Metapath embedding-based miRNA-disease association prediction model construction method

Abstract

The invention belongs to the technical field related to biomedicine, and particularly relates to a miRNA-disease association prediction method based on meta-path embedding, which comprises the following steps: based on a miRNA and disease association database, taking miRNA and diseases in the miRNA and disease association database as nodes in a network, taking a connecting line of known associated miRNA and disease nodes as edges in the network, and distinguishing the types of the miRNA nodes and the disease nodes, so that the miRNA-disease association prediction method based on meta-path embedding is realized. A miRNA-disease heterogeneous network is constructed; the method comprises the following steps: adopting a preset meta path, taking each node in a network as an initial node, guiding random walk, and generating a series of walk node sequences corresponding to the nodes; by means of heterogeneous Skip-gram, based on all the walk node sequences, feature embedding of each miRNA and disease in the database is obtained through extraction; and embedding and constructing a dichotomy training sample set by adopting the characteristics corresponding to the miRNA and the disease, and training a classifier to obtain a miRNA-disease association prediction model. According to the invention, the correlation accuracy of miRNA and diseases can be improved.

Description

Technical Field

[0001] The present invention belongs to the field of biomedicine-related technologies, and more specifically, relates to a method for constructing a miRNA-disease association prediction model based on meta-path embedding. Background Art

[0002] miRNA is a class of small non-coding RNA molecules that inhibit the expression of target genes by inducing mRNA cleavage and degradation, translation inhibition or other morphological control mechanisms (Bartel DP. MicroRNAs: genomics, biogenesis, mechanism, and function [J]. cell, 2004, 116 (2): 281-297.). In recent years, many studies have shown that miRNA is related to many important biological processes (Gao P, Wong CCL, Tung EKK, et al. Deregulation of microRNA expression occurs early and accumulates in early stages of HBV-associated multistep hepatocarcinogenesis [J]. Journal of hepatology, 2011, 54 (6): 1177-1184.). Therefore, discovering more information related to miRNA and disease can provide a deeper understanding of the molecular mechanism of the disease. Predicting the potential association between miRNA and disease can effectively prevent potential disease targets at the first time and further improve the diagnosis, prognosis and treatment of complex human diseases (Hart M, Wach S, Nolte E, et al. The proto-oncogene ERG is a target of micro RNA miR-145 in prostate cancer [J]. The FEBS journal, 2013, 280 (9): 2105-2116.).

[0003] However, obtaining miRNA-disease association information through biological experiments is both laborious and expensive. Therefore, researchers are increasingly inclined to develop computational methods and algorithmic models, using existing databases to construct a network associated with miRNA and disease (i.e., miRNA-disease network), and use this network to predict potential miRNA-disease association information, thereby reducing the cost of biological experiments. Many network model-based methods have been successful in this regard (Xuan P, Han K, Guo Y, et al. Prediction of potential disease-associated microRNAs based on random walk [J]. Bioinformatics, 2015, 31 (11): 1805-1815.). As machine learning has become increasingly prominent in research, researchers have increasingly adopted these methods to train miRNA and disease-related data, and have achieved remarkable success (Chen X, Wang L, Qu J, et al. Predicting miRNA-disease association based on inductive matrix completion [J]. Bioinformatics, 2018, 34 (24): 4256-4265.). Graph machine learning methods for network models are also developing day by day. Network embedding methods are an important application of machine learning on network models. These network embedding methods can automatically extract feature embeddings of nodes from the network. Under the premise of retaining the network structure information as much as possible, low-dimensional feature vectors are used to represent the nodes in the network, making downstream link prediction, node classification and other tasks more efficient. Many general network embedding methods have been proposed, such as DeepWalk, node2vec, SDNE, etc., and have been effectively applied in bioinformatics research (Chen Z, Wang X, Gao P, et al. Predicting disease related microRNA based on similarity and topology [J]. Cells, 2019, 8 (11): 1405.). However, it is worth noting that they mainly focus on homogeneous networks, while miRNA-disease association networks are heterogeneous. Therefore, the current general network embedding methods cannot effectively mine the structural information in the miRNA-disease network to obtain the association information between miRNA and disease. A network embedding method that can be applied to the heterogeneous network of miRNA-disease association and match the appropriate algorithm model is needed to predict the miRNA-disease association. Summary of the invention

[0004] In view of the above defects or improvement needs of the prior art, the present invention provides a method for constructing a miRNA-disease association prediction model based on meta-path embedding, which aims to propose a network embedding method suitable for miRNA-disease association heterogeneous networks to improve the accuracy of miRNA and disease association.

[0005] To achieve the above object, according to one aspect of the present invention, a method for constructing a miRNA-disease association prediction model based on metapath embedding is provided, comprising:

[0006] Based on the preset miRNA and disease association database, the miRNAs and diseases in it are taken as nodes in the network, the lines connecting the known associated miRNA and disease nodes are taken as edges in the network, and the types of miRNA and disease nodes are distinguished to construct a miRNA-disease heterogeneous network;

[0007] Using a preset meta-path, taking each node in the network as the starting node, random walks are guided on the miRNA-disease heterogeneous network to generate a series of walk node sequences corresponding to the node; using heterogeneous Skip-gram, based on all walk node sequences, feature embeddings of each miRNA and disease in the miRNA and disease association database are extracted;

[0008] The features corresponding to the miRNA and the disease in the miRNA and disease association database are used to embed and construct a binary classification training sample set, and a classifier is trained to obtain a miRNA-disease association prediction model.

[0009] Furthermore, the meta-pathway is: disease→miRNA→disease.

[0010] Furthermore, the classifier is an XGBoost classifier.

[0011] Furthermore, the XGBoost classifier is trained by ensemble learning.

[0012] Furthermore, the training sample set includes:

[0013] The positive samples are constructed by embedding the features corresponding to all pairs of miRNAs and diseases with known associations in the miRNA and disease association database, and the negative samples with the same number of positive samples are constructed by embedding the features corresponding to miRNAs and diseases without association in the miRNA and disease association database.

[0014] According to another aspect of the present invention, a miRNA-disease association prediction method is provided, comprising:

[0015] The miRNA-disease association prediction model constructed by the miRNA-disease association prediction model construction method as described above predicts the association between the miRNA and the disease based on the feature embedding of the miRNA to be predicted and the disease.

[0016] According to another aspect of the present invention, an electronic device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed by a processor, the device where the storage medium is located is controlled to execute the steps of the method described above.

[0018] According to another aspect of the present invention, a computer program product is provided, comprising a computer program or instructions, wherein when the computer program or instructions are executed by a processor, the steps of the method described above are implemented.

[0019] In general, compared with the prior art, the above technical solution conceived by the present invention has the following beneficial effects:

[0020] 1. The present invention proposes a method for constructing a miRNA-disease association prediction model based on metapath embedding. When predicting potential miRNA-disease associations in a biomedical network model, the heterogeneity of the network must be taken into account. The present invention proposes a miRNA-disease heterogeneous network construction method, which contains different types of nodes-miRNA and disease. Further, in order to effectively process this heterogeneous information, the metapath is a key tool of the method of the present invention. By embedding specific semantic information (i.e., type path) in the metapath, random walks can be guided according to these paths, rather than completely random. This method enables the generated node sequence to retain more heterogeneous information, thereby enhancing the effect of subsequent predictions. In addition, when generating feature embedding using the node sequence generated by random walks, the heterogeneous Skip-gram model can consider different types of nodes without losing the heterogeneous information of the network, further improving the prediction ability of potential miRNA-disease associations. Therefore, the model constructed by the method of the present invention can effectively screen out potential miRNA-disease associations.

[0021] 2. The present invention also proposes a preferred meta-path: disease → miRNA → disease. On the one hand, the alternating appearance of disease and miRNA retains more heterogeneous information, thereby improving the predictive performance of potential associations in the miRNA-disease heterogeneous network; on the other hand, this meta-path can reveal the indirect associations between different diseases through shared miRNAs. Many diseases share the same miRNA regulatory network, which indicates that they have commonalities in molecular mechanisms. By analyzing this path, potential association information between diseases, such as comorbidity or common pathological mechanisms, can be captured. Integrating the association information of multiple diseases helps to enhance the model's predictive ability for miRNA-disease associations. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flowchart of a method for constructing a miRNA-disease association prediction model based on meta-path embedding provided by an embodiment of the present invention;

[0023] Figure 2 is a schematic diagram of a miRNA-disease association heterogeneous network provided by an embodiment of the present invention;

[0024] Figure 3 is a schematic diagram of a metapath-guided random walk on a miRNA-disease association heterogeneous network provided by an embodiment of the present invention, wherein "d" represents a disease, "r" represents a miRNA, and different numbers represent different diseases or miRNAs;

[0025] Figure 4 It is a heterogeneous Skip-gram framework diagram provided by an embodiment of the present invention.

[0026] Figure 5 is a schematic diagram of training an XGBoost model and predicting potential miRNA-disease associations provided by an embodiment of the present invention;

[0027] Figure 6 1 is a schematic diagram comparing ROC curves and AUC values ​​of the method of Example 1 provided in an embodiment of the present invention and other network embedding methods;

[0028] Figure 7 It is a schematic diagram comparing Precision and Recall of the method of Example 1 provided in an embodiment of the present invention and other network embedding methods;

[0029] Figure 8 This is a performance display diagram of the method of Example 1 provided in an embodiment of the present invention under the disturbance of original network data. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0031] Embodiment 1

[0032] A method for constructing a miRNA-disease association prediction model based on metapathway embedding, such as Figure 1 As shown, including:

[0033] Based on the preset miRNA and disease association database, the miRNAs and diseases in it are taken as nodes in the network, the lines connecting the known associated miRNA and disease nodes are taken as edges in the network, and the types of miRNA and disease nodes are distinguished to construct a miRNA-disease heterogeneous network;

[0034] Using the preset meta-path, taking each node in the network as the starting node, we guided random walks on the miRNA-disease heterogeneous network to generate a series of walk node sequences corresponding to the node; using heterogeneous Skip-gram, based on all walk node sequences, we extracted the feature embedding of each miRNA and disease in the miRNA-disease association database;

[0035] The feature embedding corresponding to miRNA and disease in the miRNA and disease association database is used to construct a binary classification training sample set, train the classifier, and obtain the miRNA-disease association prediction model.

[0036] In view of the high cost of traditional biological experiments, existing computational methods and algorithm models are difficult to effectively apply to miRNA-disease heterogeneous networks. This embodiment proposes a miRNA-disease association prediction method based on meta-path embedding, which aims to effectively screen out potential miRNA-disease associations based on existing relevant biomedical data before conducting biological experiments, thereby reducing the high cost of biological experiments.

[0037] When predicting potential miRNA-disease associations in biomedical network models, the heterogeneity of the network must be taken into account. This embodiment proposes a method for constructing a miRNA-disease heterogeneous network, which contains different types of nodes-miRNA and disease. Further, in order to effectively process this heterogeneous information, the metapath is a key tool of the method of this embodiment. By embedding specific semantic information (i.e., type path) in the metapath, random walks can be guided according to these paths rather than completely random. This method enables the generated node sequence to retain more heterogeneous information, thereby enhancing the effect of subsequent predictions. In addition, when using the node sequence generated by random walks to generate feature embeddings, the heterogeneous Skip-gram model can consider different types of nodes without losing the heterogeneous information of the network, further improving the prediction ability of potential miRNA-disease associations. Therefore, the model constructed by the method of this embodiment can effectively screen out potential miRNA-disease associations.

[0038] In the method of this embodiment, regarding the heterogeneous network, miRNA and disease are taken as nodes in the network, known miRNA-disease associations are taken as edges in the network, and the types of miRNA and disease nodes are distinguished to obtain an undirected and unweighted heterogeneous network G=(V, E, T V ), E represents the edge set of the network, V represents the node set of the network, T V ={miRNA, disease}, which represents the type set of nodes. For any type of node v (denoted as t), there is a mapping function that can be expressed as: where t∈TV ， V t .

[0039] For example, taking the human miRNA-disease association database HMDD v2.0 as an example (Li Y, Qiu C, Tu J, et al. HMDD v2.0: a database for experimentally supported human microRNA and disease associations [J]. Nucleic acids research, 2014, 42 (D1): D1070-D1074.), a miRNA-disease association heterogeneous network as described in step 1 is constructed. The data in the database needs to be preprocessed. First, all records of different miRNA copies that produce the same mature miRNA are merged into a group, for example, hsa-mir-376a-1 and hsa-mir-376a-2 are merged into the same has-mir-376a. Then, the disease terms of standard medical subject headings (MeSH) are used to unify the disease names. In order to better describe the association between miRNA and disease, an adjacency matrix is ​​used to record the association between human miRNA and disease downloaded and sorted from the HMDD v2.0 database. The elements in the matrix are 1 or 0, where 1 indicates that there is a known association between miRNA and disease, and 0 indicates that there is no known association. The constructed miRNA-disease association network contains 495 miRNAs and 383 diseases, and there are 5430 miRNA-disease associations, such as Figure 2 shown.

[0040] In the method of this embodiment, regarding the meta-path-guided random walk, the traditional random walk cannot distinguish different node types in heterogeneous networks, cannot effectively capture the semantic information in heterogeneous networks, and is not suitable for the undirected and unweighted heterogeneous network constructed above. The meta-path-guided random walk starts the random walk process by predefining a meta-path specific to the heterogeneous network. The meta-path guides the random walk so that the embedding learning is consistent with the semantics of different node types and their interconnections.

[0041] Specifically, for the following general meta-path scheme P:

[0042]

[0043] in, Indicates the node type V 1 To V l Path.

[0044] It should be noted that in the scenario of the present invention, there are only two types of nodes, as mentioned above. V={miRNA, disease}, after permutation and combination, there are four possible directed edges in the path, which can be considered as the preferred method, that is, only two directed edge combinations are considered, namely disease to miRNA and miRNA to disease. t represents the set of nodes of type t in an undirected and unweighted heterogeneous network, that is,

[0045] The transition probability of the heterogeneous random walk guided by the meta-path scheme P is as follows:

[0046]

[0047] Where t∈T V , t+1∈T V , where t and t+1 represent two node types in the network. The meta-path scheme P must be satisfied. represents the initial node of the i-th walk, whose node type is t, that is, Representation Node The set of nodes of type t+1 among the neighboring nodes (i.e. associated nodes) in the network, i.e. The node representing the next random walk (i+1th step) needs to satisfy the following conditions under the guidance of the meta-path scheme P: Only probability As the node of the next random walk, otherwise the probability is 0.

[0048] In addition, the meta-path is usually defined in a symmetric form, that is, in the node sequence generated based on the meta-path random walk, the first node V 1 With the last node V l The same, that is:

[0049] Considering the specific miRNA-disease association prediction, the node sequence generated by the walk contains more semantic information and is interpretable. The following meta-path is preferably defined: "disease-miRNA-disease".

[0050] For example, in the above-mentioned heterogeneous network of miRNA-disease association, the metapath “disease-miRNA-disease” can be defined symmetrically. Algorithm 1 shows the random walk process of generating a set of node sequences under the guidance of a metapath. Figure 3 a sequence of lines in .

[0051] Algorithm 1 Path-guided Random Walk

[0052]

[0053] In the method of this embodiment, regarding the extraction of feature embedding, the above-mentioned feature embedding of miRNA and disease extracted by heterogeneous Skip-gram is to obtain a node sequence containing semantic information and interpretability under the guidance of the meta-path as described above. For a given node v in the sequence, it is necessary to maximize the probability of other nodes appearing in the context window of the given node in the sequence, that is, it is necessary to maximize the probability of the adjacent nodes of each node appearing around the node, and learn the effective representation of each node. The overall learning objectives are as follows:

[0054]

[0055] Among them, consider all nodes v∈V, as mentioned above T V Represents the set of node types in the heterogeneous network. For each node type t∈T in the network V , N t (v) is the set of nodes of type t in the context nodes of the given node v (i.e., the nodes within the window size range centered on the node v). t ∈N t (v), p(c t |v;θ) is the probability to be maximized, which is usually defined as a softmax function, and θ is the parameter. Furthermore, in order to consider different types of nodes in the process of learning feature embedding, heterogeneous Skip-gram is used in the softmax function to calculate N t Any node c in (v) t Perform softmax calculation as follows:

[0056]

[0057] X v It is the row corresponding to node v in the feature matrix X, that is, the embedded feature vector of node v.

[0058] Note that the softmax function used in the heterogeneous Skip-gram process has the denominator It is necessary to calculate all nodes of type t, which obviously causes huge computational costs, especially on large networks, this cost is even more unbearable. In the face of such a problem, the heterogeneous negative sampling method is used to optimize the process. Specifically, setting the negative sample sampling amount to M, the softmax function can be updated to:

[0059]

[0060] Its gradient can be derived as follows:

[0061]

[0062] Among them, σ(·) is the sigmoid function, that is, Indicates randomly sampling a negative sample of category t And repeat M times, noting that It represents a node Is it node c? t The indicator function stipulates that when m = 0, That is as follows:

[0063]

[0064] By traversing all nodes v∈V in the network and repeating meta-path-guided random walks and heterogeneous Skip-gram, feature embeddings of miRNAs and diseases can be extracted.

[0065] For example, for each set of node sequences, the above heterogeneous Skip-gram is used to obtain the feature embedding of each miRNA and disease. Figure 4 As shown, the meaning is that for a given context node v, only the context nodes related to the target node c are considered. t Nodes u∈V of the same category t , only they may appear in the context node set N of node type t of node v t (v), among these nodes, the target node c t At the same time, the heterogeneous negative sampling method is considered to optimize the heterogeneous Skip-gram process, which greatly reduces the computational cost. The specific process is shown in Algorithm 2 below.

[0066] Algorithm 2 Heterogeneous Skip-gram optimized by heterogeneous negative sampling method

[0067]

[0068]

[0069] Taking each node as the starting node, the meta-path-guided random walk and heterogeneous Skip-gram processes are repeated, and finally high-quality miRNA and disease feature embedding containing network structure information and semantic information is obtained. The specific process is shown in Algorithm 3.

[0070] Algorithm 3: miRNA and disease feature embedding

[0071]

[0072] In the method of this embodiment, the construction of the training sample set can be used as an optimal binary classification data set for generating a balanced positive and negative sample. Potential miRNA-disease association prediction can be regarded as a binary classification problem. The miRNA embedding features and disease embedding features extracted above are spliced, and the spliced ​​vector is used as the input of the binary classification, and the association between the two is used as the output of the binary classification.

[0073] A good binary classification model usually needs to be trained using positive and negative samples. i and diseases j As a positive sample (r i , d j , 1). For negative samples, if all miRNA-disease pairs without known association are taken as negative samples, too many negative samples will be generated, resulting in an extreme imbalance of positive and negative samples in the training set, affecting the performance of the binary classification model. The present invention generates negative samples with the same number and form as the positive samples through sampling, that is, (r m , d n , 0), and finally a binary classification data set with balanced positive and negative samples can be generated.

[0074] In the method of this embodiment, regarding classifier training, it can be preferred that an XGBoost classifier be selected for training. The XGBoost classifier has excellent performance for the classification of high-dimensional feature vectors. It can be further preferred that an ensemble learning method be selected for training, which has good generalization ability. The XGBoost classifier is trained using the data set generated above.

[0075] The trained XGBoost classifier is used to predict the unknown miRNA-disease association. The features of the miRNA-disease pairs with unknown association are spliced ​​and input into the XGBoost classifier. If the output is 1, it is predicted that there is an association between the two, and if the output is 0, there is no association.

[0076] For example, after the above-mentioned feature embedding of miRNA and disease is obtained, all the miRNA r i and diseases j As a positive sample (r i , d j , 1). For negative samples, if all miRNA-disease pairs with no known association are taken as negative samples, too many negative samples will be generated, resulting in an extreme imbalance of positive and negative samples in the training set, affecting the performance of the binary classification model. A typical approach is to sample miRNA-disease pairs with unknown associations to generate negative samples with the same number and form as the positive samples, i.e. (r m , dn , 0), and finally a binary classification data set with balanced positive and negative samples can be generated. This data set is used to train the XGBoost classifier, and the trained XGBoost classifier can be used to predict unknown associated miRNA-disease pairs. The unknown associated miRNA-disease pairs are spliced ​​with features and input into the XGBoost classifier. If the output is 1, it is predicted that there is an association between the two, and if the output is 0, there is no association, such as Figure 5 shown.

[0077] The effectiveness of the method proposed in the present invention was verified through a series of evaluation methods, such as various evaluation indicators under five-fold cross validation, and case analysis of two diseases, renal cell carcinoma and breast tumor, as follows.

[0078] The following indicators are now introduced to verify the effectiveness of the present invention in discovering potential miRNA-disease associations:

[0079] (1) Confusion matrix: It can be used to measure the accuracy of the model prediction results. The confusion matrix is ​​usually described by four quantities, namely, true positive (TP), false positive (FP), true negative (TN) and false negative (FN). Among them, the true positive refers to the positive sample that is actually predicted to be a positive sample, the false positive refers to the negative sample that is actually predicted to be a positive sample, the true negative refers to the negative sample that is actually predicted to be a negative sample, and the false negative refers to the positive sample that is actually predicted to be a negative sample. It can be expressed as the following Table 1:

[0080] Table 1 Confusion matrix

[0081]

[0082] (2) True Positive Rate (TPR) / Recall: It indicates the proportion of samples that are actually positive examples that are correctly judged as positive examples by the classifier. That is, among all the true positive examples, the proportion of the number of samples that are classified as positive examples to the total number of positive examples. The calculation formula of TPR / Recall is:

[0083] (3) False Positive Rate (FPR): It indicates the proportion of samples that are misclassified as positive examples among all negative examples. That is, among all negative examples, the proportion of samples that are misclassified as positive examples to the total number of negative examples. The calculation formula of FPR is:

[0084] (4) Receiver Operating Characteristic Curve (ROC): It is a curve plotted with different thresholds as the horizontal axis and the true positive rate (y-axis) and false positive rate (x-axis) as the vertical axis. The threshold can be the class probability output by the classifier. It reflects the sensitivity and specificity of the model at different thresholds.

[0085] (5) Area Under ROC Curve (AUC): Indicates the area covered under the ROC curve. A larger AUC value means that the model has a higher prediction accuracy. Usually, the AUC value ranges from 0.5 to 1, where 0.5 represents random guessing and 1 represents a perfect classifier.

[0086] (6) Precision: refers to the proportion of samples that are actually positive among all samples predicted to be positive. In other words, Precision measures the proportion of true positive samples in the model prediction results. Its calculation formula is: The value range of Precision is also 0 to 1. The higher the value, the higher the accuracy of the positive samples in the model prediction results.

[0087] First, the ability of the present invention to discover potential miRNA-disease associations was verified by a five-fold cross-validation method and compared with other methods. Given that most diseases in the HMDD v2.0 database are associated with only a few miRNAs, we only evaluated 15 common diseases associated with at least 80 miRNAs. Specifically, for a specific disease d, the miRNAs associated with d were randomly divided into 20% as a test set and 80% as a training set. Table 2 shows the AUC values ​​of different methods for predicting miRNAs associated with different diseases. It can be found that the present invention always maintains a high accuracy rate and effectively predicts miRNA-disease associations.

[0088] Table 2 AUC values ​​of different methods for predicting miRNAs associated with different diseases

[0089]

[0090] The present invention uses a metapath-based embedding method and compares it with other network embedding methods, including Deepwalk, GraRep, and SDNE. The rest of the settings are consistent with the methods involved in the present invention. The top 15 common diseases associated with miRNAs were also used for 5-fold cross-validation evaluation. For the final miRNA-disease association prediction results, the three indicators of AUC, Precision, and Recall were used to measure the prediction performance of the model. The AUC values ​​are shown in Table 3 below, and the ROC curves are shown in Figure 6 In particular, for Precision and Recall, the first k samples in the predicted associated prediction candidate set are selected for evaluation, as Figure 7 The results show that the method of the present invention is superior to other methods as a whole, achieving the highest AUC value of 0.958 among the 15 common diseases evaluated, which may be attributed to its meta-path-guided random walk and heterogeneous Skip-gram, which is more suitable for the heterogeneous network of miRNA-disease.

[0091] Table 3 AUC values ​​of different network embedding methods for predicting miRNAs associated with different diseases

[0092]

[0093] The method described in the present invention has good robustness by randomly modifying the association between miRNA and disease. Specifically, 1% of the edges in the network are randomly deleted or inserted each time. After each deletion or insertion, the association network is reconstructed using the data, predicted using the present invention, and the performance is evaluated under 5-fold cross validation. The average AUC under 15 disease data is selected as the evaluation index and compared with the model performance under the original data. The comparison results are shown in Figure 2. Figure 8 It can be found that the AUC shows only small fluctuations as the proportion of link modifications changes, and shows a slight downward trend as the percentage of deletions increases. This result demonstrates the robustness of the present invention, and the stability of the performance despite network perturbations highlights the reliability of the method described in the present invention in capturing and utilizing the inherent structure and semantic relationships within the network.

[0094] In order to further evaluate the effectiveness of the present invention, two important human diseases, renal cell carcinoma and breast tumor, were used as examples to perform validation analysis using three miRNA-disease association databases: HMDD v2.0, dbDEMC (Yang Z, Ren F, Liu C, et al. dbDEMC: a database of differentially expressed miRNAs in human cancers [C] / / BMC genomics. BioMed Central, 2010, 11: 1-8.) and miR2Disease (Jiang Q, Wang Y, Hao Y, et al. miR2Disease: amanually curated database for microRNA deregulation in human disease [J]. Nucleic acids research, 2009, 37 (suppl_1): D98-D104.). First, the present invention was used to predict potential miRNAs associated with renal cell carcinoma, and the predicted miRNA candidate set was validated. Other candidate miRNAs in HMDD v2.0 that are not related to the studied disease are predicted, and they are ranked according to the predicted association scores. The top 30 miRNA candidate sets with the association scores are verified in two other miRNA-disease association databases, dbDEMC and miR2Disease. The verification results are shown in Table 4. As a case study result of renal cell carcinoma, the top 10 miRNAs obtained by the prediction method of the present invention are all related to renal cell carcinoma, and 28 of the top 30 miRNAs are related to renal cell carcinoma through verification by dbDEMC and miR2Disease databases. This case study demonstrates the excellent performance of the method of the present invention in predicting the association between miRNA and disease from another perspective.

[0095] Table 4 Prediction results of miRNAs related to renal cell carcinoma

[0096]

[0097]

[0098] In addition, a second case study on breast tumors was conducted, which verified that the method described in the present invention is applicable to new diseases with unknown associated miRNAs. Breast tumors were considered as new diseases by deleting all known associations between breast tumors and miRNAs in HMDD V2.0. Slightly different from the verification of renal cell carcinoma, the HMDD v2.0 association database was added to verify the results. The results are shown in Table 5. Among the top 10 and top 30 predicted breast tumor-related miRNAs, 10 and 29 miRNAs passed the verification of HMDD v2.0, dbDEMC and miR2Disease, respectively. In the list of predicted candidate miRNA results, the higher the prediction score, the higher the ranking, and the higher the possibility of potential association with breast cancer. Obviously, hsa-mir-210, which ranks first, has the greatest potential relevance to breast tumors. The relevance of hsa-mir-210 overexpression in the contralateral unaffected breast of breast cancer patients has been revealed (Shidfar A, Costa FF, Scholtens D, et al. Expression of miR-18a and miR-210 in normal breast tissue as candidate biomarkers of breast cancer risk [J]. Cancer Prevention Research, 2017, 10 (1): 89-97.). This case study shows that the method of the present invention is also reliable in predicting miRNAs associated with new diseases.

[0099] Table 5 Results of miRNAs predicted to be associated with breast tumors (considered as new diseases)

[0100]

[0101]

[0102] The present invention elaborates on its principle and implementation method by applying the HMDD v2.0 public authoritative database and verifies its effectiveness.

[0103] The method of this embodiment can further capture the common associations between diseases: the disease-miRNA-disease metapathway can reveal the indirect associations between different diseases through shared miRNAs. Many diseases may share the same miRNA regulatory network, which means that they may have commonalities in molecular mechanisms. In addition, the underlying biological mechanisms can be revealed: miRNA plays a key role in regulating gene expression and cell function, and these functions are often associated with a variety of diseases. The disease-miRNA-disease metapathway helps to deeply understand the specific mechanism of action of miRNA in different diseases. For example, by analyzing the regulatory pattern of a certain miRNA in multiple related diseases, its key role in the occurrence and development of the disease can be revealed, thereby guiding experimental verification and functional research.

[0104] Embodiment 2

[0105] The present application also relates to an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0106] The electronic device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The memory may be used to store computer programs and / or modules, and the processor may perform various functions of the electronic device by running or executing computer programs and / or modules stored in the memory, and calling data stored in the memory.

[0107] The relevant technical solutions are the same as above and will not be described in detail here.

[0108] Embodiment 3

[0109] The present application also relates to a computer-readable storage medium having a computer program stored thereon, and the steps of the above method are implemented when the computer program is executed by a processor.

[0110] Specifically, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as a hard disk, an internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0111] The relevant technical solutions are the same as above and will not be described again here.

[0112] Embodiment 4

[0113] The embodiment of the present application provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the steps of the method of the above embodiment of the present application.

[0114] The relevant technical solutions are the same as above and will not be described again here.

[0115] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for constructing a miRNA-disease association prediction model based on metapath embedding, characterized in that: include: Based on the preset miRNA and disease association database, the miRNAs and diseases in it are taken as nodes in the network, the lines connecting the known associated miRNA and disease nodes are taken as edges in the network, and the types of miRNA and disease nodes are distinguished to construct a miRNA-disease heterogeneous network; Using a preset meta-path, taking each node in the network as the starting node, guiding random walks on the miRNA-disease heterogeneous network, and generating a series of walk node sequences corresponding to the node; using heterogeneous Skip-gram, based on all walk node sequences, extracting the feature embedding of each miRNA and disease in the miRNA and disease association database; The features corresponding to the miRNA and the disease in the miRNA and disease association database are used to embed and construct a binary classification training sample set, and a classifier is trained to obtain a miRNA-disease association prediction model.

2. The method for constructing a miRNA-disease association prediction model according to claim 1, wherein: The meta-pathway is: disease→miRNA→disease.

3. The method for constructing a miRNA-disease association prediction model according to claim 1, wherein: The classifier is an XGBoost classifier.

4. The method for constructing a miRNA-disease association prediction model according to claim 3, wherein: The XGBoost classifier is trained by ensemble learning.

5. The method for constructing a miRNA-disease association prediction model according to claim 1, wherein: The training sample set includes: The positive samples are constructed by embedding the features corresponding to all pairs of miRNAs and diseases with known associations in the miRNA and disease association database, and the negative samples with the same number of positive samples are constructed by embedding the features corresponding to miRNAs and diseases without association in the miRNA and disease association database.

6. A miRNA-disease association prediction method, characterized in that: include: The miRNA-disease association prediction model constructed by the miRNA-disease association prediction model construction method according to any one of claims 1 to 5 is used to predict the association between the miRNA and the disease based on the feature embedding of the miRNA to be predicted and the disease.

7. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 and / or the steps of the method according to claim 6 are implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed by a processor, the device where the storage medium is located is controlled to execute the steps of the method according to any one of claims 1 to 5 and / or the steps of the method according to claim 6.

9. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 5 and / or the steps of the method according to claim 6 are implemented.

Images