Association prediction model training method, association prediction method and association prediction system

Abstract

The invention provides a correlation prediction model training method, a correlation prediction method and a correlation prediction system, and the method comprises the steps: obtaining a plurality of training samples which are obtained through the construction of first sample multi-source features and second sample multi-source features with correlation labels, the first sample multi-source feature is a multi-source feature corresponding to sample micro ribonucleic acid, and the second sample multi-source feature is a multi-source feature corresponding to a sample physiological state type; inputting the plurality of training samples into a preset model, and training the preset model based on vertex vector representation output by a fusion feature layer of the preset model to obtain a trained association prediction model; and the fusion feature layer is used for carrying out fusion processing on the global attention fusion feature representation and the local attention fusion feature representation to obtain vertex vector representation. According to the invention, the association prediction accuracy of miRNA and physiological status types is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of biological information association prediction, and in particular to an association prediction model training method, an association prediction method and a system. Background Art

[0002] MicroRNA (miRNA) is a type of non-coding single-stranded RNA molecule with a length of about 22 nucleotides encoded by endogenous genes. It has many important regulatory functions in living organisms, including cell differentiation, proliferation and apoptosis, as well as transcription and transport of genetic material.

[0003] Existing graph neural network-based miRNA and physiological state type (such as biological development, metabolism and other physiological processes) association prediction models usually only use a single information source, and use the same set of attention coefficients when fusing multiple perspective features of different vertices, resulting in a low final association prediction accuracy.

[0004] Therefore, there is an urgent need for an association prediction model training method, an association prediction method and a system to solve the above problems. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides an association prediction model training method, an association prediction method and a system.

[0006] The present invention provides an association prediction model training method, comprising: Acquire multiple training samples, where the training samples are constructed from a first sample multi-source feature and a second sample multi-source feature having an associated relationship label, wherein the first sample multi-source feature is a multi-source feature corresponding to the sample microRNA, and the second sample multi-source feature is a multi-source feature corresponding to the sample physiological state type; Inputting a plurality of the training samples into a preset model, and training the preset model based on the vertex vector representation output by the fusion feature layer of the preset model to obtain a trained association prediction model; Among them, the fusion feature layer is used to fuse the global attention fusion feature representation output by the global attention fusion layer of the preset model and the local attention fusion feature representation output by the local attention fusion layer of the preset model to obtain the vertex vector representation.

[0007] According to an association prediction model training method provided by the present invention, the obtaining of multiple training samples includes: Acquire a first similarity network and a second similarity network, wherein the first similarity network is a similarity network between the sample microRNAs, and the second similarity network is a similarity network between the sample physiological state types; Constructing corresponding node aggregation embedding according to network structure information of each node in the first similarity network and the second similarity network, wherein the network structure information is extracted based on a graph convolutional network; According to the node aggregation embedding corresponding to the first similarity network, a network node representation of the sample microRNA is obtained; according to the node aggregation embedding corresponding to the second similarity network, a network node representation of the sample physiological state type is obtained; When it is determined that there is a correlation between the sample microRNA and the sample physiological state type, the network node representation of the sample microRNA is used as the first sample multi-source feature, and the network node representation of the sample physiological state type corresponding to the sample microRNA is used as the second sample multi-source feature to construct the training sample.

[0008] According to a method for training an association prediction model provided by the present invention, after inputting the plurality of training samples into a preset model, the method further comprises: Based on the preset activation function in the global attention fusion layer and the corresponding network node representation, a first attention coefficient and a second attention coefficient are calculated, wherein the first attention coefficient is the attention coefficient corresponding to the network node representation of the sample microRNA, and the second attention coefficient is the attention coefficient corresponding to the network node representation of the sample physiological state type; Based on the first preset convolutional neural network in the global attention fusion layer, a first global attention fusion feature representation is calculated according to the first attention coefficient and the network node representation of the sample microRNA, wherein the first global attention fusion feature representation is the global attention fusion feature representation corresponding to the sample microRNA; a second global attention fusion feature representation is calculated according to the second attention coefficient and the network node representation of the sample physiological state type, wherein the second global attention fusion feature representation is the global attention fusion feature representation corresponding to the sample physiological state type; Based on the global attention fusion layer, the first global attention fusion feature representation and the second global attention fusion feature representation are input into the fusion feature layer.

[0009] According to a method for training an association prediction model provided by the present invention, after inputting the plurality of training samples into a preset model, the method further comprises: Based on the attention mechanism in the local attention fusion layer and the second preset convolutional neural network, a first local attention fusion feature representation is calculated according to the first node aggregate embedding, wherein the first node aggregate embedding is a node aggregate embedding corresponding to a node in the first similarity network; Based on the attention mechanism in the local attention fusion layer and the second preset convolutional neural network, a second local attention fusion feature representation is calculated according to the second node aggregate embedding, wherein the second node aggregate embedding is the node aggregate embedding corresponding to the node in the second similarity network; Based on the local attention fusion layer, the first local attention fusion feature representation and the second local attention fusion feature representation are input into the fusion feature layer.

[0010] According to a method for training an association prediction model provided by the present invention, the method further includes: Based on the attention mechanism of the fusion feature layer, the first global attention fusion feature representation and the first local attention fusion feature representation are fused and embedded to obtain a first graph aggregation embedding, wherein the first graph aggregation embedding is a graph aggregation embedding corresponding to the nodes in the first similarity network; Based on the attention mechanism of the fusion feature layer, the second global attention fusion feature representation and the second local attention fusion feature representation are fused and embedded to obtain a second graph aggregation embedding, wherein the second graph aggregation embedding is a graph aggregation embedding corresponding to the nodes in the second similarity network; Calculating the interaction embedding between the nodes in the first similarity network and the nodes in the second similarity network according to the interaction embedding of the first graph aggregate embedding and the second graph aggregate embedding; The first graph aggregate embedding, the interaction embedding and the second graph aggregate embedding are processed by a connection operation to obtain the vertex vector representation.

[0011] According to a method for training an association prediction model provided by the present invention, the preset model is trained based on the vertex vector representation output by the fusion feature layer of the preset model to obtain a trained association prediction model, including: Based on the fully connected layer in the preset model, according to the vertex vector representation, predicting the corresponding association prediction result between the first sample multi-source feature and the second sample multi-source feature in the training sample; Based on the cross entropy loss function, performing loss value analysis on the association prediction result and the association relationship label corresponding to the training sample to obtain a loss value; When it is determined that the loss value meets the preset threshold, the trained association prediction model is obtained.

[0012] The present invention also provides an association prediction method, comprising: The multi-source features corresponding to the target micro RNA and the multi-source features corresponding to the target physiological state type are input into the trained association prediction model to obtain the association prediction result between the target micro RNA and the target physiological state type output by the trained association prediction model, wherein the trained association prediction model is trained based on the above-mentioned association prediction model training method.

[0013] The present invention also provides an association prediction model training system, comprising: A sample acquisition module, used to acquire a plurality of training samples, wherein the training samples are constructed by a first sample multi-source feature and a second sample multi-source feature having an associated relationship label, wherein the first sample multi-source feature is a multi-source feature corresponding to the sample microRNA, and the second sample multi-source feature is a multi-source feature corresponding to the sample physiological state type; A model training module, used for inputting a plurality of the training samples into a preset model, and training the preset model based on the vertex vector representation output by the fusion feature layer of the preset model to obtain a trained association prediction model; Among them, the fusion feature layer is used to fuse the global attention fusion feature representation output by the global attention fusion layer of the preset model and the local attention fusion feature representation output by the local attention fusion layer of the preset model to obtain the vertex vector representation.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the associated prediction model training method described above is implemented, or the associated prediction method described above is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described association prediction model training methods, or the above-described association prediction method.

[0016] The association prediction model training method, association prediction method and system provided by the present invention obtain the final vector representation of the vertex through the steps of multi-view feature extraction, vertex global attention fusion feature representation acquisition, vertex local attention fusion feature representation acquisition and fusion feature integration, and based on the final vector representation of the vertex, predict the association relationship between potential miRNA and physiological state type by training the fully connected network in the preset model, thereby improving the accuracy of association prediction between miRNA and physiological state type. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or 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 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.

[0018] Figure 1 A schematic diagram of the process of the association prediction model training method provided by the present invention; Figure 2 A schematic diagram of a feature information aggregation structure based on a graph neural network provided by the present invention; Figure 3 A schematic diagram of the structure of the global attention fusion layer provided by the present invention; Figure 4 A schematic diagram of the structure of the local attention fusion layer provided by the present invention; Figure 5 A schematic diagram of the structure of the fusion feature layer provided by the present invention; Figure 6 A schematic diagram of the structure of the association prediction model training system provided by the present invention; Figure 7 This is a schematic structural diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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.

[0020] In the field of bioinformatics, the prediction of the association between miRNA and physiological state types is an important and challenging task. In addition to associating the physiological state types with miRNAs in the existing database, new miRNAs added to the database can also be associated with the corresponding physiological state types, which helps to reveal the mechanism of action of miRNA in the occurrence and development of physiological states.

[0021] Graph Neural Networks (GNN) are widely used in the field of bioinformatics due to their excellent ability to process complex graph structured data. However, a common limitation of the current GNN-based miRNA and physiological state type association prediction models is that these existing technologies often rely only on a single information source for modeling, which limits the further improvement of the model prediction performance. In addition, existing multi-source information integration models generally only use traditional attention mechanisms to learn the global importance of features from different perspectives. The same set of attention coefficients are used when different vertices fuse multiple perspective features, but different perspectives may not contribute the same amount to the feature representation of different vertices, which also affects the association prediction results.

[0022] In response to the above-mentioned problems existing in the prior art, the present invention provides a multi-view graph neural network model based on global and local attention fusion (Global and Local Attention-based Link Prediction, referred to as GLALP) for predicting the association between miRNA and physiological state types.

[0023] Figure 1 A flow chart of the association prediction model training method provided by the present invention is shown in FIG. Figure 1 As shown, the present invention provides an association prediction model training method, comprising: Step 101, obtain multiple training samples, where the training samples are constructed by a first sample multi-source feature and a second sample multi-source feature having an associated relationship label, wherein the first sample multi-source feature is a multi-source feature corresponding to the sample microRNA, and the second sample multi-source feature is a multi-source feature corresponding to the sample physiological state type.

[0024] In the graph structure, nodes represent entities (such as miRNA and genes, etc.), and edges represent the relationship between entities (such as the binding of miRNA and target genes, etc.). In the field of bioinformatics, graph neural networks can effectively integrate and utilize these complex biological network data.

[0025] In order to train the graph neural network model, the present invention first collects a large number of data sources, including multi-source feature data of miRNA and multi-source feature data of physiological state types, which come from different biological databases or experimental measurements. Among them, the first sample multi-source features (miRNA related) may include miRNA expression levels, sequence features, interaction networks with other molecules, and functional annotations, etc. These features together constitute a multidimensional vector describing the characteristics of miRNA. The second sample multi-source features (physiological state related) may include biochemical indicators (such as blood test results), imaging manifestations, and genetic variations, etc. These features reflect the physiological state from different angles.

[0026] Furthermore, miRNAs and physiological state types were used as nodes in the graph structure, and edges were constructed based on the associations between them (such as known miRNA-physiological state associations, correlations between miRNA expression changes and physiological states, etc.), thus forming a complex graph network containing miRNA nodes, physiological state nodes, and the relationships between them.

[0027] Furthermore, one or more association labels are assigned to each training sample. These labels represent the specific association or classification results (such as a certain metabolic state) between miRNA and physiological state. These labels are the goals of model training and are used to guide the model to learn the mapping relationship between miRNA features and physiological state.

[0028] Step 102, inputting a plurality of the training samples into a preset model, and training the preset model based on the vertex vector representation output by the fusion feature layer of the preset model to obtain a trained association prediction model; Among them, the fusion feature layer is used to fuse the global attention fusion feature representation output by the global attention fusion layer of the preset model and the local attention fusion feature representation output by the local attention fusion layer of the preset model to obtain the vertex vector representation.

[0029] In the present invention, multiple training samples containing known associations are input into a preset model. These training samples contain data from different sources, such as multi-source features of sample microRNA (miRNA) and multi-source features of sample physiological state types. There are associations between them that need to be learned and predicted by the model.

[0030] One of the core parts of the preset model in the present invention is its fusion feature layer, which is used to integrate feature information from different levels, that is, to fuse the feature representations output by the global attention fusion layer and the local attention fusion layer in order to more comprehensively understand the complex relationship between the data. Figure 2 This is a schematic diagram of the feature information aggregation structure based on the graph neural network provided by the present invention, which can be referenced Figure 2 As shown, the present invention adaptively learns the importance contribution of different perspectives on multi-perspective features, thereby obtaining a vector representation of the vertex.

[0031] In the present invention, the global attention fusion layer focuses on capturing and integrating feature information in a global scope, analyzing the nodes and edges in the entire data set or graph to generate a global attention fusion feature representation. This global perspective helps the model understand the macro trends and global structures in the data. In contrast, the local attention fusion layer focuses more on the local neighborhood information around the node. It generates a local attention fusion feature representation by considering the direct neighbors or closer neighbors of the node. This local perspective helps the model capture the local dependencies and interaction details between nodes.

[0032] In the present invention, the fusion feature layer combines the global attention fusion feature representation and the local attention fusion feature representation through weighted sum, splicing or nonlinear transformation, and generates the final vertex vector representation. This vertex vector representation contains both global macro information and local micro details, and can more comprehensively reflect the characteristics and attributes of the node (or sample). Based on these vertex vector representations, the preset model is trained. The goal of the training is to optimize the parameters of the model so that the model can accurately predict the association relationship in the training samples. For example, the model parameters are adjusted by comparing the difference between the model's prediction results and the true label (such as cross entropy loss). After multiple iterations of training, the model will gradually converge to a stable and good performance state, and a trained association prediction model will be obtained. This model can use the rules and features learned from the training data to accurately predict the association of new and unseen samples.

[0033] The association prediction model training method provided by the present invention obtains the final vector representation of the vertex through the steps of multi-view feature extraction, vertex global attention fusion feature representation acquisition, vertex local attention fusion feature representation acquisition and fusion feature integration, and based on the final vector representation of the vertex, predicts the association relationship between potential miRNA and physiological state type by training the fully connected network in the preset model, thereby improving the accuracy of association prediction between miRNA and physiological state type.

[0034] Based on the above embodiment, the step of obtaining a plurality of training samples includes: Acquire a first similarity network and a second similarity network, wherein the first similarity network is a similarity network between the sample microRNAs, and the second similarity network is a similarity network between the sample physiological state types; Constructing corresponding node aggregation embedding according to network structure information of each node in the first similarity network and the second similarity network, wherein the network structure information is extracted based on a graph convolutional network; According to the node aggregation embedding corresponding to the first similarity network, a network node representation of the sample microRNA is obtained; according to the node aggregation embedding corresponding to the second similarity network, a network node representation of the sample physiological state type is obtained; When it is determined that there is a correlation between the sample microRNA and the sample physiological state type, the network node representation of the sample microRNA is used as the first sample multi-source feature, and the network node representation of the sample physiological state type corresponding to the sample microRNA is used as the second sample multi-source feature to construct the training sample.

[0035] In the present invention, the multi-source features of sample miRNA and sample physiological state type, i.e., the first sample multi-source features and the second sample multi-source features, can be extracted respectively through the multi-view feature extraction layer in the preset model, so as to construct training samples. The present invention first obtains the similarity network corresponding to each of the sample miRNA and the sample physiological state type. In the graph network structure, the nodes represent the sample miRNA or the sample physiological state type, and the edges represent the similarity between the sample miRNAs or the similarity between the sample physiological state types, for example, the functional similarity between the sample miRNAs, the sequence similarity between the sample miRNAs, and the Gaussian interaction similarity between the sample miRNAs, or the semantic similarity between the physiological state types, the target similarity between the physiological state types, and the Gaussian interaction association similarity between the physiological state types. In the present invention, the existing similarity calculation method can be selected to calculate the similarity network of the sample miRNA and the sample physiological state type, so as to provide a data basis for the subsequent association prediction method of the miRNA and the sample physiological state type.

[0036] Specifically, after similarity calculation, a similarity network of sample miRNA (i.e., a first similarity network) and a similarity network of sample physiological state types (i.e., a second similarity network) are obtained. The first similarity network can be specifically expressed as:

[0037] in, , and , is the adjacency matrix of the sample miRNA node pair association network in the graph network structure, which respectively represents the functional similarity between sample miRNAs, the sequence similarity between sample miRNAs, and the Gaussian interaction similarity between sample miRNAs; The second similarity network can be specifically expressed as:

[0038] in, , and , is the adjacency matrix of the association network of sample physiological state type nodes in the graph network structure. Taking the graph network structure constructed by the physiological state type information in the current database as an example, they represent the semantic similarity between physiological state types, the target similarity between physiological state types, and the Gaussian interaction association similarity between physiological state types.

[0039] Furthermore, based on the graph convolutional network, the present invention extracts various node network structure information in the similarity network and constructs node aggregation embedding on different similarity networks, which can be obtained by the following node aggregation embedding formula: , ;

[0040] in, , , , is the initial feature vector of the association network of sample miRNA node pairs in the graph network structure; is the initial feature vector of the association network of the sample physiological state type node pair in the graph network structure; represents the identity matrix; yes The degree matrix of yes The degree matrix of and Represents the graph convolutional network model Node aggregation embedding of layers; and represents the corresponding learnable weight matrix, is a non-linear activation function.

[0041] In the present invention, a two-layer graph convolutional network model is constructed for the association prediction of sample miRNA and sample physiological state type. After obtaining the corresponding node aggregation embedding through the above node aggregation embedding formula, two types of node representations of sample miRNA and sample physiological state type can be obtained: ; ; in, A network node representation representing the sample microRNA, i.e., the first sample multi-source feature; The network node representation representing the sample physiological state type, i.e., the second sample multi-source feature.

[0042] On the basis of the above embodiment, after inputting the plurality of training samples into the preset model, the method further includes: Based on the preset activation function in the global attention fusion layer and the corresponding network node representation, a first attention coefficient and a second attention coefficient are calculated, wherein the first attention coefficient is the attention coefficient corresponding to the network node representation of the sample microRNA, and the second attention coefficient is the attention coefficient corresponding to the network node representation of the sample physiological state type; Based on the first preset convolutional neural network in the global attention fusion layer, a first global attention fusion feature representation is calculated according to the first attention coefficient and the network node representation of the sample microRNA, wherein the first global attention fusion feature representation is the global attention fusion feature representation corresponding to the sample microRNA; a second global attention fusion feature representation is calculated according to the second attention coefficient and the network node representation of the sample physiological state type, wherein the second global attention fusion feature representation is the global attention fusion feature representation corresponding to the sample physiological state type; Based on the global attention fusion layer, the first global attention fusion feature representation and the second global attention fusion feature representation are input into the fusion feature layer.

[0043] In the present invention, based on the global attention fusion layer and convolutional neural network, the preset model adaptively learns the importance contribution of different perspectives on multi-perspective features to obtain the global attention fusion feature representation of the vertex. Figure 3 The structural diagram of the global attention fusion layer provided by the present invention can be referred to Figure 3 As shown in the figure, after the training samples are input into the preset model, the global attention fusion layer calculates the attention coefficients of sample miRNA and sample physiological state type nodes in different types of similarities and different hierarchical views. The attention coefficient calculation formula is as follows:

[0044] in, represents the first attention coefficient; represents the second attention coefficient; is the global average pooling corresponding to the sample miRNA, It is the global average pooling corresponding to the sample physiological state type; and is a nonlinear activation function; , , , is the trainable weight matrix; , , , is the offset vector; Furthermore, based on the first preset convolutional neural network, the global attention fusion feature representations corresponding to the sample miRNA and the sample physiological state type are calculated respectively, and the calculation formula of the global attention fusion feature representation is as follows:

[0045] in, represents the first global attention fusion feature representation; represents the second global attention fusion feature representation; and It is a two-dimensional convolutional neural network used to integrate different similarity perspectives and graph neural network layers, i.e., the first preset convolutional neural network; It is the multiplication operation of the element-level attention coefficient and the feature representation of different views.

[0046] On the basis of the above embodiment, after inputting the plurality of training samples into the preset model, the method further includes: Based on the attention mechanism in the local attention fusion layer and the second preset convolutional neural network, a first local attention fusion feature representation is calculated according to the first node aggregate embedding, wherein the first node aggregate embedding is a node aggregate embedding corresponding to a node in the first similarity network; Based on the attention mechanism in the local attention fusion layer and the second preset convolutional neural network, a second local attention fusion feature representation is calculated according to the second node aggregate embedding, wherein the second node aggregate embedding is the node aggregate embedding corresponding to the node in the second similarity network; Based on the local attention fusion layer, the first local attention fusion feature representation and the second local attention fusion feature representation are input into the fusion feature layer.

[0047] In the present invention, based on the attention mechanism and convolutional neural network in the local attention fusion layer, a preset model learns the attention coefficients of different vertices when fusing multi-view features to obtain the local attention fusion feature representation. Figure 4 The structural diagram of the local attention fusion layer provided by the present invention can be referred to Figure 4 As shown, based on the network node representation of the sample microRNA and the network node representation of the sample physiological state type in the above embodiment, the node aggregation embedding of each corresponding node is obtained, and then these node aggregation embeddings are applied to the attention mechanism in the local attention fusion layer, and the corresponding local attention fusion feature representation is calculated by the local attention fusion feature representation calculation formula, and the formula is as follows:

[0048] ,

[0049] ,

[0050] in, The node aggregate embedding corresponding to the i-th node in the network node representation of the sample microRNA, i.e., the first node aggregate embedding; The node aggregate embedding corresponding to the jth node in the network node representation representing the sample physiological state type, i.e., the second node aggregate embedding; and is the trainable weight matrix; and is the normalization coefficient; and It is a one-dimensional convolutional neural network used to integrate different similarities and perspectives of graph neural network layers, that is, the second preset convolutional neural network.

[0051] Based on the above embodiment, the method further includes: Based on the attention mechanism of the fusion feature layer, the first global attention fusion feature representation and the first local attention fusion feature representation are fused and embedded to obtain a first graph aggregation embedding, wherein the first graph aggregation embedding is a graph aggregation embedding corresponding to the nodes in the first similarity network; Based on the attention mechanism of the fusion feature layer, the second global attention fusion feature representation and the second local attention fusion feature representation are fused and embedded to obtain a second graph aggregation embedding, wherein the second graph aggregation embedding is a graph aggregation embedding corresponding to the nodes in the second similarity network; Calculating the interaction embedding between the nodes in the first similarity network and the nodes in the second similarity network according to the interaction embedding of the first graph aggregate embedding and the second graph aggregate embedding; The first graph aggregate embedding, the interaction embedding and the second graph aggregate embedding are processed by a connection operation to obtain the vertex vector representation.

[0052] In the present invention, based on the global attention fusion feature representation and the local attention fusion feature representation obtained in the above embodiments, the final vector representation of the vertex, that is, the vertex vector representation, is obtained. Figure 5 For the structural diagram of the fusion feature layer provided by the present invention, please refer to Figure 5 As shown in Figure 1, by fusing the attention mechanism of the feature layer, the global attention fusion feature representation and the local attention fusion feature representation are combined, and the corresponding fusion embedding calculation formula is:

[0053] in, represents the graph aggregation embedding corresponding to the i-th node in the first similarity network, that is, the first graph aggregation embedding; represents the graph aggregation embedding corresponding to the j-th node in the second similarity network, that is, the second graph aggregation embedding; is a trainable parameter; Furthermore, the interaction embedding between the two types of nodes (i.e., the nodes corresponding to the sample microRNA and the nodes corresponding to the sample physiological state type) is calculated by element-wise multiplication. The calculation formula of the interaction embedding is:

[0054] in, represents the interaction embedding between the i-th node in the first similarity network and the j-th node in the second similarity network; Based on the connection operation, the two types of nodes and interaction embeddings are combined to obtain the final representation, namely the vertex vector representation:

[0055] in, is the vertex vector representation, For connection operation.

[0056] On the basis of the above embodiment, the vertex vector representation outputted by the fusion feature layer of the preset model is used to train the preset model to obtain a trained association prediction model, including: Based on the fully connected layer in the preset model, according to the vertex vector representation, predicting the corresponding association prediction result between the first sample multi-source feature and the second sample multi-source feature in the training sample; Based on the cross entropy loss function, performing loss value analysis on the association prediction result and the association relationship label corresponding to the training sample to obtain a loss value; When it is determined that the loss value meets the preset threshold, the trained association prediction model is obtained.

[0057] In the present invention, a multilayer perceptron (MLP) decoder is applied to perform link prediction, and then the loss value between the association prediction result and the association relationship label is analyzed through the loss function and the vertex vector representation obtained in the above embodiment, so as to obtain a trained association prediction model after determining that the preset training conditions are met, and the model can predict the association relationship between potential miRNA and physiological state types. Specifically, the cross entropy loss function used in the present invention is:

[0058] in, It is the association prediction result (association probability) between the sample miRNA and the sample physiological state type predicted by the fully connected layer; is the trainable weight matrix, Represents the loss value calculated using the cross entropy loss function; Indicates the association label between sample miRNA and sample physiological state type.

[0059] In one embodiment, five common evaluation indicators are measured: AUC (Area Under Curve), area under the precision and recall curve (AUPR), accuracy, F1 score, precision and recall, to verify that the performance of the association prediction model training method provided by the present invention exceeds the existing 8 association prediction models that can be used for miRNA and physiological state types.

[0060] Specifically, in this embodiment, the association process between miRNA and physiological state type data is explained, wherein the miRNA-physiological state type association data comes from the MicroRNA database (HMDD v3.2), HMDD v3.2 provides the latest and comprehensive collection of miRNA-physiological state type associations, and identifies 12,446 experimentally confirmed associations involving 853 miRNAs and 591 physiological state types, wherein positive samples are designated as known miRNA-physiological state type associations (indicated as 1), while negative samples indicate that there is no discernible association between miRNA and physiological state type, indicated as 0.

[0061] Furthermore, among the samples labeled as 0, an equal number of negative samples are randomly selected as positive samples to maintain sample balance. Subsequently, the dataset is divided into a training set and a test set, of which the test set accounts for 20%. In order to ensure that the model has robust generalization performance, a 5-fold cross-validation is used in the training set to fine-tune the model hyperparameters. Specifically, all samples in the dataset are divided into five subsets of equal size, and the execution is repeated five times, each subset is used as a test set in turn, while the other four subsets are used as training sets, and then tested on an independent test set and compared with other baseline models.

[0062] In this embodiment, GLALP was implemented based on Torch, and Adam was used to optimize the trainable model variables. In order to avoid overfitting, a standard dropout technique was used. During the training phase, hidden units were randomly dropped after the final fully connected layer. Based on 5-fold cross-validation, the learning rate was finally set to 0.0005, the dropout rate was set to 0.5, the number of GNN layers L = 2, and the embedding dimension k = 128. The model outputs a probability between 0 and 1 for each potential edge of the miRNA-physiological state type pair, using 0.5 as the threshold for link prediction. If the probability is greater than or equal to 0.5, the pair is predicted to be linked (i.e., it is a positive pair); otherwise, the pair is unlinked.

[0063] Furthermore, five common evaluation metrics are used to measure performance: AUC, Area Under the Precision and Recall Curve (AUPR), Accuracy, F1 Score, Precision, and Recall. AUC describes the probability that the model ranks positive samples before negative samples; AUPR is particularly effective for unbalanced datasets and situations where the positive class is more concerned. The other three evaluation metrics are F1 score, accuracy, and recall.

[0064] Then, GLALP is compared with eight other link prediction methods, including NIMCGCN, MMGCN, ERMDA, HGANMDA, AGAEMD, MINIMDA, MAGCN and AMHMDA, where all experiments are conducted on the HMDD v3.2 dataset and the parameters of each method are kept in their original states.

[0065] NIMCGCN and MMGCN use graph convolutional networks as the main method for extracting node embeddings. NIMCGCN extracts embeddings from one feature dimension, while MMGCN applies multi-view GCN to extract multi-channel features. ERMDA adopts a resampling strategy and uses ensemble learning and feature selection to integrate various data sources to predict the association between miRNA and physiological state types. HGANMDA is a deep learning model that uses a hierarchical graph attention network to predict the association between miRNA and physiological state types, focusing on building a heterogeneous graph containing miRNA, physiological state types, and lncRNA associations, and applying node-level and semantic-level attention to prioritize connections to improve prediction accuracy. GAEMD uses a node-level attention autoencoder within the miRNA physiological state type framework to collect data and reconstruct a miRNA physiological state type connection network. MINIMDA develops various multimodal networks and synthesizes embedding representations of miRNA and physiological state types by integrating various high-order neighborhood data from these networks, aiming to infer the association between miRNA and physiological state types. MAGCN uses a graph convolutional network with a multi-channel attention mechanism and a convolutional neural network combiner to process the constructed lncRNA-miRNA and miRNA-physiological state type networks, and then uses a bilinear decoder to reconstruct the miRNA-physiological state type association. AMHMDA uses an attention-aware multi-view similarity network and hypergraph learning, combined with a graph convolutional network and an attention mechanism to predict the association between miRNA and physiological state types.

[0066] Table 1 is the comparison results of GLALP and other model experiments, which can be referred to as shown in Table 1: Table 1 Comparison results of GLALP and other model experiments

[0067] Among them, the AUC, AUPR, accuracy, F1-score, recall rate and precision of GLALP are 0.9451, 0.9459, 0.8727, 0.8674 and 0.8781 respectively. It can be seen that these six evaluation indicators are the highest among all methods. In particular, the F1-score of GLALP is 0.75% higher than that of AMHMDA, and the recall rate of GLALP is 1.25% higher than that of AMHMDA.

[0068] Furthermore, the fusion of global attention features is for each layer in the graph neural network, while the fusion of local attention features is focused on a single miRNA and drug node. Both fusion methods use different feature embeddings. In this embodiment, in order to evaluate the impact of combining local and global feature fusion methods, an ablation experiment was conducted. Table 2 shows the indicator results of global and local feature fusion, which can be referred to as shown in Table 2: Table 2 Index results of global and local feature fusion

[0069] In Table 2, the integration of local and global strategies produced the most favorable results in five indicators. It can be seen that the fusion of local and global attention showed quite good performance in the association prediction task of miRNA and physiological state types, although the combined strategy did not significantly surpass the other two methods. However, further research highlighted its superior performance on the selected dataset.

[0070] Further, several experiments were conducted on the selected datasets by AMHMDA, using only 20% of the HMDD v3.2 data for testing, while the remaining 80% was used to train the AMHDA model. Using a similar approach, this embodiment evaluated GLALP on various datasets. For model training, 80% of the HMDDv3.2 data was split into ratios of 1:1, 1:3, 1:5, 1:7, and 1:9, and the remaining 20% ​​was reserved for the final prediction task. Table 3 shows the experimental performance comparison results between GLALP and AMHMDA, which can be referred to as shown in Table 3: Table 3 Experimental performance comparison results between GLALP and AMHMDA

[0071] Table 3 shows the performance comparison between GLALP and AMHMDA. GLALP outperforms the existing AMHMDA in all indicators under different class imbalances. The difference is most obvious in the AUC and F1 score columns, where GLALP always shows higher values. When the data is divided 1:1, GLAMDA's AUC is 10% higher than AMHMDA, GLALP's F1 score is 11% higher than AMHMDA, and all other indicators GLAMDA is at least 1% higher. These results show that the classifier's overall ranking ability (AUC) and the balance between precision and recall (F1 score) are better. The trend shows that as the class imbalance increases (from 1:1 to 1:9), the performance of both methods on all indicators decreases, which is more common in class imbalance.

[0072] The present invention also provides an association prediction method, comprising: The multi-source features corresponding to the target micro RNA and the multi-source features corresponding to the target physiological state type are input into the trained association prediction model to obtain the association prediction result between the target micro RNA and the target physiological state type output by the trained association prediction model, wherein the trained association prediction model is trained based on the association prediction model training method described in the above embodiments.

[0073] In the present invention, for miRNA and physiological state, feature information needs to be collected from multiple sources, and these features may include but are not limited to gene expression levels, protein interactions, metabolite concentrations, environmental factors (such as diet, lifestyle), and clinical manifestations, etc. The use of multi-source features helps to more fully understand the relationship between miRNA and physiological state.

[0074] The association prediction model is a pre-trained machine learning or deep learning model used to predict the association between miRNA and a specific physiological state. The model establishes prediction rules by learning a large amount of data on known miRNA-physiological state associations. After the model training is completed, the multi-source features of new, unknown associated miRNAs and multi-source features of the target physiological state can be input into the model. These input features will be used as input variables of the model to generate prediction results.

[0075] Furthermore, based on the input multi-source features, the model outputs a prediction result, which indicates the degree or probability of association between the target miRNA and the target physiological state type. This result can be quantitative (such as probability value or score) or qualitative (such as "associated" or "not associated").

[0076] The association prediction method provided by the present invention obtains the final vector representation of the vertex through the steps of multi-view feature extraction, vertex global attention fusion feature representation acquisition, vertex local attention fusion feature representation acquisition and fusion feature integration, and based on the final vector representation of the vertex, predicts the association relationship between potential miRNA and physiological state type by training the fully connected network in the preset model, thereby improving the accuracy of association prediction between miRNA and physiological state type.

[0077] The following is a description of the association prediction model training system provided by the present invention. The association prediction model training system described below and the association prediction model training method described above can be referenced to each other.

[0078] Figure 6 The schematic diagram of the structure of the association prediction model training system provided by the present invention is as follows: Figure 6As shown, the present invention provides an association prediction model training system, including a sample acquisition module 601 and a model training module 602, wherein the sample acquisition module 601 is used to obtain multiple training samples, and the training samples are constructed by a first sample multi-source feature and a second sample multi-source feature having an association relationship label, wherein the first sample multi-source feature is a multi-source feature corresponding to the sample micro RNA, and the second sample multi-source feature is a multi-source feature corresponding to the sample physiological state type; the model training module 602 is used to input multiple of the training samples into a preset model, and based on the vertex vector representation output by the fusion feature layer of the preset model, train the preset model to obtain a trained association prediction model; wherein the fusion feature layer is used to fuse the global attention fusion feature representation output by the global attention fusion layer of the preset model and the local attention fusion feature representation output by the local attention fusion layer of the preset model to obtain the vertex vector representation.

[0079] The association prediction model training system provided by the present invention obtains the final vector representation of the vertex through the steps of multi-view feature extraction, vertex global attention fusion feature representation acquisition, vertex local attention fusion feature representation acquisition and fusion feature integration, and based on the final vector representation of the vertex, predicts the association relationship between potential miRNA and physiological state type by training the fully connected network in the preset model, thereby improving the accuracy of association prediction between miRNA and physiological state type.

[0080] The system provided by the present invention is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for the specific processes and detailed contents, which will not be repeated here.

[0081] Figure 7 A schematic diagram of the structure of an electronic device provided by the present invention, such as Figure 7As shown, the electronic device may include: a processor (Processor) 701, a communication interface (Communications Interface) 702, a memory (Memory) 703 and a communication bus 704, wherein the processor 701, the communication interface 702, and the memory 703 communicate with each other through the communication bus 704. The processor 701 may call the logic instructions in the memory 703 to execute the association prediction model training method, the method comprising: obtaining a plurality of training samples, wherein the training samples are constructed by a first sample multi-source feature and a second sample multi-source feature having an association relationship label, wherein the first sample multi-source feature is a multi-source feature corresponding to the sample micro RNA, and the second sample multi-source feature is a multi-source feature corresponding to the sample physiological state type; inputting a plurality of the training samples into a preset model, and training the preset model based on the vertex vector representation output by the fusion feature layer of the preset model to obtain a trained association prediction model; wherein the fusion feature layer is used to fuse the global attention fusion feature representation output by the global attention fusion layer of the preset model and the local attention fusion feature representation output by the local attention fusion layer of the preset model to obtain the vertex vector representation; Or, execute an association prediction method, which includes: inputting the multi-source features corresponding to the target micro RNA and the multi-source features corresponding to the target physiological state type into a trained association prediction model, and obtaining the association prediction result between the target micro RNA and the target physiological state type output by the trained association prediction model, wherein the trained association prediction model is trained based on the association prediction model training method described in the above embodiments.

[0082] In addition, the logic instructions in the above-mentioned memory 703 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk.

[0083] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the association prediction model training method provided by the above methods, and the method includes: obtaining multiple training samples, wherein the training samples are constructed by a first sample multi-source feature and a second sample multi-source feature having an association relationship label, wherein the first sample multi-source feature is a multi-source feature corresponding to the sample micro RNA, and the second sample multi-source feature is a multi-source feature corresponding to the sample physiological state type; inputting multiple of the training samples into a preset model, and training the preset model based on the vertex vector representation output by the fusion feature layer of the preset model to obtain a trained association prediction model; wherein the fusion feature layer is used to fuse the global attention fusion feature representation output by the global attention fusion layer of the preset model and the local attention fusion feature representation output by the local attention fusion layer of the preset model to obtain the vertex vector representation; Or, execute an association prediction method, which includes: inputting the multi-source features corresponding to the target micro RNA and the multi-source features corresponding to the target physiological state type into a trained association prediction model, and obtaining the association prediction result between the target micro RNA and the target physiological state type output by the trained association prediction model, wherein the trained association prediction model is trained based on the association prediction model training method described in the above embodiments.

[0084] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the processor executes the association prediction model training method provided in the above embodiments, the method comprising: obtaining a plurality of training samples, wherein the training samples are constructed by a first sample multi-source feature having an association relationship label and a second sample multi-source feature, wherein the first sample multi-source feature is a multi-source feature corresponding to the sample microRNA, and the second sample multi-source feature is a multi-source feature corresponding to the sample physiological state type; inputting the plurality of the training samples into a preset model, and training the preset model based on the vertex vector representation output by the fusion feature layer of the preset model to obtain a trained association prediction model; wherein the fusion feature layer is used to fuse the global attention fusion feature representation output by the global attention fusion layer of the preset model and the local attention fusion feature representation output by the local attention fusion layer of the preset model to obtain the vertex vector representation; Or, execute an association prediction method, which includes: inputting the multi-source features corresponding to the target micro RNA and the multi-source features corresponding to the target physiological state type into a trained association prediction model, and obtaining the association prediction result between the target micro RNA and the target physiological state type output by the trained association prediction model, wherein the trained association prediction model is trained based on the association prediction model training method described in the above embodiments.

[0085] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0086] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention 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. However, 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 invention.

Claims

1. A method for training an association prediction model, characterized in that: include: Acquire multiple training samples, where the training samples are constructed from a first sample multi-source feature and a second sample multi-source feature having an associated relationship label, wherein the first sample multi-source feature is a multi-source feature corresponding to the sample microRNA, and the second sample multi-source feature is a multi-source feature corresponding to the sample physiological state type; Inputting a plurality of the training samples into a preset model, and training the preset model based on the vertex vector representation output by the fusion feature layer of the preset model to obtain a trained association prediction model; Among them, the fusion feature layer is used to fuse the global attention fusion feature representation output by the global attention fusion layer of the preset model and the local attention fusion feature representation output by the local attention fusion layer of the preset model to obtain the vertex vector representation.

2. The association prediction model training method according to claim 1, characterized in that: The obtaining of multiple training samples comprises: Acquire a first similarity network and a second similarity network, wherein the first similarity network is a similarity network between the sample microRNAs, and the second similarity network is a similarity network between the sample physiological state types; Constructing corresponding node aggregation embedding according to network structure information of each node in the first similarity network and the second similarity network, wherein the network structure information is extracted based on a graph convolutional network; According to the node aggregation embedding corresponding to the first similarity network, a network node representation of the sample microRNA is obtained; according to the node aggregation embedding corresponding to the second similarity network, a network node representation of the sample physiological state type is obtained; When it is determined that there is a correlation between the sample microRNA and the sample physiological state type, the network node representation of the sample microRNA is used as the first sample multi-source feature, and the network node representation of the sample physiological state type corresponding to the sample microRNA is used as the second sample multi-source feature to construct the training sample.

3. The association prediction model training method according to claim 2, characterized in that: After inputting the plurality of training samples into the preset model, the method further comprises: Based on the preset activation function in the global attention fusion layer and the corresponding network node representation, a first attention coefficient and a second attention coefficient are calculated, wherein the first attention coefficient is the attention coefficient corresponding to the network node representation of the sample microRNA, and the second attention coefficient is the attention coefficient corresponding to the network node representation of the sample physiological state type; Based on the first preset convolutional neural network in the global attention fusion layer, a first global attention fusion feature representation is calculated according to the first attention coefficient and the network node representation of the sample microRNA, wherein the first global attention fusion feature representation is the global attention fusion feature representation corresponding to the sample microRNA; a second global attention fusion feature representation is calculated according to the second attention coefficient and the network node representation of the sample physiological state type, wherein the second global attention fusion feature representation is the global attention fusion feature representation corresponding to the sample physiological state type; Based on the global attention fusion layer, the first global attention fusion feature representation and the second global attention fusion feature representation are input into the fusion feature layer.

4. The association prediction model training method according to claim 3, characterized in that: After inputting the plurality of training samples into the preset model, the method further comprises: Based on the attention mechanism in the local attention fusion layer and the second preset convolutional neural network, a first local attention fusion feature representation is calculated according to the first node aggregate embedding, wherein the first node aggregate embedding is a node aggregate embedding corresponding to a node in the first similarity network; Based on the attention mechanism in the local attention fusion layer and the second preset convolutional neural network, a second local attention fusion feature representation is calculated according to the second node aggregate embedding, wherein the second node aggregate embedding is the node aggregate embedding corresponding to the node in the second similarity network; Based on the local attention fusion layer, the first local attention fusion feature representation and the second local attention fusion feature representation are input into the fusion feature layer.

5. The association prediction model training method according to claim 4, characterized in that: The method further comprises: Based on the attention mechanism of the fusion feature layer, the first global attention fusion feature representation and the first local attention fusion feature representation are fused and embedded to obtain a first graph aggregation embedding, wherein the first graph aggregation embedding is a graph aggregation embedding corresponding to the nodes in the first similarity network; Based on the attention mechanism of the fusion feature layer, the second global attention fusion feature representation and the second local attention fusion feature representation are fused and embedded to obtain a second graph aggregation embedding, wherein the second graph aggregation embedding is a graph aggregation embedding corresponding to the nodes in the second similarity network; Calculating the interaction embedding between the nodes in the first similarity network and the nodes in the second similarity network according to the interaction embedding of the first graph aggregate embedding and the second graph aggregate embedding; The first graph aggregate embedding, the interaction embedding and the second graph aggregate embedding are processed by a connection operation to obtain the vertex vector representation.

6. The association prediction model training method according to any one of claims 1 to 5, characterized in that: The vertex vector representation outputted by the fusion feature layer of the preset model is used to train the preset model to obtain a trained association prediction model, including: Based on the fully connected layer in the preset model, according to the vertex vector representation, predicting the corresponding association prediction result between the first sample multi-source feature and the second sample multi-source feature in the training sample; Based on the cross entropy loss function, performing loss value analysis on the association prediction result and the association relationship label corresponding to the training sample to obtain a loss value; When it is determined that the loss value meets the preset threshold, the trained association prediction model is obtained.

7. A method for association prediction, characterized in that: include: The multi-source features corresponding to the target micro RNA and the multi-source features corresponding to the target physiological state type are input into a trained association prediction model to obtain an association prediction result between the target micro RNA and the target physiological state type output by the trained association prediction model, wherein the trained association prediction model is trained based on the association prediction model training method as described in any one of claims 1 to 6.

8. A correlation prediction model training system, characterized in that: include: A sample acquisition module, used to acquire a plurality of training samples, wherein the training samples are constructed by a first sample multi-source feature and a second sample multi-source feature having an associated relationship label, wherein the first sample multi-source feature is a multi-source feature corresponding to the sample microRNA, and the second sample multi-source feature is a multi-source feature corresponding to the sample physiological state type; A model training module, used for inputting a plurality of the training samples into a preset model, and training the preset model based on the vertex vector representation output by the fusion feature layer of the preset model to obtain a trained association prediction model; Among them, the fusion feature layer is used to fuse the global attention fusion feature representation output by the global attention fusion layer of the preset model and the local attention fusion feature representation output by the local attention fusion layer of the preset model to obtain the vertex vector representation.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the association prediction model training method according to any one of claims 1 to 6 is implemented, or the association prediction method according to claim 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the association prediction model training method according to any one of claims 1 to 6 is implemented, or the association prediction method according to claim 7.

Images