Virus host prediction method and device, equipment and storage medium

Abstract

The invention discloses a virus host prediction method and device, equipment and a storage medium. The method comprises the following steps: acquiring a virus protein sequence; constructing a virus protein cluster network structure based on the virus protein sequence; performing community division on the virus protein cluster network structure to obtain a community distribution result; based on the community distribution result and the virus protein cluster network structure, performing iterative training on a to-be-trained graph convolutional neural network to obtain a virus host prediction model; and inputting the information of the to-be-predicted virus into the virus host prediction model, and predicting the host of the to-be-predicted virus. By constructing the virus protein cluster network structure, complex interaction between the virus and the host can be easily and comprehensively understood, and in combination with the community distribution result of the virus protein cluster network structure, the difficulty of extracting the feature information of the virus and the protein can be reduced; therefore, the accuracy of predicting the host of the virus by the virus host prediction model obtained by training can be improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a virus host prediction method, device, equipment and storage medium. Background Art

[0002] Viruses are tiny, submicroscopic particles without a complete cellular structure that can replicate using host cell systems. Composed primarily of nucleic acids (DNA or RNA) and a protein coat, viruses are highly parasitic and completely dependent on the host cell's energy and metabolic systems for vital functions. Predicting a virus's host cell can help understand the interactions between viruses and host cells, thereby exploiting these interactions to more effectively treat diseases caused by bacterial infections.

[0003] Existing virus-host prediction methods primarily use viral genomes or proteins for host prediction. For example, features are extracted from the genome and searched against a protein database. The extracted features are then converted into a numerical matrix and input into a multi-layer perceptron for host prediction. However, extracting features solely from viral genomes or protein sequences only partially captures the correlation between the virus and its host and cannot accurately predict a virus's host. Summary of the Invention

[0004] This application proposes a virus host prediction method, apparatus, computer equipment and storage medium to solve the problem that features extracted only from the viral genome or protein sequence only reflect partial correlation between the virus and the host and cannot accurately predict the virus host.

[0005] In a first aspect, a method for predicting a virus host is provided, comprising:

[0006] Obtain viral protein sequences;

[0007] Based on the viral protein sequence, constructing a viral protein cluster network structure;

[0008] Performing community division on the viral protein cluster network structure to obtain a community distribution result;

[0009] Based on the community distribution results and the viral protein cluster network structure, iteratively training the graph convolutional neural network to be trained to obtain a viral host prediction model;

[0010] The information of the virus to be predicted is input into the virus host prediction model to predict the host of the virus to be predicted.

[0011] In a second aspect, a virus host prediction device is provided, comprising:

[0012] An acquisition module, used to obtain viral protein sequences;

[0013] A construction module for constructing a viral protein cluster network structure based on the viral protein sequence;

[0014] A partitioning module is used to perform community partitioning on the viral protein cluster network structure to obtain community distribution results;

[0015] A training module is used to iteratively train the graph convolutional neural network to be trained based on the community distribution results and the viral protein cluster network structure to obtain a virus host prediction model;

[0016] The prediction module is used to input the information of the virus to be predicted into the virus host prediction model to predict the host of the virus to be predicted.

[0017] Optionally, the building blocks include:

[0018] A first evaluation submodule is used to perform similarity evaluation between each protein sequence in the viral protein sequence to obtain similarity results between each protein sequence;

[0019] A first construction submodule is used to construct a protein similarity graph based on the similarity results between the protein sequences in the viral protein sequence;

[0020] A first clustering submodule is used to cluster the protein similarity graph to obtain protein clusters;

[0021] The second construction submodule is used to construct the viral protein cluster network structure according to the correspondence between the virus and protein sequences in the viral protein sequence and the protein cluster.

[0022] Optionally, the division module includes:

[0023] A mapping submodule, configured to perform spatial mapping on the viral protein cluster network structure to obtain node latent variables;

[0024] The second clustering submodule is used to cluster the node hidden variables to implement community division of the node hidden variables and obtain community center vectors;

[0025] A second evaluation submodule is used to perform similarity evaluation on the node hidden variable and the community center vector to obtain a similarity distribution, wherein the similarity distribution represents the similarity between each virus node and each protein cluster node and each center vector in the community center vector;

[0026] The extraction submodule is used to extract the community distribution result from the similarity distribution.

[0027] Optionally, the mapping submodule includes:

[0028] A first calculation unit is used to calculate the correlation coefficient between each virus node and each neighbor node of the virus node and the correlation coefficient between each protein cluster node and each neighbor node of the protein cluster node in the virus protein cluster network structure;

[0029] a normalization unit, used for normalizing the correlation coefficients;

[0030] The second calculation unit is used to calculate each normalized correlation coefficient using an activation function to obtain the node hidden variable.

[0031] Optionally, the training module includes:

[0032] A prediction submodule, configured to use a graph convolutional neural network to be trained to perform host prediction on the viral protein cluster network structure to obtain a host prediction value for each viral node;

[0033] An updating submodule, configured to calculate the gradient of the loss function by backpropagation based on the host prediction value of each virus node, and to update the model parameters of the graph convolutional neural network to be trained using an optimization algorithm;

[0034] An enhancer module, for enhancing the viral protein cluster network structure based on the community distribution results and the host prediction value of each viral node;

[0035] A determination submodule is used to return to execute the host prediction of the viral protein cluster network structure using the graph convolutional neural network to be trained, obtain the host prediction value of each viral node, until the number of iterations meets the number threshold, and use the graph convolutional neural network to be trained as the viral host prediction model.

[0036] Optionally, the prediction submodule includes:

[0037] an acquisition unit, configured to acquire a node representation of each virus node in the viral protein cluster network structure based on a graph convolutional layer in the graph convolutional neural network to be trained;

[0038] The prediction unit is used to perform classification prediction on the node representation of each virus node to obtain a host prediction value of each virus node.

[0039] Optionally, the enhancer module includes:

[0040] a first determining unit, configured to determine a high-credibility node based on the host prediction value of each virus node, wherein the high-credibility node represents a first virus node that meets a first preset condition in each community representing the community distribution result;

[0041] an extraction unit, configured to extract a second virus node that meets a second preset condition from each community in the community distribution result;

[0042] a second determining unit, configured to determine, if a second virus node without a host label exists in the community of the community distribution result, the community corresponding to the second virus node without a host label as a target community;

[0043] An enhancement unit is configured to assign pseudo labels to all virus nodes without host labels in the target community according to the host label of the first virus node in the target community, so as to enhance the virus protein cluster network structure, wherein the pseudo label is consistent with the host label of the first virus node.

[0044] In a third aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned virus host prediction method when executing the computer program.

[0045] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned virus host prediction method are implemented.

[0046] The present application provides a virus host prediction method, device, computer equipment and storage medium, which obtains the virus protein sequence; constructs the virus protein cluster network structure based on the virus protein sequence; divides the virus protein cluster network structure into communities to obtain community distribution results; iteratively trains the graph convolutional neural network to be trained based on the community distribution results and the virus protein cluster network structure to obtain a virus host prediction model; inputs the information of the virus to be predicted into the virus host prediction model to predict the host of the virus to be predicted. The present application constructs the virus protein cluster network structure, constructs the biological properties contained between the phage and the protein cluster into a relationship in the network structure, links the virus and the protein cluster, avoids ignoring the important biological factors between the virus and the host, and comprehensively understands the complex interaction between the virus and the host. Combined with the community distribution results of the virus protein cluster network structure, the virus nodes in the same community are more closely connected in the virus protein cluster network structure, which is conducive to reducing the difficulty of extracting the characteristic information of the virus and the protein. In this way, the graph convolutional neural network to be trained is iteratively trained, which is conducive to improving the accuracy of the virus host prediction model trained to predict the host of the virus. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0048] Figure 1 A diagram illustrating the application environment of the virus host prediction method provided in an embodiment of the present application;

[0049] Figure 2 Flowchart of the virus host prediction method provided in the embodiment of the present application;

[0050] Figure 3 A schematic flow chart of a virus host prediction method provided in yet another embodiment of the present application;

[0051] Figure 4 A model framework diagram of the virus host prediction model provided in the embodiments of the present application;

[0052] Figure 5 A comparison chart of the prediction results of the virus host prediction model provided in the embodiment of the present application and the prior art;

[0053] Figure 6 A structural block diagram of a virus host prediction device provided in an embodiment of the present application;

[0054] Figure 7 This is a structural block diagram of the computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0056] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner.In the following description, many specific details are provided so as to provide a full understanding of the embodiments of the present application. However, it will be appreciated by those skilled in the art that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps etc. can be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring the various aspects of the application.

[0057] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically separate entities. That is, these functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0058] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, while others may be combined or partially combined. Therefore, the actual execution order may vary depending on the actual situation.

[0059] The virus host prediction method provided by the embodiment of the present invention can be applied in the following Figure 1 In the application environment. Among them, the computer device 110 communicates with the server 120 through the network 130. The computer device 110 can obtain the viral protein sequence; based on the viral protein sequence, construct a viral protein cluster network structure; perform community division on the viral protein cluster network structure to obtain a community distribution result; based on the community distribution result and the viral protein cluster network structure, iteratively train the graph convolutional neural network to be trained to obtain a viral host prediction model; input the information of the virus to be predicted into the viral host prediction model, predict the host of the virus to be predicted, and display it through the computer device 110. In the present invention, by constructing a viral protein cluster network structure, the biological properties contained between the phage and the protein cluster are constructed into a relationship in the network structure, and the virus and protein cluster are linked to avoid ignoring the important biological factors between the virus and the host, so as to fully understand the complex interaction between the virus and the host, and combine the community distribution results of the viral protein cluster network structure to make the viral nodes in the same community more closely connected in the viral protein cluster network structure, which is conducive to reducing the difficulty of extracting the characteristic information of the virus and the protein. In this way, iterative training of the graph convolutional neural network to be trained is conducive to improving the accurate characterization of the host predicted by the trained viral host prediction model. The computer device 110 may be, but is not limited to, various smart phones 110 - 1 , tablet computers 110 - 2 , and notebook computers 110 - 3 . The present invention will be described in detail below through specific embodiments.

[0060] See also Figure 2 As shown, Figure 2 This is a flow chart of a virus host prediction method provided by an embodiment of the present invention. This method can be applied to both a terminal and a server. This embodiment uses the server as an example. The virus host prediction method includes the following steps:

[0061] S101: Obtain viral protein sequence.

[0062] Wherein, the viral protein sequence refers to the amino acid sequence of the protein encoded by the viral genome. Exemplarily, the viral protein sequence can be a bacteriophage protein sequence.

[0063] For example, a bacteriophage protein sequence can be represented as: Among them, v i represents the i-th phage, represents all protein sequences of the i-th phage, the number of all protein sequences is m, represents the jth protein sequence of the i-th phage, a i Indicates one of the 20 amino acids, y i ∈{1,...,n h} represents h host tags of phage.

[0064] S102: Construct a viral protein cluster network structure based on the viral protein sequence.

[0065] Taking the viral protein sequence as a phage protein sequence as an example, the constructed viral protein cluster network structure is a phage protein cluster network structure, and the viral protein cluster network structure is expressed as G = {V, E, X}, where represents the total number of phages and protein clusters N, E = {(v i ,v j )|1≤i,j≤N,i≠j} represents the bacteriophage v i -Protein cluster edge v j The set of X=[x1;...;x N ]∈R N×F A∈R represents the F-dimensional feature vector of N phage and protein clusters, R is the real number field, and x1 represents the feature vector of the first phage or protein cluster in the real number field. N×N The adjacency matrix representing the phage protein cluster network structure has a size of N*N and is defined as follows:

[0066]

[0067] Among them, v i 、v j Represent phage and protein cluster respectively. If there is an edge between them <v i ,v j >, then the adjacency matrix A i,j is assigned a value of 1, otherwise 0.

[0068] In one embodiment, based on the viral protein sequence, a viral protein cluster network structure is constructed, including:

[0069] Performing similarity evaluation between each protein sequence in the viral protein sequence to obtain similarity results between each protein sequence;

[0070] Constructing a protein similarity graph based on the similarity results between each protein sequence in the viral protein sequence;

[0071] Clustering the protein similarity graph to obtain protein clusters;

[0072] According to the correspondence between virus and protein sequences and protein clusters in the viral protein sequence, a viral protein cluster network structure is constructed.

[0073] For example, a protein sequence alignment tool (double index alignment of next-generation sequencing data, DIAMOND) is used to perform pairwise protein sequence alignment between all protein sequence pairs in the viral protein sequence. If the protein pair in the DIAMOND alignment result is lower than a given threshold E value (the default threshold is set to 1e-5), it means that there is significant similarity between the paired protein sequences. Based on the alignment results, a protein similarity graph can be constructed, which is represented by G protein ={V protein ,E protein}, where V protein Represents the vertex set of the protein similarity graph, where the vertices are protein sequences, E protein The set of edges representing the protein similarity graph, where each edge represents the similarity between two vertices.

[0074] Use the Markov clustering algorithm MCL to identify the protein similarity graph G protein The set of disjoint protein clusters in is expressed as:

[0075]

[0076] Among them, C i represents a protein cluster, represents the i-th protein sequence vertex, represents the edge of the i-th protein sequence, The vertex representing the i-th protein sequence belongs to the vertex set of the protein similarity graph and The edge set of the i-th protein sequence belongs to the edge set of the protein similarity graph. In addition, for 1≤i,j≤n c And i≠j,n c represents the total number of protein clusters, Indicates that the vertex set of the i-th protein sequence and the vertex set of the j-th protein sequence are disjoint and Indicates that all protein sequence vertices constitute the vertex set of the protein similarity graph. If the virus v i Contains protein Contains protein sequences Protein cluster v j =C i , then for protein cluster v j and virus v i Construct an edge <v i ,v j >. Finally, a viral protein cluster network structure was constructed based on the correspondence between virus and protein sequences and the protein clusters obtained by clustering.

[0077] S103: Perform community division on the viral protein cluster network structure to obtain community distribution results.

[0078] The community distribution result refers to the collection of communities obtained after clustering the network structure of viral protein clusters, and each community contains similar viruses and protein clusters.

[0079] In one embodiment, the viral protein cluster network structure is divided into communities, and the community distribution results obtained include:

[0080] Perform spatial mapping on the viral protein cluster network structure to obtain node latent variables;

[0081] Cluster the node hidden variables to realize community division of the node hidden variables and obtain the community center vector;

[0082] The similarity between the node latent variables and the community center vector is evaluated to obtain the similarity distribution, which represents the similarity between each virus node and each protein cluster node and each center vector in the community center vector;

[0083] The community distribution results are extracted from the similarity distribution.

[0084] The viral protein cluster network structure G can be embedded into a low-dimensional space using a variational graph attention auto-encoder module (VGAAM) to achieve spatial mapping and generate node latent variables. These node latent variables simultaneously preserve the topological structure of the viral protein cluster network and the attributes of the viruses and protein clusters, including structural features, functions, and interactions between virus and protein sequences. The variational graph attention auto-encoder module can be a neural network.

[0085] VGAAM can include the inference model q θ (.) and the generative model pφ (.), where the inference model q θ (.) By learning the topological structure of the viral protein cluster network structure G and the attribute information of each node, the node hidden variables are obtained and the model p is generated. φ (.) Optimizing the inference model q by reconstructing the viral protein cluster network θ (.) The node hidden variables learned.

[0086] In the viral protein cluster network, since the phage nodes in the same community are more closely connected, their node representations are more similar. Therefore, a community detection module can be constructed to generate the model p φ (.) Collaboratively optimize the node latent vectors of each node in the viral protein cluster network, and ultimately obtain stable node latent variables and community division results. The community detection module can be a module composed of a neural network.

[0087] Specifically, for the node hidden variable Z obtained by the variational graph attention autoencoder module, the point hidden variable Z is clustered based on the K-means algorithm in the community detection network to obtain a community center vector containing K center vectors, which is expressed as μ = {μ1,μ2,...,μ k}.

[0088] The similarity between the node hidden variable Z and the community center vector μ is evaluated by the Student t distribution algorithm of the community detection network, which is expressed as:

[0089]

[0090] Among them, z i represents the node hidden variable of node i, μ j represents the jth community center vector, α≥1 represents the degree of freedom of the Student t distribution, the default value is 1, q ij represents the probability that node i belongs to the jth community.

[0091] By calculating the similarity between each node (including virus nodes and protein cluster nodes) and each community center vector, a similarity distribution Q∈R can be obtained. N×K , where N represents the number of nodes and K represents the number of center vectors in the community center vector.

[0092] After obtaining stable node latent variables and community distributions, we can obtain the cluster with the highest probability from the similarity distribution as the community of the corresponding node and obtain the community distribution result. The calculation process is as follows:

[0093] c j =argmax(q ij )

[0094] Among them, cj represents the jth community, q ij represents the probability of a node being assigned to a community.

[0095] In order to obtain the topological structure of the virus protein cluster network structure, the attribute characteristics of viruses and proteins, we can use the inference model q θ (Z|X,A), obtain the node variational approximate probability distribution of the node feature representation of each virus node and the node feature representation of the protein cluster node. The process can be expressed as:

[0096]

[0097] Where Z∈R N×F represents the node hidden variable, Z belongs to the real field matrix of N*F dimension, N represents the number of nodes (including virus nodes and protein cluster nodes) in the virus protein cluster network structure, F' represents the hidden layer dimension, q θ (z i |X,A) is the approximate true posterior distribution of node i based on Gaussian distribution, X represents the initialization features of virus nodes and protein cluster nodes, and A represents the adjacency matrix.

[0098] By generating a model p φ (.)Use p φ (A|Z) reconstructs the adjacency matrix A. Specifically, the cross entropy based on dot multiplication is used to reconstruct the adjacency matrix A of the viral protein cluster network structure. The specific calculation formula is as follows:

[0099]

[0100]

[0101] Among them, σ represents the activation function, z i represents the hidden vector of the i-th node, z j represents the jth node hidden vector, A ij represents the edge between the hidden vector of the i-th node and the hidden vector of the j-th node, N represents the number of nodes (including virus nodes and protein cluster nodes) in the virus protein cluster network structure, represents the transpose of the j-th node hidden vector.

[0102] After calculating the similarity distribution Q, the auxiliary distribution P can be calculated based on the similarity distribution Q, making the similarity distribution Q and the auxiliary distribution P as similar as possible, so that the similarity distribution Q learns and optimizes from the high-confidence community assignment. The degree of freedom α of the Student t distribution can be adjusted to adjust the concentration between the auxiliary distribution P and the similarity distribution Q. The higher the degree of freedom, the more concentrated the auxiliary distribution P is compared to the similarity distribution Q.

[0103] The auxiliary distribution P is identified by the similarity distribution Q, which can enhance the clustering effect of the node dependent variable and the allocation of high-confidence nodes. After obtaining the community center vector, the community center vector can be normalized to prevent large communities from distorting the latent vector space. The clustering loss can force the similarity distribution Q to approach the auxiliary distribution P, so the similarity distribution Q and the auxiliary distribution P can supervise and optimize each other through continuous iterative updates. The auxiliary distribution P is used to assign a greater probability to similar nodes in the latent vector space and a lower probability to different nodes in the latent vector space. The auxiliary distribution P can be expressed as:

[0104]

[0105] Among them, q ij represents the probability that node i is assigned to the jth community, represents the frequency of occurrence of the jth community.

[0106] By using the loss function L VGAAM and L CDM The node hidden variables in the viral protein cluster network structure and their final community distribution are learned respectively.

[0107] Among them, the loss function of the variational graph attention autoencoder module is as follows:

[0108]

[0109] in, Represents the generative model p φ (A|Z) reconstructs the deviation of the adjacency matrix A, a ij |z i , z j Represents the hidden variable z at given node i and node j i and z j Under the condition of , the probability of the existence of edges between nodes, N represents the number of nodes in the viral protein cluster network structure, Represents the prior distribution p(Z) of the node hidden variable Z in the viral protein cluster network structure and the inference model q θ The error between (Z|X,A).

[0110] The loss function of the community detection module is as follows:

[0111]

[0112] Where P and Q represent the auxiliary distribution and similarity distribution, respectively.

[0113] L VGAAM and L CDM The combination is used for joint optimization learning, and the unsupervised learning objective function is obtained as follows:

[0114] Loss UL =L VGAAM +L CDM

[0115] When obtaining node latent variables and community distribution results, the Adam optimization algorithm is used to minimize the unsupervised learning objective function. The resulting loss is then used to update the weights of the variational graph attention autoencoder module and the community detection module through backpropagation. After multiple iterations of training, the optimization stops after reaching the specified number of training iterations, resulting in stable node latent variables and community distribution results.

[0116] Since the viral protein cluster network is a heterogeneous network, learning node representation features in heterogeneous networks is relatively complex. In order to capture the network topology and node attributes, the node representation is updated based on the correlation between the node and its neighbors. The topology and attribute information of each node in the viral protein cluster network structure can be captured based on the attention mechanism. In one embodiment, the viral protein cluster network structure is spatially mapped, and the node latent variables obtained include:

[0117] Calculate the correlation coefficient between each virus node and its neighbor nodes in the virus protein cluster network structure, as well as the correlation coefficient between each protein cluster node and its neighbor nodes;

[0118] Normalize each correlation coefficient;

[0119] The activation function is used to calculate each normalized correlation coefficient to obtain the node hidden variable.

[0120] The attention mechanism can be used to determine the node v according to the following formula i and its neighbor node v j The correlation coefficient r ij :

[0121] r ij =LeakyReLU(w T ReLU(Wx i )+w T ReLU(Wx j ))

[0122] Where W∈R F′×F and w∈R F′ Represents the training weight parameters in the generative model, x i and x j Represents node v i (virus node or protein cluster node) and node vj (neighbor node of virus node or neighbor node of protein cluster node), LeakyReLU and ReLU represent activation functions respectively.

[0123] The correlation coefficient r is obtained using the SoftMax function ij Normalized, the calculation formula is as follows:

[0124]

[0125] Among them, α ij represents the normalized correlation coefficient, N i Represents node v i The neighbor node set of node v i The calculation formula for the latent variable embedding node representation is as follows:

[0126]

[0127] Where W is the weight matrix and σ represents the activation function.

[0128] By performing the same operation on each node (including virus nodes and protein cluster nodes), the latent variable representation Z of all nodes V can be obtained.

[0129] S104: Based on the community distribution results and the virus protein cluster network structure, the graph convolutional neural network to be trained is iteratively trained to obtain a virus host prediction model.

[0130] The graph convolutional neural network (GCN) to be trained refers to a neural network model that has not been trained or needs further training. In this application, it is used to process the viral protein cluster network structure. The viral protein cluster network structure is input into the graph neural network to be trained to predict the host prediction value of the virus, and the graph convolutional neural network to be trained is iteratively trained based on the community distribution results and the host prediction value of the virus predicted by the graph neural network to be trained and the loss function until the number of iterations meets the preset number, and the graph convolutional neural network to be trained is used as the virus host prediction model.

[0131] In one embodiment, based on the community distribution results and the viral protein cluster network structure, the graph convolutional neural network to be trained is iteratively trained to obtain a viral host prediction model including:

[0132] Using the graph convolutional neural network to be trained, the host prediction of the viral protein cluster network structure is performed to obtain the host prediction value of each viral node;

[0133] The gradient of the loss function is calculated by backpropagation based on the host prediction value of each virus node, and the model parameters of the graph convolutional neural network to be trained are updated using an optimization algorithm;

[0134] Enhance the viral protein cluster network structure based on community distribution results and host prediction value of each viral node;

[0135] Return to execute host prediction for the viral protein cluster network structure using the graph convolutional neural network to be trained, obtain the host prediction value for each viral node, until the number of iterations meets the number threshold, and use the graph convolutional neural network to be trained as the viral host prediction model.

[0136] The loss function may be a cross loss function.

[0137] Specifically, the virus protein cluster network structure is input into the graph convolutional neural network to be trained for host prediction, and the host prediction value of each virus node is obtained; the cross entropy loss function is used to calculate the error between the host prediction value and the host's true label. gcn It can be expressed as:

[0138]

[0139] Among them, n h represents the number of host categories, Z true Indicates the true host category label of the virus, Z pre Represents the host prediction value of the virus.

[0140] The Adam optimization algorithm is used to minimize the cross-entropy loss function. Specifically, the cross-entropy loss function is used to calculate the error between the host prediction value and the true label as the loss value. The loss value is then back-propagated to obtain the error gradient (i.e., the partial derivative of the loss with respect to the model parameters). Finally, the Adam optimization algorithm is used to update the model parameters of the graph convolutional neural network to be trained, such as weights, based on the error gradient.

[0141] The viral protein cluster network structure is enhanced based on the community distribution results and the host prediction value of each viral node.

[0142] Then, the enhanced viral protein cluster network is input into the graph convolutional neural network to be trained for host prediction. The step of obtaining the host prediction value of each viral node is continued until the number of iterations meets the number threshold, the optimization stops, and the graph convolutional neural network to be trained is used as the virus host prediction model.

[0143] In one embodiment, a graph convolutional neural network to be trained is used to perform host prediction on the viral protein cluster network structure to obtain a host prediction value for each viral node, including:

[0144] Obtaining a node representation of each virus node in the viral protein cluster network structure based on the graph convolutional layer in the graph convolutional neural network to be trained;

[0145] The node representation of each virus node is classified and predicted to obtain the host prediction value of each virus node.

[0146] The graph convolution layer uses a single-layer graph convolution model and uses the single-layer graph convolution network as a semi-supervised classifier. The virus protein cluster network structure G = {V, E, X} is input into the graph convolution to obtain the node representation of the virus node in the virus protein cluster network structure. Then, an activation function SoftMax is used to output the host prediction value Z of each virus node. pre , the specific calculation is as follows:

[0147]

[0148] Among them, V in G={v i ,v j} i＝1,2,...,S,j＝S+1,S+2,...,N Indicates that the network contains S viruses and NS protein clusters, and E in G = {(v i ,v j )|1≤i,j≤N,i≠j} represents the edge set between viruses and protein clusters in the virus protein cluster network structure, (v i ,v j ) indicates virus v i and protein cluster v j There are edges between them. It is obtained by adding the adjacency matrix A and the identity matrix I. is The calculated diagonal matrix, W gcn It is the trainable parameter matrix of the single-layer graph convolution model, where X in G = [x1; ...; x N ]∈R N×F represents the initialization features of virus nodes and protein cluster nodes, and ReLU represents the activation function.

[0149] In one embodiment, enhancing the viral protein cluster network structure based on the community distribution results and the host prediction value of each viral node includes:

[0150] Determine a high-credibility node based on the host prediction value of each virus node, where the high-credibility node represents the first virus node in each community of the community distribution result that meets the first preset condition;

[0151] Extracting a second viral node that meets a second preset condition from each community of the community distribution result;

[0152] If there is a second virus node without a host label in the community of the community distribution result, the community corresponding to the second virus node without a host label is determined as the target community;

[0153] According to the host label of the first virus node in the target community, pseudo labels are assigned to all virus nodes without host labels in the target community to enhance the virus protein cluster network structure, wherein the pseudo label is consistent with the host label of the first virus node.

[0154] Among them, the probability of the first virus node that meets the first preset condition in each community of the community distribution result being the first l virus nodes with the host prediction value is a high-credible node. The high-credible node is marked with a host label, and the community where the high-credible node is located in the community distribution result can be determined. Therefore, the high-credible node It can be expressed as:

[0155]

[0156] Among them, Z pre represents the host prediction value of the virus node v, and the i-th community represents the community where the high-trusted node l is located.

[0157] Highly trusted nodes can be determined based on community distribution results Community

[0158] Extract the virus node v that meets the second preset condition, i.e., the first q host prediction value, from each community in the community distribution result, i.e., the second virus node, which can be expressed as:

[0159]

[0160] in, represents the host prediction value of the virus node v, i represents the i-th community c where the virus node v is located i .

[0161] If there is a second virus node with or without a host label in the community in the community distribution result, the community is determined as the target community, and pseudo labels are assigned to all virus nodes without host labels in the target community according to the host label of the first virus node in the target community. That is, pseudo labels are added to the unlabeled virus nodes belonging to the target community in the virus protein cluster network structure to enhance the virus protein cluster network structure.

[0162] For example, for a phage v1 with a known host label of 1, the phage v1 is a high-trust node and belongs to the first community c1. Then the pseudo labels of all phages with unknown hosts in the first community c1 are set to 1.

[0163] In this way, during the iterative training of the graph convolutional neural network to be trained, the number of reliable samples in the viral protein cluster network structure will continue to increase after each training, thereby improving the prediction accuracy of the host prediction model obtained by the final training.

[0164] S105: Input the information of the virus to be predicted into the virus host prediction model to predict the host of the virus to be predicted.

[0165] The information of the virus to be predicted may be the sequence information of the virus to be predicted, and the sequence information may be the phage protein cluster network structure of the phage protein sequence of the virus to be predicted constructed using the virus host prediction method of the present application. The sequence information of the virus to be predicted is input into the virus host prediction model to predict the host of the virus to be predicted.

[0166] Take the phage protein sequence as an example, Figures 3 to 4 As shown, a flowchart of a virus host prediction method and a model framework diagram of a virus host prediction model are provided, as follows:

[0167] Obtain a phage protein sequence; construct a phage protein cluster network structure based on the phage protein sequence; construct a variational graph autoencoder module and a community detection module, construct an unsupervised learning objective function, optimize the unsupervised learning objective function until the optimization number is reached, obtain the node latent variables and community divisions (i.e., community distribution results) output by the variational graph autoencoder module and the community detection module, construct a semi-supervised classifier module, and the semi-supervised classifier module is a graph convolutional neural network. The graph convolutional neural network is collaboratively optimized through the community distribution results for iterative training; during the iterative training process, the loss function of the graph convolutional neural network (i.e., the semi-supervised learning objective function), such as the cross entropy function, is optimized until the number of iterations is reached, and the trained graph convolutional neural network is determined to be a host prediction model. The host prediction model can be used to predict the host of the phage based on the sequence information of the phage to predict the host of the phage. Iterative training of a graph convolutional neural network using collaborative optimization of community distribution results can be performed by identifying high-confidence nodes from the host prediction values ​​predicted by the graph convolutional neural network and determining the community in which the high-confidence node resides in the community distribution results. If a virus node without a host label is present among the top q virus nodes in the community distribution results, the community is determined to be the target community. In this way, pseudo-labels can be assigned to all virus nodes without a host label in the target community based on the host labels of the high-confidence nodes in the target community. Nodes assigned with pseudo-labels are then used as training nodes in the phage protein cluster network structure, greatly enriching the feature representations of phages and protein clusters, making adjacent phages in the same community more closely connected, node representations more similar, and easier to extract feature information between phages and protein clusters, thereby improving the prediction accuracy of the trained virus-host prediction model. Finally, the virus-host prediction model can be used to predict which host bacteria a phage from an unknown host can infect.

[0168] This application predicts two different phage data sets, namely, Dataset1 data set, including 1003 phages and Dataset2 data set, including 1059 phages. The training parameter parameters of the virus host prediction model of this application are set as the hidden layer dimension F' of the node hidden variable Z is set to 1000, the training rounds of unsupervised learning and semi-supervised learning are set to 200, the learning rate is set to 0.05, the dimension F of the phage node and protein cluster node feature X is set to 256, and the initial feature of the phage node is set to 4-mer frequency.

[0169] The performance of the virus host prediction model trained based on the above training parameters in the phage host prediction task can be evaluated by the accuracy rate (Accuracy), and the calculation formula is as follows:

[0170]

[0171] Among them, TP is the number of correctly predicted positive examples, TN is the number of correctly predicted negative examples, FP is the number of negative examples incorrectly predicted as positive examples, and FN is the number of positive examples incorrectly predicted as negative examples.

[0172] refer to Figure 3The comparison of the prediction results shown shows that the prediction accuracy of the virus host prediction model of the present application is 0.8724 and 0.8349 on Dataset 1 and Dataset 2, respectively, using VGAACD-PCG, that is, the virus host prediction model trained by the present application. This is far higher than the prediction accuracy of other existing methods. The prior art is explained as follows: HostPhinder uses k-mer (short sequences of length k) to evaluate the genetic similarity between phages of unknown hosts and reference datasets, and the phage host with the highest similarity is used as the prediction result. vHULK first extracts features from the phage genome and searches against the protein database, then converts these features into a numerical matrix and finally inputs them into a multi-layer perceptron for host prediction. DeepHost uses k-mer to encode phage genomes of different lengths into a 3D matrix and trains a convolutional neural network model for host prediction. Graphage converts each phage sequence into a gap pattern graph, then uses a multi-layer graph convolutional neural network to convert each gap pattern graph into a low-dimensional feature embedding representation, and finally inputs the phage features into the neural network for host classification. PhageTB is a phage ensemble prediction model that integrates four prediction methods to predict phage-host interactions. These prediction methods include constructing comparative models through the similarities between phage-phage, host-host, and phage-Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR). An integrated model is also developed by combining a hybrid model with a comparative model for host prediction tasks.

[0173] The above is the virus host prediction process of this application.

[0174] As mentioned above, the present application provides a virus host prediction method, device, computer equipment and storage medium, by obtaining a virus protein sequence; constructing a virus protein cluster network structure based on the virus protein sequence; performing community division on the virus protein cluster network structure to obtain a community distribution result; based on the community distribution result and the virus protein cluster network structure, iteratively training the graph convolutional neural network to be trained to obtain a virus host prediction model; inputting the information of the virus to be predicted into the virus host prediction model to predict the host of the virus to be predicted. The present application constructs a virus protein cluster network structure, constructs the biological properties contained between the phage and the protein cluster into a relationship in the network structure, links the virus and the protein cluster, avoids ignoring the important biological factors between the virus and the host, and comprehensively understands the complex interaction between the virus and the host, combines the community distribution results of the virus protein cluster network structure, makes the virus nodes in the same community more closely connected in the virus protein cluster network structure, and helps to reduce the difficulty of extracting the characteristic information of the virus and protein. In this way, iterative training of the graph convolutional neural network to be trained is conducive to improving the accuracy of the virus host prediction model trained to predict the host of the virus.

[0175] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0176] In one embodiment, a virus host prediction device is provided, which corresponds to the virus host prediction method in the above embodiment. Figure 6 As shown, the virus host prediction device includes:

[0177] Acquisition module 201, used to obtain viral protein sequences;

[0178] Construction module 202, for constructing a viral protein cluster network structure based on the viral protein sequence;

[0179] A partitioning module 203 is used to perform community partitioning on the viral protein cluster network structure to obtain community distribution results;

[0180] A training module 204 is used to iteratively train the graph convolutional neural network to be trained based on the community distribution results and the viral protein cluster network structure to obtain a viral host prediction model;

[0181] The prediction module 205 is used to input the information of the virus to be predicted into the virus host prediction model to predict the host of the virus to be predicted.

[0182] In this embodiment, by constructing a virus protein cluster network structure, the biological properties contained between the phage and the protein cluster are constructed into a relationship in the network structure, and the virus and protein cluster are linked to avoid ignoring important biological factors between the virus and the host, so as to fully understand the complex interactions between the virus and the host. Combined with the community distribution results of the virus protein cluster network structure, the virus nodes in the same community are more closely connected in the virus protein cluster network structure, which is conducive to reducing the difficulty of extracting characteristic information of the virus and protein. In this way, iterative training of the graph convolutional neural network to be trained is conducive to improving the accuracy of the virus host prediction model obtained by training to predict the host of the virus.

[0183] Optionally, building blocks include:

[0184] The first evaluation submodule is used to evaluate the similarity between each protein sequence in the viral protein sequence to obtain the similarity results between each protein sequence;

[0185] The first construction submodule is used to construct a protein similarity graph based on the similarity results between the protein sequences in the viral protein sequence;

[0186] The first clustering submodule is used to cluster the protein similarity graph to obtain protein clusters;

[0187] The second construction submodule is used to construct a viral protein cluster network structure based on the correspondence between the virus and protein sequences and the protein clusters in the viral protein sequence.

[0188] Optionally, the partitioning module includes:

[0189] The mapping submodule is used to perform spatial mapping on the viral protein cluster network structure to obtain node latent variables;

[0190] The second clustering submodule is used to cluster the node hidden variables to realize community division of the node hidden variables and obtain the community center vector;

[0191] The second evaluation submodule is used to perform similarity evaluation on the node hidden variables and the community center vector to obtain a similarity distribution, which represents the similarity between each virus node and each protein cluster node and each center vector in the community center vector;

[0192] The extraction submodule is used to extract the community distribution results from the similarity distribution.

[0193] Optionally, the mapping submodule includes:

[0194] A first calculation unit is used to calculate the correlation coefficient between each virus node and its neighboring nodes in the virus protein cluster network structure, and the correlation coefficient between each protein cluster node and its neighboring nodes;

[0195] Normalization unit, used to normalize each correlation coefficient;

[0196] The second calculation unit is used to calculate each normalized correlation coefficient using an activation function to obtain a node latent variable.

[0197] Optionally, the training module includes:

[0198] The prediction submodule is used to use the graph convolutional neural network to be trained to perform host prediction on the viral protein cluster network structure and obtain the host prediction value of each viral node;

[0199] The update submodule is used to calculate the gradient of the loss function based on the host prediction value of each virus node through backpropagation, and to update the model parameters of the graph convolutional neural network to be trained using an optimization algorithm;

[0200] Enhancer module, used to enhance the viral protein cluster network structure based on the community distribution results and the host prediction value of each viral node;

[0201] The submodule is determined to return to execute host prediction of the viral protein cluster network structure using the graph convolutional neural network to be trained, obtain the host prediction value of each viral node, until the number of iterations meets the number threshold, and use the graph convolutional neural network to be trained as the viral host prediction model.

[0202] Optionally, the prediction submodule includes:

[0203] an acquisition unit, configured to acquire a node representation of each virus node in the virus protein cluster network structure based on a graph convolutional layer in the graph convolutional neural network to be trained;

[0204] The prediction unit is used to perform classification prediction on the node representation of each virus node to obtain the host prediction value of each virus node.

[0205] Optionally, the enhancer module includes:

[0206] a first determining unit, configured to determine a high-credibility node based on the host prediction value of each virus node, wherein the high-credibility node represents a first virus node that meets a first preset condition in each community representing the community distribution result;

[0207] an extraction unit, configured to extract a second virus node that meets a second preset condition from each community in the community distribution result;

[0208] A second determining unit is configured to determine, if there is a second virus node without a host label in the community of the community distribution result, the community corresponding to the second virus node without a host label as the target community;

[0209] The enhancement unit is used to assign pseudo labels to all virus nodes without host labels in the target community according to the host label of the first virus node in the target community, so as to enhance the virus protein cluster network structure, wherein the pseudo label is consistent with the host label of the first virus node.

[0210] In one embodiment, a computer device is provided. The internal structure diagram of the computer device can be as follows: Figure 7 As shown. The computer device includes a processor, memory, network interface, display screen, and input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When executed by the processor, the computer program implements the functions or steps of a virus host prediction method.

[0211] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented:

[0212] Obtain the viral protein sequence; construct a viral protein cluster network structure based on the viral protein sequence; divide the viral protein cluster network structure into communities to obtain community distribution results; iteratively train the graph convolutional neural network to be trained based on the community distribution results and the viral protein cluster network structure to obtain a viral host prediction model; input the information of the virus to be predicted into the viral host prediction model to predict the host of the virus to be predicted.

[0213] In this embodiment, by constructing a virus protein cluster network structure, the biological properties contained between the phage and the protein cluster are constructed into a relationship in the network structure, and the virus and protein cluster are linked to avoid ignoring important biological factors between the virus and the host, so as to fully understand the complex interactions between the virus and the host. Combined with the community distribution results of the virus protein cluster network structure, the virus nodes in the same community are more closely connected in the virus protein cluster network structure, which is conducive to reducing the difficulty of extracting characteristic information of the virus and protein. In this way, iterative training of the graph convolutional neural network to be trained is conducive to improving the accuracy of the virus host prediction model obtained by training to predict the host of the virus.

[0214] In one embodiment, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the following steps are implemented:

[0215] Obtain the viral protein sequence; construct a viral protein cluster network structure based on the viral protein sequence; divide the viral protein cluster network structure into communities to obtain community distribution results; iteratively train the graph convolutional neural network to be trained based on the community distribution results and the viral protein cluster network structure to obtain a viral host prediction model; input the information of the virus to be predicted into the viral host prediction model to predict the host of the virus to be predicted.

[0216] In this embodiment, by constructing a virus protein cluster network structure, the biological properties contained between the phage and the protein cluster are constructed into a relationship in the network structure, and the virus and protein cluster are linked to avoid ignoring important biological factors between the virus and the host, so as to fully understand the complex interactions between the virus and the host. Combined with the community distribution results of the virus protein cluster network structure, the virus nodes in the same community are more closely connected in the virus protein cluster network structure, which is conducive to reducing the difficulty of extracting characteristic information of the virus and protein. In this way, iterative training of the graph convolutional neural network to be trained is conducive to improving the accuracy of the virus host prediction model obtained by training to predict the host of the virus.

[0217] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can be found in the relevant descriptions of the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0218] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0219] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0220] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. 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. 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 various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A virus host prediction method, characterized in that: include: Obtain viral protein sequences; Based on the viral protein sequence, constructing a viral protein cluster network structure; Performing community division on the viral protein cluster network structure to obtain a community distribution result; Based on the community distribution results and the viral protein cluster network structure, iteratively training the graph convolutional neural network to be trained to obtain a viral host prediction model; The information of the virus to be predicted is input into the virus host prediction model to predict the host of the virus to be predicted.

2. The virus host prediction method according to claim 1, characterized in that The method of constructing a viral protein cluster network structure based on the viral protein sequence comprises: Performing similarity evaluation on each protein sequence in the viral protein sequence to obtain similarity results between each protein sequence; constructing a protein similarity graph based on the similarity results between the protein sequences in the viral protein sequence; Clustering the protein similarity graph to obtain protein clusters; The viral protein cluster network structure is constructed based on the correspondence between the virus and protein sequences in the viral protein sequence and the protein cluster.

3. The virus host prediction method according to claim 1, characterized in that The community division of the viral protein cluster network structure to obtain the community distribution result includes: Performing spatial mapping on the viral protein cluster network structure to obtain node latent variables; Clustering the node hidden variables to achieve community division of the node hidden variables and obtain community center vectors; Performing similarity evaluation on the node hidden variable and the community center vector to obtain a similarity distribution, wherein the similarity distribution represents the similarity between each virus node and each protein cluster node and each center vector in the community center vector; The community distribution result is extracted from the similarity distribution.

4. The virus host prediction method according to claim 3, characterized in that The spatial mapping of the viral protein cluster network structure to obtain node hidden variables includes: Calculating the correlation coefficient between each virus node and each neighboring node of the virus node and the correlation coefficient between each protein cluster node and each neighboring node of the protein cluster node in the virus protein cluster network structure; Normalizing each of the correlation coefficients; An activation function is used to calculate each normalized correlation coefficient to obtain the node latent variable.

5. The virus host prediction method according to claim 1, characterized in that The iterative training of the graph convolutional neural network to be trained based on the community distribution results and the viral protein cluster network structure to obtain a viral host prediction model includes: Using the graph convolutional neural network to be trained, host prediction is performed on the viral protein cluster network structure to obtain a host prediction value for each viral node; Back-propagating the gradient of the loss function based on the host prediction value of each virus node, and updating the model parameters of the graph convolutional neural network to be trained using an optimization algorithm; enhancing the viral protein cluster network structure based on the community distribution results and the host prediction value of each viral node; Return to executing the host prediction of the viral protein cluster network structure using the graph convolutional neural network to be trained, obtain the host prediction value of each viral node, until the number of iterations meets the number threshold, and use the graph convolutional neural network to be trained as the viral host prediction model.

6. The virus host prediction method according to claim 5, characterized in that The graph convolutional neural network to be trained is used to perform host prediction on the viral protein cluster network structure to obtain a host prediction value for each viral node, including: Obtaining a node representation of each virus node in the viral protein cluster network structure based on a graph convolutional layer in the graph convolutional neural network to be trained; Classification prediction is performed on the node representation of each virus node to obtain a host prediction value of each virus node.

7. The virus host prediction method according to claim 5, characterized in that The step of enhancing the viral protein cluster network structure based on the community distribution result and the host prediction value of each viral node includes: Determining a high-credibility node based on the host prediction value of each virus node, wherein the high-credibility node represents the first virus node in each community of the community distribution result that meets a first preset condition; Extracting a second viral node that meets a second preset condition from each community of the community distribution result; If there is a second virus node without a host label in the community of the community distribution result, the community corresponding to the second virus node without a host label is determined as the target community; According to the host label of the first virus node in the target community, pseudo labels are assigned to all virus nodes without host labels in the target community to enhance the virus protein cluster network structure, wherein the pseudo labels are consistent with the host label of the first virus node.

8. A virus host prediction device, characterized in that: include: An acquisition module, used to obtain viral protein sequences; A construction module for constructing a viral protein cluster network structure based on the viral protein sequence; A partitioning module is used to perform community partitioning on the viral protein cluster network structure to obtain community distribution results; A training module is used to iteratively train the graph convolutional neural network to be trained based on the community distribution results and the viral protein cluster network structure to obtain a virus host prediction model; The prediction module is used to input the information of the virus to be predicted into the virus host prediction model to predict the host of the virus to be predicted.

9. A computer 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 steps of the virus host prediction method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the virus host prediction method according to any one of claims 1 to 7 are implemented.

Images