Virus and disease association prediction model, construction method of prediction model and prediction method
By employing a dual-model complementary design and ensemble learning strategy, the sequence similarity of viruses, semantic similarity of diseases, and functional similarity features are integrated, thus solving the problem of insufficient accuracy in predicting the association between viruses and diseases. This achieves efficient and systematic prediction of the association between viruses and diseases, thereby improving prediction performance.
Patent Information
- Application Number
- CN202511754275.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-26
- Publication Date
- 2026-03-06
AI Technical Summary
Existing technologies struggle to systematically integrate multi-source biological characteristics, resulting in insufficient accuracy in predicting the association between viruses and diseases. Furthermore, traditional methods are costly, time-consuming, and highly dependent on samples.
A dual-model complementary design is adopted, in which sequence similarity features of the virus and semantic similarity features of the disease are extracted by independent prediction model A and independent prediction model B, respectively, and combined with functional similarity features. A bagging ensemble strategy is used to generate a prediction model for the association between the virus and the disease, thus integrating multimodal biological information.
It achieves more comprehensive prediction of virus-disease associations, improves the accuracy and robustness of predictions, significantly enhances the performance of AUROC and AUPRC, and has good scalability and versatility.
Smart Images
Figure CN121617655A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of virus and disease interaction prediction technology, and in particular to a virus and disease association prediction model, a method for constructing the prediction model, and a prediction method. Background Technology
[0002] Viruses play a crucial role in the development and progression of many human diseases. However, a systematic understanding of the link between viruses and diseases remains lacking. Traditional research methods primarily rely on epidemiological surveys, retrospective clinical analyses, or animal model experiments. For example, Homaira et al. found through a population cohort study that infants infected with respiratory syncytial virus (RSV) had a significantly increased risk of developing asthma in adulthood. While such surveys are highly reliable, their high cost, long duration, and strong sample dependence make it difficult to comprehensively and systematically uncover potential virus-disease associations.
[0003] In recent years, computational methods based on protein-protein interaction networks have been increasingly applied to predicting disease-disease relationships. For example, Menche et al. proposed that proteins corresponding to related diseases often exhibit spatial clustering in human protein-protein interaction networks. This network topological regularity provides a basis for inferring potential disease associations. These computational methods show promising application prospects in predicting viral disease associations, but they still lack integration of viral characteristics and contain limited multimodal information such as disease phenotypes and functional annotations.
[0004] Therefore, there is still an urgent need to develop a computational framework that integrates multi-source biological features for the efficient and systematic prediction of potential virus-disease associations. Summary of the Invention
[0005] This invention provides a novel method for predicting the association between viruses and diseases—VDP (Virus-DiseasePredictor). This method consists of two independent prediction models, and the results are fused through an integration strategy, which can efficiently and accurately predict the association between viruses and diseases.
[0006] To achieve the above objectives, this invention provides a method for constructing a virus-disease association prediction model, specifically including the following steps:
[0007] S1. Collect and preprocess virus and disease data;
[0008] S2. Extract features from the virus and disease data in S1 using an independent prediction model A. The feature extraction of the independent prediction model A includes: extracting features based on the sequence similarity of the virus; and extracting features based on the semantics of the disease.
[0009] S3. Extract features from the independent prediction model B on the virus and disease data in S1; the feature extraction of the independent prediction model B is as follows: extract features based on functional similarity between viruses; extract features based on functional similarity between diseases; extract features based on functional similarity between viruses and diseases.
[0010] S4. Based on five-fold cross-validation and grid search, determine the best machine learning algorithm for independent prediction model A and independent prediction model B respectively, and construct independent prediction model A and independent prediction model B based on the features extracted in S2 and S3.
[0011] S5. Based on the bagging ensemble strategy, integrate the independent prediction model A and the independent prediction model B to generate a virus-disease association prediction model, and use the virus-disease association prediction model to predict the association probability of the virus and the disease.
[0012] Preferably, the virus and disease data in step S1 specifically include: viral genome sequence, viral classification ID (taxid), virus-related human target genes, GO functional annotations of virus-related human target genes, disease mesh tree, disease name, disease-related human target genes, and GO functional annotations of disease-related human target genes; the preprocessing specifically includes: retaining viruses with no fewer than 10 human target genes, complete genomes, and GO annotations greater than 0, as well as diseases with no fewer than 10 human target genes and GO annotations greater than 0.
[0013] Preferably, the feature extraction of viral sequence similarity in step S2 specifically includes: extracting the embedding representation of the viral genome using the Protein Set Transformer model, calculating the cosine similarity of the embeddings between viruses as the sequence similarity feature of the virus, and the specific definition of cosine similarity is given in formula (1):
[0014] (1)
[0015] in It is the dot product of the viral genome embedding representation vectors. and It is the norm of the viral genome embedding representation vector.
[0016] Preferably, the semantic feature extraction of diseases in step S2 specifically includes: calculating the semantic similarity of different diseases based on their structural positions in the MeSH classification tree. For any disease A, its MeSH classification tree can be represented as: DAGA=( , , ), where DAGA represents the directed acyclic graph of the mesh classification tree corresponding to disease A. The node representing disease A, Represents the set of all ancestor nodes of disease A, including nodes. itself, This represents the set of all links in the graph;
[0017] Diseases in DAGA The semantic contribution to disease A is defined as follows: Diseases in DAGA The semantic contribution to disease A is defined as follows: The calculation formula is as follows: Formula (2):
[0018] (2)
[0019] in Indicates connection to disease and his son's illness The semantic contribution factor of the edge, DAGA neutron disease The semantic contribution to disease A is defined as follows: The semantic contribution of disease A to itself is defined as 1. For disease A, the contribution of distant ancestor nodes to disease A is smaller. Therefore, the semantic value of disease A is defined as:
[0020] (3)
[0021] By considering the relative positions of two diseases in the MeSH classification tree, a semantic similarity measurement method is proposed. It is assumed that diseases with more shared parts in the DAG tend to have higher semantic similarity. The semantic similarity between disease A and disease B is... Defined as:
[0022] (4)
[0023] in, Represents the set of all ancestor nodes of disease A, including nodes. itself, Represents the set of all ancestor nodes of disease B, including nodes itself, Let t be the semantic value related to disease A. The semantic value of disease t related to disease B. For the semantic value of disease A, This is the semantic value of disease B.
[0024] Preferably, the method for extracting the functional similarity features between viruses, the functional similarity features between diseases, and the functional similarity features between viruses and diseases in step S3 specifically includes:
[0025] S3.1. Based on the set of human genes related to viruses and diseases, functional enrichment is used to obtain the set of GO terms for the genes.
[0026] S3.2 Calculate virus-virus functional similarity separately. Disease-disease functional similarity Functional similarity to viruses and diseases The calculation process and definition of functional similarity are shown in the following formulas (5) and (6):
[0027] (5)
[0028] in, Represents a set of GO terms. To represent a single GO term, Represents the first term in the GO term set. GO terms Indicates a pair of terms, Represents the set of GO terms Size; express Similarity of GO term pairs;
[0029] Functional similarity is defined as the similarity between two sets of GO terms. Calculate the similarity between a single GO term and a set of GO terms. The similarity of GO term pairs Calculated using the GOGO tool;
[0030] (6)
[0031] in, This represents the first set of GO terms. Represents the set of GO terms Size; This represents the second set of GO terms. Represents the set of GO terms Size; express The Middle GO terms and Similarity; express The Middle GO terms and The similarity.
[0032] Preferably, the step S4 of determining the optimal machine learning algorithm based on five-fold cross-validation and grid search specifically includes: performing network search and cross-validation (GridSearchCV) on the training set of five-fold cross-validation, and determining the optimal model algorithm based on the AUC value; the optimal machine learning algorithm is selected from the following algorithms: Artificial Neural Network (ANN), Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XgBoost).
[0033] Based on the determined optimal machine learning algorithm and GridSearchCV, hyperparameters are selected, and independent prediction models A and B are constructed based on the features extracted in S2 and S3.
[0034] Preferably, step S5, which involves integrating independent prediction model A and independent prediction model B using a bagging ensemble strategy to generate a virus-disease association prediction model, specifically includes:
[0035] Bagging ensemble is used to train a series of sub-models for independent prediction models A and B. The prediction probability of the final test set samples is derived from the mean of the prediction probabilities of all sub-models of independent prediction models A and B. Its calculation formula is as follows: Formula (7):
[0036] (7)
[0037] in, This represents the bagging ensemble of independent prediction model A and independent prediction model B. express The predicted probabilities of a series of sub-models, express The predicted probabilities of a series of sub-models.
[0038] Under the same technical concept, the present invention also provides a probability prediction model for the association between viruses and diseases, which is constructed using the method for constructing the prediction model for the association between viruses and diseases.
[0039] Under the same technical concept, the present invention also provides a method for predicting the probability of association between viruses and diseases. The prediction method uses a prediction model constructed by the method for constructing the prediction model of association between viruses and diseases, or uses the prediction model to predict the probability of association between viruses and diseases.
[0040] Preferably, in the prediction method, for a certain unknown virus and disease association pair, after collecting and preprocessing data in step S1, the required features for building A and B models are generated through steps S2 and S3. The features are then input into the prediction model constructed using the prediction model or the virus and disease association prediction model construction method to predict the association probability of the virus and disease.
[0041] The above-described solution of the present invention has the following beneficial effects:
[0042] (1) The novel virus and disease association prediction method VDP provided by this invention proposes a new idea of multimodal biological information integration based on existing research; by simultaneously considering the molecular sequence information of the virus, the semantic structure information of the disease and the GO function annotation information of both, a more comprehensive feature representation is achieved, which effectively makes up for the problems of single information source and insufficient prediction accuracy of existing methods.
[0043] (2) A dual-model complementary design is adopted, which independently optimizes two feature systems, namely virus sequence-disease semantics and virus-disease functional information, during the modeling stage, and improves the overall prediction performance and robustness through an ensemble learning strategy. Cross-validation shows that VDP outperforms the single model in terms of AUROC, AUPRC, accuracy, precision, and recall, with an average AUROC of 0.905 and an AUPRC of 0.891, demonstrating its practical application value.
[0044] (3) VDP has good scalability and versatility, and can be applied to other pathogen-human disease association prediction tasks in the future. Attached Figure Description
[0045] Figure 1 A flowchart illustrating the dataset construction and feature extraction process for a virus-disease association prediction method according to the present invention; Figure 1 (A) Flowchart of the process for building the dataset. Figure 1 (B) represents the feature extraction for models A and B;
[0046] Figure 2 This invention provides a flowchart of algorithm selection, model training, and evaluation results for a method for predicting the association between viruses and diseases. Figure 2 (C) is a flowchart of algorithm selection and model training using five-fold cross-validation combined with grid search; Figure 2 (D) shows the AUROC and AUPRC plots of the final model evaluation results;
[0047] Figure 3 This is a Mesh classification tree diagram of breast neoplasms, as an example in Embodiment 1 of the present invention. Detailed Implementation
[0048] To make the technical problems, solutions, and advantages of this invention clearer, a detailed description will be provided below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0049] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.
[0050] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a locking connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.
[0051] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0052] Example 1:
[0053] like Figure 1 , 2 As shown, the specific embodiment of the present invention is as follows:
[0054] Step 1: As Figure 1 (A); Collect and preprocess virus and disease data
[0055] The dataset for this invention is derived from high-quality viral disease association datasets collected by our laboratory. It retains only viruses with no fewer than 10 human target genes, complete genomes, and GO annotations greater than 0, as well as diseases with no fewer than 10 human target genes and GO annotations greater than 0.
[0056] The final dataset contains 1286 known association pairs, involving 424 viruses and 60 diseases, and preserves viral genome sequences, viral tax IDs (taxid), virus-associated human target genes, GO functional annotations of virus-associated human target genes, disease mesh trees, disease names, disease-associated human target genes, and GO functional annotations of disease-associated human target genes.
[0057] For diseases, the corresponding Mesh tree data is obtained by querying the Mesh database.
[0058] by Figure 3 For example, Figure 3 This demonstrates the MeSH classification structure of breast tumors. One path uses "C04; Tumors" as the general category, passing through "C04.588; Tumors by Location" to reach "C04.588.180; Breast Tumors," reflecting a classification from an oncology perspective. The other path uses "C17; Diseases of the Skin and Connective Tissue" as the general category, passing through "C17.800; Dermatology" and "C17.800.090; Breast Diseases" to reach "C17.800.500; Breast Tumors," reflecting a classification based on disease systems. Therefore, "C04.588.180; C17.800.500; Breast Tumors" lies at the intersection of these two paths, reflecting its dual classification within the MeSH system.
[0059] Step 2: As Figure 1 (B) Perform feature extraction for independent prediction model A.
[0060] Step 2.1: Viral sequence similarity: The embedding representation of the viral genome was extracted using the Protein Set Transformer model. After obtaining the embedding representation of the virus, the number of clusters was set to 1286 by K-means clustering, and the center sample of each cluster was taken as the negative sample to form a negative sample dataset. The cosine similarity of the embedding representation between each pair of viruses was calculated according to formula (1) as the sequence similarity between viruses.
[0061] (1)
[0062] in It is the dot product of the viral genome embedding representation vectors. and It is the norm of the viral genome embedding representation vector.
[0063] Step 2.2: Disease semantic similarity: Based on formulas (2), (3) and (4), and combined with the Mesh classification tree data corresponding to the diseases, calculate the semantic similarity between diseases.
[0064] Semantic similarity is calculated based on the structural position of different diseases in the MeSH classification tree. For any disease A, its MeSH classification tree can be represented as: DAGA=( , , ), where DAGA represents the directed acyclic graph of the mesh classification tree corresponding to disease A. The node representing disease A, Represents the set of all ancestor nodes of disease A, including nodes. itself, This represents the set of all links in the graph;
[0065] Diseases in DAGA The semantic contribution to disease A is defined as follows: Diseases in DAGA The semantic contribution to disease A is defined as follows: The calculation formula is as follows: Formula (2):
[0066] (2)
[0067] in Indicates connection to disease and his son's illness The semantic contribution factor of the edge, DAGA neutron disease The semantic contribution to disease A is defined as follows: The semantic contribution of disease A to itself is defined as 1. For disease A, the contribution of distant ancestor nodes to disease A is smaller. Therefore, the semantic value of disease A is defined as:
[0068] (3)
[0069] By considering the relative positions of two diseases in the MeSH classification tree, a semantic similarity measurement method is proposed. It is assumed that diseases with more shared parts in the DAG tend to have higher semantic similarity. The semantic similarity between disease A and disease B is... Defined as:
[0070] (4)
[0071] in, Represents the set of all ancestor nodes of disease A, including nodes. itself, Represents the set of all ancestor nodes of disease B, including nodes itself, Let t be the semantic value related to disease A. The semantic value of disease t related to disease B. For the semantic value of disease A, This is the semantic value of disease B.
[0072] Step 3: As Figure 1 (B). Feature extraction of independent prediction model B
[0073] Step 3.1: Based on the set of human genes related to viruses and diseases, use the clusterProfiler package in R language to perform functional enrichment analysis to obtain the biological function annotations corresponding to these genes, thereby obtaining a significantly enriched set of GO terms.
[0074] Step 3.2: Based on formulas (5) and (6), and combining the respective GO term sets of viruses and diseases, calculate the virus-virus functional similarity Sim(V,V), disease-disease functional similarity Sim(D,D), and virus-disease functional similarity Sim(V,D) respectively. Calculate the virus-virus functional similarity separately. Disease-disease functional similarity Functional similarity to viruses and diseases The calculation process and definition of functional similarity are shown in the following formulas (5) and (6):
[0075] (5)
[0076] in, Represents a set of GO terms. To represent a single GO term, Represents the first term in the GO term set. GO terms Indicates a pair of terms, Represents the set of GO terms Size; express Similarity of GO term pairs;
[0077] Functional similarity is defined as the similarity between two sets of GO terms. Calculate the similarity between a single GO term and a set of GO terms. The similarity of GO term pairs Calculated using the GOGO tool;
[0078] (6)
[0079] in, This represents the first set of GO terms. Represents the set of GO terms Size; This represents the second set of GO terms. Represents the set of GO terms Size; express The Middle GO terms and Similarity; express The Middle GO terms and The similarity.
[0080] Step 4: As Figure 2 (C). Model algorithm selection.
[0081] First, five-fold cross-validation is performed on the entire dataset to obtain five training sets (TrainSet). The optimal model algorithm is determined based on the average AUC value of the five-fold cross-validation on the training sets (TrainSet1 to TrainSet5) for different algorithms. Each time five-fold cross-validation is performed on the training sets, GridSearchCV is used on the non-test data to automatically determine the optimal hyperparameters. The best machine learning algorithm is selected from the following: Artificial Neural Network (ANN), Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XgBoost).
[0082] Based on the results in Table 1, the algorithm used for the final independent prediction model A in this invention is RF, and the algorithm used for the independent prediction model B is XGBoost.
[0083] Table 1. AUC values of different algorithms for Model A and Model B on the training set during cross-validation.
[0084] Step 5: As Figure 2 (C); bagging integration
[0085] For each five-fold cross-validation, we perform random sampling with replacement on the training set to form a series of training subsets.
[0086] A series of base learners are built by training independent prediction models A and B on these training subsets. For each base learner, GridSearchCV is used to automatically determine the optimal hyperparameters on the training subsets. The prediction probability of the final test set samples is derived from the mean of the prediction probabilities of all sub-models of independent prediction models A and B. Its calculation formula is as follows: Formula (7):
[0087] (7)
[0088] in, This represents the bagging ensemble of independent prediction model A and independent prediction model B. express The predicted probabilities of a series of sub-models, express The predicted probabilities of a series of sub-models.
[0089] According to Formula 7, the predicted probability of a sample 1 is ultimately composed of two parts: the mean probability from the base learners of independent prediction model A and independent prediction model B. Figure 2 (C) All values shown in the table are means. The final predicted probability of sample 1 is the mean probability of independent prediction model A and independent prediction model B.
[0090] Validation of the invention: The performance of the invention is evaluated by implementing 5-fold cross-validation.
[0091] In the 5-fold cross-validation, all virus-disease association data were randomly divided into 5 parts. One part was used as the test set each time, and the remaining 4 parts were used as the training set. The area under the ROC curve (AUROC) and the area under the PR curve (AUPRC) were used to quantitatively evaluate the performance of the invention. Specific performance data are as follows: Figure 2 As shown in (D), in terms of the AUROC metric, Model A and Model B achieved scores of 0.891 and 0.885, respectively, while the final model VDP of this invention, after bagging ensemble, further improved the performance to 0.905. In terms of the AUPRC metric, Model A and Model B scored 0.874 and 0.867, respectively, while the VDP model reached 0.891. Therefore, it is evident that the VDP model of this invention performs better in classification performance than the single model, exhibiting the most outstanding overall performance.
Claims
1. A method for constructing a virus and disease association prediction model, characterized in that, Specifically comprising the following steps: S1, collecting and preprocessing virus and disease data; S2, feature extraction of independent prediction model A on virus and disease data in S1; the feature extraction of the independent prediction model A is: feature extraction of sequence similarity of viruses; feature extraction of semantics of diseases; S3, feature extraction of independent prediction model B on virus and disease data in S1; the feature extraction of the independent prediction model B is: feature extraction of functional similarity features between viruses; feature extraction of functional similarity features between diseases; feature extraction of functional similarity features between viruses and diseases; S4, based on five-fold cross-validation and grid search, determine the best machine learning algorithm for independent prediction model A and independent prediction model B respectively, and construct independent prediction model A and independent prediction model B based on the features extracted in S2, S3; S5, based on bagging ensemble strategy, integrate independent prediction model A and independent prediction model B to generate virus and disease association prediction model.
2. The construction method of claim 1, wherein, The virus and disease data in step S1 specifically includes: virus genome sequence, virus classification ID, virus related human target gene, GO function annotation of virus related human target gene, disease Mesh tree, disease name, disease related human target gene, GO function annotation of disease related human target gene; the preprocessing specifically includes: retaining viruses with human target genes not less than 10, complete genome, and GO annotation quantity greater than 0, and diseases with human target genes not less than 10 and GO annotation quantity greater than 0.
3. The construction method of claim 1, wherein, The feature extraction of sequence similarity of viruses in step S2 specifically includes: using Protein Set Transformer model to extract embedded representation of virus genome, calculating cosine similarity of virus embedding as sequence similarity feature of virus, and the definition of cosine similarity is specifically shown in formula (1): （1） wherein is the dot product of the viral genome embedding representation, and is the norm of the viral genome embedding representation.
4. The construction method of claim 1, wherein, Step S2, which involves extracting semantic features of diseases, specifically includes: calculating the semantic similarity of different diseases based on their structural positions in the MeSH classification tree. For any disease A, its MeSH classification tree can be represented as: DAGA=( , , ), where DAGA represents the directed acyclic graph of the mesh classification tree corresponding to disease A. The node representing disease A, Represents the set of all ancestor nodes of disease A, including nodes. itself, This represents the set of all links in the graph; Disease A in DAG A The semantic contribution of disease A is defined as Disease A in DAG A The semantic contribution of disease A is defined as The formula is as follows (2): （2） wherein represents the semantic contribution factor of the edge connecting disease and its child diseases in the DAGA, child diseases The semantic contribution of disease A is defined as , the semantic contribution of disease A to itself is defined as 1, and the farther the ancestor node is from disease A, the smaller the contribution of disease A is, and therefore, the semantic value of disease A is defined as: （3） By considering the relative position of the two diseases in the MeSH classification tree, a semantic similarity measurement method is proposed, assuming that the diseases sharing more parts in the DAG often have higher semantic similarity. The semantic similarity of disease A and disease B is defined as: where is the semantic similarity of disease A and disease B. （4） wherein, denotes the set of all ancestor nodes of disease A, including node itself, denotes the set of all ancestor nodes of disease B, including node itself, is the semantic value of disease t related to disease A, is the semantic value of disease t related to disease B, is the semantic value of disease A, is the semantic value of disease B.
5. The construction method of claim 1 wherein, The extraction method of functional similarity features between viruses, between diseases, and between viruses and diseases in step S3 specifically includes: S3.1, according to the virus and disease related human gene set, function enrichment is obtained for the GO term set of the gene; S3.2, respectively calculating virus-virus functional similarity disease-disease functional similarity and virus-disease functional similarity The calculation process and definition of functional similarity are shown in the following formulas (5), (6): （5） wherein, represents a set of GO terms, represents a single GO term, represents the i-th GO term in a set of GO terms, represents a pair of GO terms, represents a pair of terms, represents the size of a set of GO terms represents the similarity of a pair of GO terms. Functional similarity is defined as the similarity between two sets of GO terms The similarity between a single GO term and a set of GO terms is computed The similarity between a pair of GO terms is computed by the GO GO tool The similarity between a pair of GO terms is computed by the GO GO tool （6） wherein, represents the size of the first set of GO terms, represents the set of GO terms represents the size of the first set of GO terms, represents the second set of GO terms, represents the set of GO terms represents the size of the second set of GO terms, represents the similarity of the th GO term in to represents the similarity of the th GO term in to 6. The construction method of claim 1 wherein, The determination of the best machine learning algorithm based on five-fold cross-validation and grid search in step S4 specifically includes: performing network search and cross-validation on the training set of five-fold cross-validation, and determining the optimal model algorithm according to the AUC value; the best machine learning algorithm is selected from the following algorithms: artificial neural network ANN, random forest RF, support vector machine SVM, and extreme gradient boosting XgBoost; According to the determined best machine learning algorithm, network search and cross-validation are used to screen hyperparameters, and independent prediction model A and independent prediction model B are constructed based on the features extracted in S2, S3.
7. The construction method of claim 1 wherein, The integration of independent prediction model A and independent prediction model B based on bagging ensemble strategy in step S5 to generate virus and disease association prediction model specifically includes: Bagging ensemble is adopted for the independent prediction model A and the independent prediction model B, a series of sub-models are trained, and the prediction probability of the final test set sample is derived from the mean of the prediction probabilities of all sub-models of the independent prediction model A and the independent prediction model B ; the calculation formula is as follows formula (7): （7） wherein, denotes a bagging ensemble of independent prediction models A and B, denotes a prediction probability of a series of submodels, denotes a prediction probability of a series of submodels.
8. A virus and disease association probability prediction model, characterized by, The prediction model is constructed by the virus and disease association prediction model construction method according to any one of claims 1-7.
9. A method of predicting a probability of association of a virus and a disease, characterized by, The prediction method uses the prediction model constructed by the virus and disease association prediction model construction method according to any one of claims 1-7 or the prediction model according to claim 9 to predict the association probability of the virus and the disease.
10. The prediction method of claim 9, wherein, For a certain unknown virus and disease association pair, after the data is collected and preprocessed in step S1, the features required for constructing the A and B models are generated in steps S2 and S3, and the features are input into the prediction model constructed by the prediction model according to claim 8 or the virus and disease association prediction model construction method according to any one of claims 1-7 to predict the association probability of the virus and the disease.