Virus particle protein prediction model, method, equipment and storage medium

Abstract

The invention is applicable to the technical field of bioinformatics, and provides a virion protein prediction model, method, equipment and medium, the virion protein prediction model comprises a data processing and translation module, a feature extraction module, a prediction module and an annotation module which are connected in sequence, the data processing and translating module is used for translating a whole genome sequence of the phage to obtain protein sequence data, and the feature extraction module is used for extracting features of the protein sequence data through a protein language model to obtain high-dimensional feature representation of the protein sequence data; the prediction module is used for processing and predicting the high-dimensional feature representation, and the annotation module is used for carrying out annotation classification on the predicted virion protein to obtain a finer annotation result of the virion protein. The protein sequences are subjected to feature extraction through the protein language model, so that the virion protein prediction model can still have excellent performance in a high-similarity environment.

Description

Technical Field

[0001] The present invention belongs to the technical field of bioinformatics, and in particular relates to a virus particle protein prediction model, method, device and storage medium. Background Art

[0002] As one of the most widely distributed and numerous biological entities in the biosphere, bacteriophages play an important role in regulating the dynamics of microbial systems by lysing bacteria and mediating the horizontal transfer of genetic material. In addition, bacteriophages also have an important impact in many application fields such as food industry, disease diagnosis, bacterial genome engineering, and phage therapy. Phage genome annotation is the basis for promoting the application of phages in these fields, especially the annotation of virion proteins (PVP) and non-virion proteins (non-PVP), which are regarded as one of the important bases for phage classification by the International Committee on Taxonomy of Viruses (ICTV). Compared with traditional wet experiments, the use of computer methods to identify virion proteins of phages can save a lot of manpower and material resources and improve efficiency.

[0003] Among the computer methods, they can be roughly divided into the following categories:

[0004] (1) Traditional machine learning: Four models (SVM, NB, RF and SCM) are often used in traditional machine learning methods. These methods often extract manual features of phage protein sequences and then use manual features to identify PVP. However, for protein sequences with high sequence similarity, only considering manual features is not enough to provide accurate identification.

[0005] (2) Ensemble learning: Ensemble learning-based methods usually use multiple models to make joint decisions to improve prediction accuracy. For example, Meta-iPVP uses a new feature representation scheme and four machine learning algorithms to encode seven input features into a probability matrix. Then, the generated probability matrix is ​​input into the support vector machine model (SVM) to classify PVP. Compared with making decisions by a single model, this type of method improves the accuracy of prediction, but due to the limitations of feature representation, the performance improvement is very limited.

[0006] (3) Deep learning: Deep learning-based methods usually use some neural networks to perform high-dimensional feature representation on the original protein sequence data, thereby improving the prediction performance of the model. For example, VirionFinder and DeePVP both use convolutional neural networks (CNN) as classifiers. By comparing existing tools, it can be found that CNN is an effective method for extracting abstract features of biological sequences.

[0007] Although the above methods have achieved good performance, they still have some limitations. First, in terms of feature extraction, they rely too much on features related to sequence similarity and cannot mine the potential high-dimensional information of the original protein sequence. This shortcoming is particularly obvious in an environment with high sequence similarity. Second, they do not consider the performance changes of the model in an environment with an imbalance of positive and negative samples. In real environments, samples are often unbalanced. Summary of the invention

[0008] The purpose of the present invention is to provide a virus particle protein prediction model, method, device and storage medium, aiming to solve the problem that the existing technology is overly dependent on features related to sequence similarity and cannot mine the potential high-dimensional information of the original protein sequence.

[0009] In one aspect, the present invention provides a virus particle protein prediction model comprising a data processing and translation module, a feature extraction module, a prediction module and an annotation module connected in sequence, wherein:

[0010] The data processing and translation module is used to translate the input full genome sequence of the bacteriophage to obtain protein sequence data corresponding to the bacteriophage;

[0011] The feature extraction module is used to extract features from the protein sequence data using a protein language model to obtain a high-dimensional feature representation of the protein sequence data;

[0012] The prediction module is used to process the high-dimensional feature representation and predict the virion protein of the bacteriophage according to the processing result;

[0013] The annotation module is used to annotate and classify the predicted virion proteins to obtain detailed classification annotation results of the virion proteins.

[0014] Preferably, the protein language model is the ProtT5 protein language model.

[0015] Preferably, the prediction module includes a neural network based on a CNN model, and the neural network based on the CNN model is used to process the high-dimensional feature representation.

[0016] Preferably, the neural network based on the CNN model includes a one-dimensional convolution layer, a batch normalization layer and a maximum pooling layer connected in sequence, wherein:

[0017] The one-dimensional convolution layer is used to upsample the high-dimensional feature representation to obtain a multi-channel feature representation;

[0018] The batch normalization layer is used to normalize the feature representation of the multi-channel to obtain a normalized feature representation;

[0019] The maximum pooling layer is used to perform a pooling operation on the normalized feature representation.

[0020] Preferably, the loss function used in training the virus particle protein prediction model is an ASL asymmetric loss function.

[0021] Preferably, the true labels are smoothed during the training of the virus particle protein prediction model.

[0022] Preferably, the ASL asymmetric loss function is:

[0023]

[0024] Among them, Log i represents the Logsoftmax function, x i represents the original prediction score of the i-th category, x j represents the original prediction score of the jth category, xspos represents the predicted probability of the virion protein prediction model for the target category, xsneg represents the complementary value of the predicted probability of the virion protein prediction model for the non-target category, and y i represents the true label of the i-th sample, asymmetricWeight i represents asymmetric weight, γ pos Represents the weight attenuation factor controlling the positive sample, γ neg Represents the weight attenuation factor controlling the negative sample, Loss i represents the loss value of the i-th sample, Loss represents the total loss value, and y ij Indicates whether the target category of the i-th sample is j. When y ij =1 means yes, y ij =0 means no.

[0025] The formula used for smoothing the label is:

[0026]

[0027] Among them, y′ i represents the smoothed label, y i represents the true label of the i-th sample, ∈ represents the label smoothing factor, and C represents the number of categories.

[0028] In another aspect, the present invention provides a method for reminding incoming calls, the method comprising the following steps:

[0029] Obtain the complete genome sequence of the phage;

[0030] The whole genome sequence is input into the virion protein prediction model as described above to obtain the detailed classification annotation results of the virion protein.

[0031] On the other hand, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the function of the viral particle protein prediction model as described above when executing the computer program.

[0032] On the other hand, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the function of the viral particle protein prediction model as described above is realized.

[0033] The virion protein prediction model in the present invention includes a data processing and translation module, a feature extraction module, a prediction module and an annotation module connected in sequence, wherein the data processing and translation module is used to translate the input phage full genome sequence to obtain protein sequence data corresponding to the phage, the feature extraction module is used to extract features from the protein sequence data through a protein language model to obtain a high-dimensional feature representation of the protein sequence data, the prediction module is used to process the high-dimensional feature representation, and predict the virion protein of the phage according to the processing result, and the annotation module is used to annotate and classify the predicted virion protein to obtain a detailed classification and annotation result of the virion protein. Thus, by combining the language model with the convolutional neural network, a virion protein prediction model is obtained, and the protein sequence is feature extracted through the protein language model to mine the potential high-dimensional feature information of the original protein sequence, thereby enabling the virion protein prediction model to still perform well in a high-similarity environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1A is a schematic diagram of the structure of the virus particle protein prediction model provided in Example 1 of the present invention;

[0035] Figure 1B This is a structural example diagram of the virus particle protein prediction model provided in Example 1 of the present invention;

[0036] Figure 2 is a flowchart of the method for predicting viral particle proteins provided in Example 2 of the present invention; and

[0037] Figure 3 It is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0039] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments:

[0040] Embodiment 1:

[0041] Figure 1A and Figure 1B The structure of the virion protein prediction model provided in the first embodiment of the present invention is shown. For the sake of convenience, only the part related to the embodiment of the present invention is shown. The virion protein prediction model includes a data processing and translation module 10, a feature extraction module 11, a prediction module 12 and an annotation module 13 connected in sequence, wherein the data processing and translation module 10 is used to translate the input whole genome sequence of the bacteriophage to obtain protein sequence data corresponding to the bacteriophage; the feature extraction module 11 is used to extract features from the protein sequence data through a protein language model to obtain a high-dimensional feature representation of the protein sequence data; the prediction module 12 is used to process the high-dimensional feature representation and predict the virion protein of the bacteriophage according to the processing result; the annotation module 13 is used to annotate and classify the predicted virion protein to obtain a detailed classification annotation result of the virion protein.

[0042] Specifically, the data processing and translation module 10 may include a Prodigal tool. The data processing and translation module 10 may receive input phage whole genome data, use the Prodigal tool to translate the phage whole genome data, and obtain a corresponding set of protein sequence data.

[0043] The above-mentioned protein language model can be ESM-1b or ESM-1v, etc. Optionally, the protein language model is a ProtT5 protein language model, so that the protein sequence is feature extracted by the ProtT5 protein language model to obtain the potential high-dimensional feature information of the original protein sequence, so that the virus particle protein prediction model can still perform well in a high-similarity environment. Specifically, the feature extraction module 11 can perform preliminary feature extraction processing on the above-mentioned protein sequence data to obtain a high-dimensional feature representation, which is usually a feature matrix, specifically a 1024-dimensional feature representation. Figure 1BAs an example, the length of each protein sequence in this group of protein sequences is M, K, Q, N, L, N, G...A, L. Assuming that the length of the protein sequence is L, after passing through the ProtT5 encoder, each amino acid will be encoded into a 1024-dimensional feature, so the protein sequence will be encoded into an L×1024-dimensional feature matrix. In order to make protein sequences of different lengths have features of the same shape, preferably, the feature matrix is ​​averaged. Continuing with the above example of a protein sequence being encoded into an L×1024-dimensional feature matrix, after Figure 1B After averaging the last hidden layer, a 1×1024 feature representation can be obtained.

[0044] Preferably, the prediction module 11 includes a neural network based on a CNN model, and the neural network based on the CNN model is used to process high-dimensional feature representations, thereby effectively increasing the resolution of the feature map, so that the virus particle protein prediction model can better understand the spatial structure of the input data and make more accurate predictions. Further preferably, the neural network based on the CNN model includes a one-dimensional convolution layer, a batch normalization layer, and a maximum pooling layer connected in sequence, wherein the one-dimensional convolution layer is used to upsample the high-dimensional feature representation to obtain a multi-channel feature representation, the batch normalization layer is used to normalize the multi-channel feature representation to obtain a normalized feature representation, and the maximum pooling layer is used to perform a pooling operation on the normalized feature representation, thereby realizing the processing of high-dimensional feature representations through the network structure based on the CNN model. In a specific implementation, in the neural network based on the CNN model, the convolution kernel size of the one-dimensional convolution layer can be 3, the step size can be 1, the number of input channels is usually 1, and the number of output channels, that is, the number of convolution kernels, can be 32. The convolution kernel size of the maximum pooling layer can be 3, and the step size can be 2. The maximum pooling layer can increase the receptive field of the network through the pooling operation, and reduce the influence of noise on the virion protein prediction model to a certain extent. Furthermore, the prediction module 11 can also include a fully connected neural network, which performs nonlinear processing on the features processed by the neural network based on the CNN model through the fully connected neural network, and finally predicts to obtain the prediction result of the protein. Among them, the prediction result can include PVP (virion protein) or non-PVP (non-virion protein), that is, which of the above-mentioned protein sequences are PVP and which are non-PVP.

[0045] Specifically, the annotation module 13 is used to make a more detailed annotation classification of the predicted PVP, and obtain a detailed classification annotation result of the virus particle protein. The annotation module 13 can use a multi-classification model to perform multi-classification on the predicted PVP, make a more detailed annotation, and output a detailed classification annotation result. Among them, the parameters of the last layer of the fully connected layer of the model can be specifically set to 6.

[0046] Optionally, the loss function used in training the virion protein prediction model is an ASL asymmetric loss function, so as to effectively alleviate the negative impact of the imbalance of positive and negative samples in real scenarios, so that the virion protein prediction model has the potential to be put into use in real scenarios.

[0047] Further optionally, the ASL asymmetric loss function is:

[0048]

[0049] Among them, Log i represents the Logsoftmax function, x i represents the original prediction score of the i-th category, x j represents the original prediction score of the jth category, xspos represents the predicted probability of the virion protein prediction model for the target category, xsneg represents the complementary value of the predicted probability of the virion protein prediction model for the non-target category, and y i represents the true label of the i-th sample, asymmetricWeight i represents asymmetric weight, γ pos Represents the weight attenuation factor controlling the positive sample, γ neg Represents the weight attenuation factor that controls the negative sample, Loss represents the loss value of the i-th sample, Loss represents the total loss value, y ij Indicates whether the target category of the i-th sample is j. When y ij =1 means yes, y ij =0 means no.

[0050] In the specific implementation, the asymmetric weight parameter γ of the positive and negative samples in the loss function can be adjusted pos and γ neg , manually set the weights of positive and negative samples, so that the virion protein prediction model can be better trained for categories with fewer samples.

[0051] Preferably, the true labels are smoothed during the training of the virion protein prediction model.

[0052] Further optionally, the formula used for label smoothing is:

[0053]

[0054] Among them, y i ′ represents the smoothed label, y i represents the true label of the i-th sample, ∈ represents the label smoothing factor, and C represents the number of categories.

[0055] In an embodiment of the present invention, the virion protein prediction model includes a data processing and translation module, a feature extraction module, a prediction module and an annotation module connected in sequence, wherein the data processing and translation module is used to translate the input phage genome sequence to obtain protein sequence data corresponding to the phage, the feature extraction module is used to extract features from the protein sequence data through a protein language model to obtain a high-dimensional feature representation of the protein sequence data, the prediction module is used to process the high-dimensional feature representation, and predict the virion protein of the phage according to the processing result, and the annotation module is used to annotate and classify the predicted virion protein to obtain a detailed classification and annotation result of the virion protein. Thus, by combining the language model with the convolutional neural network, a virion protein prediction model is obtained, and the protein sequence is feature extracted through the protein language model to mine the potential high-dimensional feature information of the original protein sequence, thereby enabling the virion protein prediction model to still perform well in a high-similarity environment.

[0056] In the embodiment of the present invention, each module of the virus particle protein prediction model can be implemented by a corresponding hardware or software module. Each module can be an independent software or hardware module or integrated into one software or hardware module, which is not intended to limit the present invention.

[0057] Embodiment 2:

[0058] Figure 2 The implementation process of the virus particle protein prediction method provided in the second embodiment of the present invention is shown. For the convenience of explanation, only the part related to the embodiment of the present invention is shown, which is described in detail as follows:

[0059] In step S101, the whole genome sequence of the bacteriophage is obtained.

[0060] The embodiment of the present invention is applicable to the annotation and classification of phage viral particle protein (PVP). In the embodiment of the present invention, the whole genome sequence of the phage can be obtained by using methods such as phage adhesion method, centrifugation method, enzyme cutting method, PCR method (polymerase chain reaction) or high-throughput sequencing method, which is not specifically limited in this embodiment.

[0061] In step S102, the whole genome sequence of the bacteriophage is input into the virion protein prediction model to obtain detailed classification annotation results of the virion protein.

[0062] In the embodiment of the present invention, the prediction model of viral particle protein can refer to the description in the aforementioned embodiment 1, and will not be described in detail in this embodiment.

[0063] In an embodiment of the present invention, the whole genome sequence of the bacteriophage is obtained, and the whole genome sequence of the bacteriophage is input into the virion protein prediction model to obtain a detailed classification annotation result of the virion protein. The virion protein prediction model includes a data processing and translation module, a feature extraction module, a prediction module and an annotation module connected in sequence, wherein the data processing and translation module is used to translate the input whole genome sequence of the phage to obtain protein sequence data corresponding to the phage, the feature extraction module is used to extract features from the protein sequence data through the protein language model to obtain a high-dimensional feature representation of the protein sequence data, the prediction module is used to process the high-dimensional feature representation, and predict the virion protein of the phage according to the processing result, and the annotation module is used to annotate and classify the predicted virion protein to obtain a detailed classification annotation result of the virion protein. Thus, by combining the language model and the convolutional neural network, a virion protein prediction model is obtained, and the protein language model is used to extract features from the protein sequence to mine the potential high-dimensional feature information of the original protein sequence, so that the virion protein prediction model can still perform well in a high-similarity environment.

[0064] Embodiment three:

[0065] Figure 3 The structure of an electronic device provided by the third embodiment of the present invention is shown. For the convenience of description, only the part related to the embodiment of the present invention is shown.

[0066] The electronic device 3 of the embodiment of the present invention includes a processor 30, a memory 31, and a computer program 32 stored in the memory 31 and executable on the processor 30. When the processor 30 executes the computer program 32, the functions of each module in the above-mentioned virion protein prediction model embodiment are realized, for example Figure 1A Alternatively, when the processor 30 executes the computer program 32, the steps in the above-mentioned virion protein prediction method embodiment are implemented, for example Figure 2 Steps S101 to S102 are shown.

[0067] In an embodiment of the present invention, the virion protein prediction model includes a data processing and translation module, a feature extraction module, a prediction module and an annotation module connected in sequence, wherein the data processing and translation module is used to translate the input phage genome sequence to obtain protein sequence data corresponding to the phage, the feature extraction module is used to extract features from the protein sequence data through a protein language model to obtain a high-dimensional feature representation of the protein sequence data, the prediction module is used to process the high-dimensional feature representation, and predict the virion protein of the phage according to the processing result, and the annotation module is used to annotate and classify the predicted virion protein to obtain a detailed classification and annotation result of the virion protein. Thus, by combining the language model with the convolutional neural network, a virion protein prediction model is obtained, and the protein sequence is feature extracted through the protein language model to mine the potential high-dimensional feature information of the original protein sequence, thereby enabling the virion protein prediction model to still perform well in a high-similarity environment.

[0068] The electronic device of the embodiment of the present invention may be a personal computer, a server, etc. The steps implemented when the processor 30 in the electronic device 3 executes the computer program 32 to implement the virus particle protein prediction method can refer to the description of the aforementioned method embodiment, which will not be repeated here.

[0069] Embodiment 4:

[0070] In an embodiment of the present invention, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the functions of each module in the above-mentioned virion protein prediction model embodiment are realized, such as Figure 1A Alternatively, when the computer program is executed by a processor, the steps in the above-mentioned viral particle protein prediction method embodiment are implemented, for example Figure 2 Steps S101 to S102 are shown.

[0071] In an embodiment of the present invention, the virion protein prediction model includes a data processing and translation module, a feature extraction module, a prediction module and an annotation module connected in sequence, wherein the data processing and translation module is used to translate the input phage genome sequence to obtain protein sequence data corresponding to the phage, the feature extraction module is used to extract features from the protein sequence data through a protein language model to obtain a high-dimensional feature representation of the protein sequence data, the prediction module is used to process the high-dimensional feature representation, and predict the virion protein of the phage according to the processing result, and the annotation module is used to annotate and classify the predicted virion protein to obtain a detailed classification and annotation result of the virion protein. Thus, by combining the language model with the convolutional neural network, a virion protein prediction model is obtained, and the protein sequence is feature extracted through the protein language model to mine the potential high-dimensional feature information of the original protein sequence, thereby enabling the virion protein prediction model to still perform well in a high-similarity environment.

[0072] The computer-readable storage medium of the embodiment of the present invention may include any entity or device or recording medium capable of carrying computer program code, for example, ROM / RAM, magnetic disk, optical disk, flash memory and other memories.

[0073] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A virus particle protein prediction model, characterized in that: The virus particle protein prediction model comprises a data processing and translation module, a feature extraction module, a prediction module and an annotation module connected in sequence, wherein: The data processing and translation module is used to translate the input full genome sequence of the bacteriophage to obtain protein sequence data corresponding to the bacteriophage; The feature extraction module is used to extract features from the protein sequence data using a protein language model to obtain a high-dimensional feature representation of the protein sequence data; The prediction module is used to process the high-dimensional feature representation and predict the virion protein of the bacteriophage according to the processing result; The annotation module is used to annotate and classify the predicted virion proteins to obtain detailed classification annotation results of the virion proteins.

2. The virus particle protein prediction model according to claim 1, characterized in that The protein language model is the ProtT5 protein language model.

3. The virus particle protein prediction model according to claim 1, characterized in that The prediction module includes a neural network based on a CNN model, and the neural network based on the CNN model is used to process the high-dimensional feature representation.

4. The virus particle protein prediction model according to claim 3, characterized in that The neural network based on the CNN model includes a one-dimensional convolution layer, a batch normalization layer and a maximum pooling layer connected in sequence, wherein: The one-dimensional convolution layer is used to upsample the high-dimensional feature representation to obtain a multi-channel feature representation; The batch normalization layer is used to normalize the feature representation of the multi-channel to obtain a normalized feature representation; The maximum pooling layer is used to perform a pooling operation on the normalized feature representation.

5. The virus particle protein prediction model according to claim 1, characterized in that The loss function used in the training of the virus particle protein prediction model is the ASL asymmetric loss function.

6. The virus particle protein prediction model according to claim 5, characterized in that The virus particle protein prediction model is trained by smoothing the real labels.

7. The virus particle protein prediction model according to claim 6, characterized in that The ASL asymmetric loss function is: Among them, Log i represents the Logsoftmax function, x i represents the original prediction score of the i-th category, x j represents the original prediction score of the jth category, xspos represents the predicted probability of the virion protein prediction model for the target category, xsneg represents the complementary value of the predicted probability of the virion protein prediction model for the non-target category, and y i represents the true label of the i-th sample, asymmetricWeight i represents asymmetric weight, γ pos Represents the weight attenuation factor controlling the positive sample, γ neg Represents the weight attenuation factor controlling the negative sample, Loss i represents the loss value of the i-th sample, Loss represents the total loss value, and y ij Indicates whether the target category of the i-th sample is j. When y ij =1 means yes, y ij =0 means no. The formula used for smoothing the label is: Among them, y′ i represents the smoothed label, y i represents the true label of the i-th sample, ∈ represents the label smoothing factor, and C represents the number of categories.

8. A method for predicting viral particle proteins, characterized in that: The method comprises: Obtain the complete genome sequence of the phage; The whole genome sequence is input into the virion protein prediction model according to any one of claims 1 to 7 to obtain a detailed classification annotation result of the virion protein.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the functions of the virus particle protein prediction model according to any one of claims 1 to 7 are realized.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the function of the virus particle protein prediction model according to any one of claims 1 to 7 is realized.

Images