A fusion sequence and network-embedded method for predicting viral-host protein-protein interactions
Through the method of fusion sequence and network embedding, the Doc2vec and Node2vec algorithms are used to extract features and combine long and short-term memory networks, which improves the accuracy of prediction of virus host protein-protein interactions, and solves the problem of insufficient prediction accuracy in the prior art. Efficient prediction results are achieved.
Patent Information
- Application Number
- CN202210651688.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-10
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2042-06-10
AI Technical Summary
The existing viral host protein-protein interaction prediction methods are insufficient in terms of accuracy, and it is difficult to effectively identify protein interactions between viruses and host cells, affecting vaccine development and drug target search.
Using the method of fusion sequence and network embedding, protein sequences and network features are extracted through Doc2vec and Node2vec algorithms, combined with long and short-term memory networks for training, and a prediction model is constructed to improve the prediction accuracy.
The accuracy of the prediction of viral host protein-protein interactions was significantly improved, and the performance was better than other methods. The area under the ROC curve (AUC) reached 0.94, verifying the effectiveness of the method.
Smart Images

Figure CN115035954B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of bioinformatics, and in particular relates to a fusion sequence and network embedded virus host protein-protein interaction prediction method. Background Art
[0002] At present, infectious diseases caused by viral infections have become a major public health problem, posing a serious threat to global human health and causing significant losses to the global economy. Viruses are non-cellular organisms that contain only one type of nucleic acid and can only complete their life cycle within host cells. In the process of viral infection of the host, the virus transports its own genetic material to the host cell nucleus through interspecies interactions of proteins, controls host cell metabolism, and destroys host cell functions. Therefore, identifying virus-host protein-protein interactions is of great significance for understanding the mechanisms of viral infection and host immune response, and for finding new drug targets and vaccine development. Summary of the invention
[0003] The purpose of the present invention is to provide a fusion sequence and network embedded virus host protein-protein interaction prediction method, which can effectively improve the accuracy of virus host protein-protein interaction prediction.
[0004] To achieve the above object, the present invention adopts the following technical solution:
[0005] A method for predicting viral host protein-protein interactions by fusion sequence and network embedding, characterized in that it comprises the following steps:
[0006] Step 1: Preprocess the data;
[0007] Step 2: Use the Doc2vec method to extract viral and host protein sequence features and vectorize the protein sequence;
[0008] Step 3: construct host protein-protein interaction network and viral protein sequence similarity network;
[0009] Step 4: Use the Node2vec algorithm to extract the features of the above two networks, vectorize the networks and perform a scalar product with the sequence features;
[0010] Step 5: Based on the obtained matrix integrating the above two features, a long short-term memory network is constructed for training, and the trained model is used for prediction.
[0011] The step 1 is specifically as follows:
[0012] Step 1.1: Preprocess the virus-host protein-protein interaction data and protein sequence information to remove outliers and missing values;
[0013] Step 1.2: Generate a virus-host protein-protein interaction matrix based on the interaction data.
[0014] The step 2 is specifically as follows:
[0015] Step 2.1: Use non-redundant protein sequences with a length between 30 and 5000 amino acids from the SwissProt database, remove redundant sequences with sequence similarity > 0.5, and use the remaining sequences as the corpus for training the Doc2vec model to train the Doc2vec model;
[0016] Step 2.2: Use the trained Doc2vec model to obtain the viral and host protein sequence features and obtain the viral and host protein sequence vectors.
[0017] The step 3 is specifically as follows:
[0018] Step 3.1: Generate a host protein-protein interaction network based on host protein interaction data;
[0019] Step 3.2: Use the Smith-Waterman algorithm to calculate the similarity between viral proteins. The calculation method is as follows:
[0020]
[0021] Among them, LS(l(i), l(j)) is the similarity matrix constructed for proteins; sw(l(i), l(j)) represents the similarity between two protein sequences.
[0022] The step 4 is specifically as follows:
[0023] Step 4.1: The random walk strategy is controlled by hyperparameters p and q to sample the node sequence of the host protein interaction network, where the transition probability between two nodes is:
[0024] π vx =α pq (t, x)·ω vx (2)
[0025] Where v is the current node, t is the node of v's previous step, x represents the position of the next step, ω is the weight of the edge between the two nodes, and α is the search bias, which is defined as:
[0026]
[0027] Among them, p and q are hyperparameters, d tx represents the shortest distance between t and x;
[0028] Step 4.2: The node sequence uses the Skip-Gram model to obtain the network characteristics of each node. The calculation method is as follows:
[0029]
[0030] in: However, due to its high computational cost, we use negative sampling technology to optimize it.
[0031] Step 4.3: perform a quantitative product of the obtained host protein interaction network features and the host protein sequence features obtained in step 2;
[0032] Step 4.4: Using the same method as in step 4.1 and step 4.2, extract network features from the viral protein sequence similarity network, and perform a quantitative product of the extracted network features and sequence features.
[0033] The step 5 is specifically as follows:
[0034] According to the obtained matrix of fusion sequence and network features, a long short-term memory network is constructed for learning modeling, and finally the prediction result is obtained.
[0035] The long short-term memory network model is specifically:
[0036] f t =σ(w f ·[h t-1 , x t ]+b f ) (5)
[0037] i t =σ(w i ·[h t-1 , x t ]+b i ) (6)
[0038]
[0039]
[0040] O t =σ(w o [h t-1 , x t ]+b o ) (9)
[0041] h t =O t tanh(C t ) (10)
[0042] In the formula, f trepresents the forget gate, σ is the sigmoid function, w is the weight matrix, h is the hidden node, x t is the input data, b is the bias direction, i t represents the input gate, Indicates the updated value of the unit state, C t Indicates the current unit state, C t-1 is the last unit state, O t Represents the output gate.
[0043] Compared with the prior art, the present invention has the following beneficial effects:
[0044] 1. The present invention takes into account the available features of different aspects of proteins, integrates the host protein interaction network and the virus protein similarity network features with the protein sequence features, and increases the effectiveness of protein feature extraction;
[0045] 2. The long short-term memory network model is used to model the virus and host protein sequences and network features, which effectively learns the sequence and network features and better utilizes the sequence and network features to improve the accuracy of interaction predictions. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] Figure 1 It is a model flow chart;
[0047] Figure 2 : It is the ROC curve of the performance comparison between the embodiment of the present invention and other methods under 5-fold cross validation. DETAILED DESCRIPTION
[0048] The present invention will be further described below with reference to the accompanying drawings and examples.
[0049] Embodiment 1:
[0050] like Figure 1 As shown, the specific embodiment process of the present invention is:
[0051] Step 1: First, 33,300 protein interaction pairs of 1,046 viral proteins and 7,459 host proteins were obtained from the VirHostnet 2.0 database. After removing duplicate, abnormal and missing data, 29,940 protein interaction pairs remained. Then, the sequence information of viral and host proteins was obtained by querying the Uniprot database. Finally, the host protein-protein interaction pair data in the String database were queried.
[0052] Step 2: Obtain non-redundant protein sequences with a length between 30 and 5000 amino acids in the SwissProt database, remove redundant (sequence similarity> 0.5) sequences, and use the remaining sequences as the corpus for training the Doc2vec model. Use the trained Doc2vec model to obtain the sequence features of the virus and host proteins and obtain the sequence vectors of the virus and host proteins.
[0053] Step 3: Use the host protein interaction pairs obtained in the String database to construct a host protein-protein interaction network. Use the Smith-Waterman algorithm to calculate the similarity between each viral protein sequence and generate a viral protein sequence similarity network. The calculation method is as follows:
[0054]
[0055] Among them, LS(l(i), l(j)) is the similarity matrix constructed for proteins; sw(l(i), l(j)) represents the similarity between two protein sequences.
[0056] Step 4: Use the Node2vec algorithm to extract the characteristics of the host protein-protein interaction network. Node2vec uses hyperparameters p and q to control the random walk strategy and sample the host protein interaction network node sequence, where the transition probability between two nodes is:
[0057] π vx =α pq (t, x)·ω vx (2)
[0058] In the formula, v is the current node, t is the node of v's previous step, x represents the position of the next step, ω is the weight of the edge between the two nodes, and α is the search bias, which is defined as:
[0059]
[0060] In the formula, p and q are hyperparameters, d tx Represents the shortest distance between t and x.
[0061] The node sequence obtains the network characteristics of each node through the deep learning Skip-Gram model. The calculation method is as follows:
[0062]
[0063] in: However, due to its high computational cost, we use negative sampling technology to optimize it.
[0064] Finally, the obtained host protein interaction network features are quantitatively multiplied with the host protein sequence features obtained in step 2. Similarly, the extracted virus protein sequence similarity network features are quantitatively multiplied with the virus protein sequence features.
[0065] Step 5: Based on the obtained matrix integrating the above two features, the long short-term memory network is used for learning to obtain the prediction result. The specific model is:
[0066] f t =σ(w f ·[h t-1 , x t ]+b f ) (5)
[0067] i t =σ(w i ·[h t-1 , x t ]+b i ) (6)
[0068]
[0069]
[0070] O t =σ(w o [h t-1 , x t ]+b o ) (9)
[0071] h t =O t tanh(C t ) (10)
[0072] In the formula, f t represents the forget gate, σ is the sigmoid function, w is the weight matrix, h is the hidden node, x t is the input data, b is the bias direction, i t represents the input gate, Indicates the updated value of the unit state, C t Indicates the current unit state, C t-1 is the last unit state, O t Represents the output gate.
[0073] Verification of the effectiveness of the present invention:
[0074] The performance of the present invention was evaluated by implementing 5-fold cross validation. In the 5-fold cross validation, all virus-host protein-protein interaction data were randomly divided into 5 parts, one of which was taken as the test set each time, and the remaining 4 were used as training sets. The area under the ROC curve (AUC) was used to quantitatively evaluate the performance of the present invention. To prevent randomness, the validation results were the average of 100 5-fold cross validations. The present invention was compared with other methods, and the results are as follows: Figure 2 As shown, the present invention achieved the highest AUC value, AUC = 0.94. The verification results show that the performance of the present invention is better than other methods.
Claims
1. A method for predicting viral-host protein-protein interactions based on fusion sequences and network embedding, It is characterized in that The following steps are involved: Step 1: Preprocess the data; Step 2: Use the Doc2vec method to extract viral and host protein sequence features and vectorize the protein sequence; Step 2.1: Use non-redundant protein sequences with a length between 30 and 5000 amino acids from the SwissProt database, remove redundant sequences with sequence similarity > 0.5, and use the remaining sequences as the corpus for training the Doc2vec model to train the Doc2vec model; Step 2.2: Use the trained Doc2vec model to obtain the viral and host protein sequence features and obtain the viral and host protein sequence vectors; Step 3: construct host protein-protein interaction network and viral protein sequence similarity network; Step 3.1: Generate a host protein-protein interaction network based on host protein interaction data; Step 3.2: Use the Smith-Waterman algorithm to calculate the similarity between viral proteins. The calculation method is as follows: Among them, LS(l(i), l(j)) is the similarity matrix constructed for proteins; sw(l(i), l(j)) represents the similarity of two protein sequences; Step 4: Use the Node2vec algorithm to extract the features of the above two networks, vectorize the networks and perform a scalar product with the sequence features; Step 4.1: The random walk strategy is controlled by hyperparameters p and q, and the node sequence of the host protein interaction network is sampled, where the transition probability between two nodes is: p vx =a pq (t,x)·ω vx (2) Where v is the current node, t is the node of v's previous step, x represents the position of the next step, ω is the weight of the edge between the two nodes, and α is the search bias, which is defined as: Among them, p and q are hyperparameters, d tx represents the shortest distance between t and x; Step 4.2: The node sequence uses the Skip-Gram model to obtain the network characteristics of each node. The calculation method is as follows: in: However, due to its high computational cost, negative sampling technique is used for optimization; Step 4.3: perform a quantitative product of the obtained host protein interaction network features and the host protein sequence features obtained in step 2; Step 4.4: Using the same method as in step 4.1 and step 4.2, extract network features from the viral protein sequence similarity network, and perform a quantitative product between the extracted network features and the sequence features; Step 5: Based on the obtained matrix integrating the above two features, a long short-term memory network is constructed for training, and the trained model is used for prediction.
2. The method for predicting viral host protein-protein interactions based on fusion sequences and network embedding according to claim 1, It is characterized in that The step 5 is specifically as follows: According to the obtained matrix of fusion sequence and network features, a long short-term memory network is constructed for learning modeling, and finally the prediction result is obtained.
3. The method for predicting viral host protein-protein interactions based on fusion sequences and network embedding according to claim 2, It is characterized in that The long short-term memory network model is specifically: f t =σ(w f ·[h t-1 ,x t ]+b f ) (5) i t =σ(wi·[h t-1 ,x t ]+b i ) (6) The t =σ(w o [h t-1 ,x t ]+b o ) (9) h t =O t ·tanh(C t ) (10) In the formula, f t represents the forget gate, σ is the sigmoid function, w is the weight matrix, h is the hidden node, x t is the input data, b is the bias direction, i t represents the input gate, Indicates the updated value of the unit state, C t Indicates the current unit state, C t-1 is the last unit state, O t Represents the output gate.
4. The method for predicting viral host protein-protein interactions based on fusion sequences and network embedding according to claim 1, It is characterized in that The step 1 is specifically as follows: Step 1.1: Preprocess the virus-host protein-protein interaction data and protein sequence information to remove outliers and missing values; Step 1.2: Generate a virus-host protein-protein interaction matrix based on the interaction data.
