η value is close to 1, the performance of the model increases rapidly. This indicates that it is better to apply more pulse signals to the disease (node) under study.

[0124] 2) Performance comparison:

[0125] In the cross-validation experiment, HNID was compared with classic network algorithms under different control sets. The classic network algorithms include:

[0126] RWRH is a classic random walk algorithm for heterogeneous networks. It integrates protein / gene networks and disease networks by using disease-gene associations to construct a heterogeneous network, and then performs random walks with restarts in the heterogeneous network to predict disease-related candidate genes.

[0127] PRINCE is a network propagation method that takes network smoothness into account. It computes a priority function that is related to its smoothness and use of prior information on the protein / gene network. Based on the similarity between the query disease and other diseases, it obtains a set of known disease-related genes of similar diseases to construct a prior vector, and then propagates the scores of the prior gene set along the edges of the protein / gene network to obtain the final association scores for all genes, where the weight of each edge is normalized by the degree of its endpoint.

[0128] CIPHER assumes that two genes that are closer to each other in a molecular interaction network usually lead to more similar phenotypes, and formulates this assumption into a regression model, from which a score is derived to assess the likelihood that a gene is involved in a specific phenotype. This score, called the consistency score, is calculated across the entire phenotypic network and protein network to measure the overall consistency between phenotypic similarity and the functional genetic relevance of disease genes.

[0129] Figure 3 It shows that HNID has higher comprehensive performance index values ​​under different control sets. Specifically, under the ALI control set, HNID exceeds the (best) baseline by 7%, 3%, 7% and 1% in AUPRC, AURecall, AUPrec and AUROC, respectively; in the RC control set, AUPRC, AURecall, AURrec and AUROC exceed the (best) baseline by 15%, 3%, 9% and 2%, respectively. These results show that HNID has good performance in cross-validation experiments.

[0130] Comprehensive performance in new disease experiments:

[0131] In this example, the performance of HNID is further evaluated in a new disease experiment, in which all known DGAs related to the disease under study are masked. The experimental results show that under different control sets, the comprehensive performance indicators of HNID are also higher than those of the classical algorithms, such as Figure 4 This confirms that HNID also has good performance in experiments with novel diseases.

[0132] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A disease gene identification method based on heterogeneous network impulse dynamics, characterized in that: The following steps are involved: S1. Obtain data; obtain gene-gene association matrix A respectively GG , disease-disease association matrix A DD and the disease-gene association matrix A DG , to characterize the gene association network GGA, disease association network DDA and disease gene association network DGA; S2. Construct an enhanced double-layer heterogeneous network; calculate the functional similarity scores between gene nodes through the DDA network and the DGA network, calculate the functional similarity scores between disease nodes through the GGA network and the DGA network, and then enhance the GGA network and the DDA network through the functional similarity scores between gene nodes and disease nodes, combine the enhanced GGA network and the DDA network with the DGA network, construct an enhanced double-layer heterogeneous network including the disease layer and the gene layer, and connect the enhanced double-layer heterogeneous network through its adjacent matrix A. H express; S3. Construct a dynamic pulse model of heterogeneous networks; in an enhanced two-layer heterogeneous network, the disease node to be identified and the corresponding known gene node are used as excitation nodes, a pulse signal is applied to the excitation node, and the dynamic response of the node to be predicted to the pulse signal as the pulse signal propagates is recorded, and it is tracked as a function of time to obtain a dynamic pulse model of heterogeneous networks; S4. Based on the heterogeneous network dynamic impulse model, the maximum impulse response is obtained as an effective impulse dynamic feature to identify disease genes.

2. A disease gene identification method based on heterogeneous network impulse dynamics according to claim 1, characterized in that: In S1, The element value indicates the strength of association between genes; The element value represents the strength of association between diseases. The element value indicates whether there is an association between the disease and the gene, R is a real number matrix, N G is the total number of genes, N D The total number of diseases.

3. The disease gene identification method based on heterogeneous network impulse dynamics according to claim 1 is characterized in that: In S2, the functional similarity scores between gene nodes are calculated through the DDA network and the DGA network, and the functional similarity scores between disease nodes are calculated through the GGA network and the DGA network. The specific contents of enhancing the GGA network and the DDA network through the functional similarity scores between gene nodes and disease nodes include: The functional similarity scores between gene nodes are calculated from the perspectives of GGA network and DGA network, and the functional similarity matrix S is constructed. GG , according to S GG The GGA network enhanced with the GGA network is expressed as The functional similarity scores between disease nodes are calculated from the perspectives of DDA network and DGA network, and the functional similarity matrix S is constructed. DD , according to S DD The DDA network enhanced with the DDA network is expressed as 4. A disease gene identification method based on heterogeneous network impulse dynamics according to claim 3, characterized in that: The specific contents of building an enhanced two-layer heterogeneous network include: Functional similarity matrix S GG The calculation method of the element value in is: Among them, the subscripts i and j represent the numbers of the two gene nodes in their respective networks, i∈{1,2,3,…,N G }, j∈{1,2,3,…,N G },N G represents the number of gene nodes, (S GG ) ij represents the functional similarity score between gene i and gene j; subscripts h and k represent the numbers of the two disease nodes in their respective networks, h∈{1,2,3,…,N D }, k∈{1,2,3,…,N D },N D represents the number of disease nodes, (A DD ) hk represents the original similarity score between disease h and disease k; (A GD ) jk Represents the association between gene i and disease j. If there is a known association, its value is equal to 1, otherwise it is equal to 0; The enhanced GGA network The element values ​​in are: Functional similarity matrix S DD The element values ​​in are: The enhanced DDA network The element values ​​in are: Based on the enhanced GGA and enhanced DDA networks, by integrating and A DG , construct an enhanced two-layer heterogeneous network, the adjacent matrix A H for: In the formula, R is a real matrix, N G is the total number of genes, N D The total number of diseases.

5. The disease gene identification method based on heterogeneous network impulse dynamics according to claim 3 is characterized in that: The dynamic pulse model of heterogeneous networks is: According to the classical dynamics model, the dynamics equation of the enhanced double-layer heterogeneous network is written as: In the formula, x it is the state of the i-th gene node in the gene layer at time t; y ht represents the state of the h-th disease node in the disease layer at time t; Represents x it The derivative of Represents y ht The derivative of k m represents the average degree of inter-layer and intra-layer diffusion contribution, and Among them, mean() is the function for finding the average value, 1 is the vector of all 1s, is the matrix considering the contribution of inter-layer and intra-layer diffusion; ω is the relative weight of intra-layer and inter-layer diffusion; b Gi Indicates whether the i-th gene node is a control node stimulated by the pulse signal; b Dh Indicates whether the hth disease node is a control node stimulated by the pulse signal; u t =δ(tt σ ) is an indicator of a periodic pulse signal, where δ is a Dirac pulse function and the pulse time constant t σ =5σ,σ=0,1,2,...; Define the matrix that takes into account the inter-layer and intra-layer diffusion contributions Then the dynamic equation of the enhanced double-layer heterogeneous network is rewritten as: Where b G Indicates b Gi The vector composed of D Indicates b Dh is a vector composed of , and η is the weight of the excited node at the initial moment.

6. The disease gene identification method based on heterogeneous network impulse dynamics according to claim 1 is characterized in that: In S4, the maximum response of the node to be predicted to the impulse signal on the dynamic impulse model of the heterogeneous network is as effective pulse dynamic signatures to identify genes associated with specific diseases.

7. A disease gene identification system based on heterogeneous network impulse dynamics, characterized in that: at least one processor; And at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute a disease gene identification method based on heterogeneous network impulse dynamics as described in any one of claims 1 to 6.

8. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions enable the computer to execute a disease gene identification method based on heterogeneous network impulse dynamics as described in any one of claims 1 to 6.

Images