Virus classification method and system based on convolutional neural network
Through the virus classification method based on convolutional neural network, the problems of low efficiency and insufficient accuracy of virus classification in traditional tools are solved, and fast and accurate virus classification is achieved, and timely treatment decisions are supported.
Patent Information
- Application Number
- CN202510557588.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-29
- Publication Date
- 2025-07-25
AI Technical Summary
In viral classification, traditional bioinformatics tools have long analysis cycles, low efficiency and insufficient accuracy due to the huge nucleotide sequence data, and cannot perform virus classification in time and effectively, affecting treatment decisions.
Virus classification method based on convolutional neural network is adopted, by obtaining long characters of viral nucleotide sequences, performing token feature extraction and preprocessing, a virus detection classification model is constructed, and a sample data set is used for model training to obtain the optimal virus detection classification model for virus classification.
It improves the accuracy and efficiency of virus classification, can quickly and accurately judge the virus category, and provides a timely and reliable basis for treatment.
Smart Images
Figure CN120372360A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of virus classification, and in particular, to a virus classification method and system based on a convolutional neural network. Background Art
[0002] With the development of the concept of precision medicine, in the context of the emergence of human infection with novel coronavirus, the clinical diagnosis of infectious diseases has received increasing attention.
[0003] At present, traditional medicine mostly uses traditional bioinformatics tools, such as using whole-genome nucleotide sequence alignment to classify coronaviruses; however, in the process of using whole-genome nucleotide sequence alignment to classify viruses by traditional bioinformatics tools, due to the extremely large genomic sequence data of ribonucleotides, the workload required for the analysis of a large number of infectious virus samples is large. However, the accuracy of the detection rules formulated according to the experience of the detector is low and the detection error is large, that is, there are problems such as a long cycle, low classification efficiency, and low classification accuracy in converting, comparing, analyzing, and classifying such a large amount of data, and it is impossible to make a judgment in a timely, effective, and rapid manner, providing a timely and reliable basis for treatment. Summary of the Invention
[0004] The present invention provides a virus classification method and system based on a convolutional neural network to overcome the above technical problems.
[0005] To achieve the above object, the technical solution of the present invention is:
[0006] A virus classification method based on a convolutional neural network specifically includes the following steps:
[0007] S1: Obtain the long character of the virus nucleotide sequence of various novel coronavirus genomes;
[0008] S2: Perform token feature extraction preprocessing on the long character of the virus nucleotide sequence to obtain a nucleotide feature value weighted vector matrix;
[0009] S3: Construct a virus detection and classification model based on a convolutional neural network;
[0010] S4: Use the nucleotide feature value weighted vector matrix as feature data, and use the virus category corresponding to the novel coronavirus genome as label data to obtain a sample data set;
[0011] And perform model training on the virus detection and classification model according to the sample data set to obtain an optimal virus detection and classification model;
[0012] S5: Obtain the novel coronavirus genome to be classified and detected after token feature extraction preprocessing, and input it into the optimal virus detection classification model, thereby realizing the classification prediction of the novel coronavirus.
[0013] Furthermore, the virus detection classification model constructed in S3 includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, and a prediction output layer connected in sequence;
[0014] The input layer is used to transmit the nucleotide feature value weighted vector matrix to the first convolutional layer;
[0015] The first convolutional layer is used to perform feature convolution operations on the nucleotide feature value weighted vector matrix to obtain a first feature convolution matrix, and transmit it to the first pooling layer;
[0016] The first pooling layer is used to perform downsampling and dimensionality reduction processing on the first feature convolution matrix to obtain a first feature pooling matrix, and transmit it to the second convolutional layer;
[0017] The second convolutional layer is used to perform convolution operations on the first feature pooling matrix to obtain a second feature convolution matrix, and transmit it to the second pooling layer;
[0018] The second pooling layer is used to perform downsampling and dimensionality reduction processing on the second feature convolution matrix to obtain a second feature pooling matrix, and transmit it to the prediction output layer;
[0019] The prediction output layer is used to predict and output the virus category of the novel coronavirus according to the second feature pooling matrix.
[0020] Furthermore, S4 specifically includes the following steps:
[0021] S41: Use the nucleotide feature value weighted vector matrix as feature data and the virus category corresponding to the novel coronavirus genome as label data to obtain a sample data set;
[0022] Randomly divide the sample data set into a training set and a test set;
[0023] S42: Train the constructed virus detection classification model according to the training set to obtain the trained virus detection classification model:
[0024] The model training includes:
[0025] Transmit the nucleotide feature value weighted vector matrix to the first convolutional layer through the input layer;
[0026] Perform feature convolution operations on the nucleotide feature value weighted vector matrix through the first convolutional layer to obtain a first feature convolution matrix, and the acquisition formula of the first feature convolution matrix is
[0027] C1(x,y) = Sigmoid(I(x,y) * K(x,y)) (1)
[0028] Where: C1(x,y) represents the convolution result of the first convolutional layer; x,y represent the characteristic point coordinates of the nucleotide eigenvalue weighted vector matrix; Sigmoid() represents the activation function; I(x,y) represents the eigenvalue of the nucleotide eigenvalue weighted vector matrix; K(x,y) represents the trainable convolutional kernel; * represents the convolution operation;
[0029] The first feature convolutional matrix is downsampled and dimension-reduced by the first pooling layer to obtain the first feature pooling matrix, and the formula for obtaining the first feature pooling matrix is
[0030] P2(x,y) = Max_C1(x*s,y*s) (2)
[0031] Where: P2(x,y) represents the element of the first feature pooling matrix; s represents the pooling window size; x*s,y*s represent the sliding positions of the pooling window;
[0032] The first feature pooling matrix is convolved by the second convolutional layer to obtain the second feature convolutional matrix, and the formula for obtaining the second feature convolutional matrix is
[0033]
[0034] Where: represents the convolution result of the second convolutional layer; i represents the matrix serial number of the first feature pooling matrix and i = 1,2,...12; represents the input neuron from the first pooling layer; j represents the serial number of the input and output feature convolutional matrices and j = 1,2,...12; W ij represents the fully connected weight parameter matrix;
[0035] The second feature convolutional matrix is downsampled and dimension-reduced by the second pooling layer to obtain the second feature pooling matrix;
[0036] The second feature pooling matrix is unfolded into a one-dimensional vector by the prediction output layer, and based on the second feature pooling matrix after the one-dimensional vector unfolding operation, the virus category of the novel coronavirus is predicted and output; and the expression for predicting and outputting the virus category of the novel coronavirus is
[0037]
[0038] Where: represents the output neuron of the prediction output layer; represents the input neuron from the second pooling layer; W ij represents the fully connected weight parameter matrix;
[0039] S43: Based on the mean square error function as the model loss function, and evaluate the trained virus detection and classification model according to the test set to determine whether the output of the trained virus detection and classification model converges;
[0040] If so, the trained virus detection and classification model at this time is the optimal virus detection and classification model;
[0041] Otherwise, implement error backpropagation based on the gradient descent method, adaptively adjust the convolution kernel and fully connected weight parameters of the trained virus detection and classification model with the goal of seeking the minimum error, and repeat step S42.
[0042] Further, the specific steps of S2 are as follows:
[0043] S21: Based on the K-mer segmentation algorithm, split the long characters of the virus nucleotide sequence into overlapping character fragments with a continuous character length feature of k, and use each overlapping character fragment as a nucleotide feature sequence;
[0044] S22: Calculate the feature occurrence frequency of each nucleotide feature sequence respectively, numerically process the nucleotide feature sequence according to the number of times the feature occurrence frequency appears, and perform normalization processing on the numerically processed virus frequency feature sequence to obtain the virus frequency eigenvalue sequence;
[0045] S23: Perform IF-IDF transformation on the virus frequency eigenvalue sequence to obtain the nucleotide feature value weighted vector matrix of the novel coronavirus genome.
[0046] Further, the method for obtaining the nucleotide feature value weighted vector matrix of the novel coronavirus genome in S23 specifically includes the following steps:
[0047] S231: Statistically obtain the frequency TF of each virus frequency eigenvalue sequence appearing in the long characters of a single virus nucleotide sequence, and calculate the inverse frequency IDF of each virus frequency eigenvalue sequence in the dataset of the entire novel coronavirus genome;
[0048] And the formula for obtaining the inverse frequency IDF is
[0049]
[0050] In the formula: N represents the total number of long characters of virus nucleotide sequences in the dataset of the entire novel coronavirus genome; n t represents the number of long characters of virus nucleotide sequences containing the nucleotide feature sequence t;
[0051] S232: Multiply the frequency TF by the inverse frequency IDF to obtain the IF-IDF weighted feature vector, i.e., the weighted vector matrix of nucleotide feature values of the novel coronavirus genome.
[0052] A virus classification system based on a convolutional neural network, comprising a long character acquisition module for virus nucleotide sequences, a feature extraction and preprocessing module, a dataset acquisition module, a virus detection and classification model construction module, and a training and deployment module;
[0053] The long character acquisition module for virus nucleotide sequences is used to acquire the long characters of virus nucleotide sequences of various novel coronavirus genomes;
[0054] The feature extraction and preprocessing module is used to perform token feature extraction and preprocessing on the long characters of virus nucleotide sequences to obtain a weighted vector matrix of nucleotide feature values;
[0055] The dataset acquisition module is used to take the weighted vector matrix of nucleotide feature values as feature data and the virus category corresponding to the novel coronavirus genome as label data to obtain a sample dataset;
[0056] The virus detection and classification model construction module is used to construct a virus detection and classification model based on a convolutional neural network;
[0057] The training and deployment module is used to electrically retrieve the sample dataset and the constructed virus detection and classification model based on a convolutional neural network, and perform model training on the virus detection and classification model according to the sample dataset to obtain an optimal virus detection and classification model, and then realize the classification prediction of the novel coronavirus through the optimal virus detection and classification model.
[0058] Beneficial effects: The present invention provides a virus classification method and system based on a convolutional neural network. By performing token feature extraction preprocessing on the long characters of the virus nucleotide sequence, a weighted vector matrix of nucleotide feature values is obtained, solving the problems of the extremely large genomic sequence data of ribonucleotides, the low accuracy of detection rules formulated according to the experience of testers, and the large detection error. It greatly improves the accuracy and efficiency of feature extraction of ribonucleotide genomic sequence data, and through the constructed virus detection and classification model based on a convolutional neural network; using the weighted vector matrix of nucleotide feature values as feature data and the virus category corresponding to the novel coronavirus genome as label data, the obtained sample data set is used to train the model to obtain the optimal virus detection and classification model; obtaining the novel coronavirus genome to be classified and detected after token feature extraction preprocessing and inputting it into the optimal virus detection and classification model to realize the classification prediction of the novel coronavirus, solving the problems of long cycle, large classification efficiency, and low classification accuracy due to the corresponding conversion, comparison, analysis, and classification of huge data, and being able to quickly and effectively make a judgment on the virus category in a timely manner, providing a timely and reliable basis for subsequent treatment. Brief Description of the Drawings
[0059] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.
[0060] Figure 1 It is a flowchart of the virus classification method based on a convolutional neural network of the present invention.
[0061] Figure 2 It is a schematic diagram of the weighted vector matrix of nucleotide feature values obtained in this embodiment;
[0062] Figure 3 It is a schematic diagram of the result of virus classification by the optimal virus detection and classification model in this embodiment. Detailed Embodiments
[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.
[0064] This embodiment provides a virus classification method and system based on a convolutional neural network, as Figure 1 shown, which specifically includes the following steps:
[0065] S1: Obtain the long character of the virus nucleotide sequence of various novel coronavirus genomes;
[0066] Specifically, for the classification and analysis of novel coronavirus in this embodiment, various novel coronavirus genome data are required, and the novel coronavirus genome data uses existing public genome fasta type data;
[0067] S2: Perform token feature extraction preprocessing on the long character of the virus nucleotide sequence to obtain a nucleotide feature value weighted vector matrix, which specifically includes the following steps:
[0068] S21: Based on the K-mer segmentation algorithm, segment the long character of the virus nucleotide sequence into overlapping character fragments with a continuous character length feature of k, and use each overlapping character fragment as a nucleotide feature sequence;
[0069] Specifically, in this embodiment, k takes 2-5 consecutive character features; for example, when k = 3, for the long character of the virus nucleotide sequence "ATGCGA", four overlapping character fragments can be obtained: "ATGCGA" → ["ATG", "TGC", "GCG", "CGA"];
[0070] S22: Calculate the feature occurrence frequency of each nucleotide feature sequence respectively, perform numerical processing on the nucleotide feature sequence according to the number of times of the feature occurrence frequency, and perform normalization processing on the numerically processed virus frequency feature sequence to obtain a virus frequency feature value sequence;
[0071] Among them, the method for numerically processing the nucleotide feature sequence in this embodiment is a publicly known technology, that is, obtaining a set of each nucleotide feature sequence according to the number of times of the feature occurrence frequency, and performing numerical processing on each nucleotide feature sequence in the sequence set to obtain a numerical signal corresponding to each nucleotide feature sequence, which will not be elaborated here;
[0072] S23: Perform an IF-IDF transform on the virus frequency feature value sequence to obtain a nucleotide feature value weighted vector matrix of the novel coronavirus genome;
[0073] In a specific embodiment, the method for obtaining the nucleotide feature value weighted vector matrix of the novel coronavirus genome in S23 specifically includes the following steps:
[0074] S231: Statistically obtain the frequency TF of each viral frequency eigenvalue sequence occurring in the long character string of a single viral nucleotide sequence, and calculate the inverse document frequency IDF of each viral frequency eigenvalue sequence in the dataset of the entire genome of the novel coronavirus;
[0075] And the formula for obtaining the inverse document frequency IDF is
[0076]
[0077] In the formula: N represents the total number of long character strings of viral nucleotide sequences in the dataset of the entire genome of the novel coronavirus; n t represents the number of long character strings of viral nucleotide sequences containing the nucleotide feature sequence t;
[0078] S232: Multiply the frequency TF by the inverse document frequency IDF to obtain the IF-IDF weighted feature vector, that is, the nucleotide eigenvalue weighted vector matrix of the novel coronavirus genome;
[0079] S3: Construct a virus detection and classification model based on a convolutional neural network;
[0080] Specifically, the constructed virus detection and classification model includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, and a prediction output layer connected in sequence;
[0081] The input layer is used to transmit the nucleotide eigenvalue weighted vector matrix to the first convolutional layer;
[0082] The first convolutional layer is used to perform feature convolution operations on the nucleotide eigenvalue weighted vector matrix to obtain a first feature convolution matrix, and transmit it to the first pooling layer;
[0083] The first pooling layer is used to perform downsampling and dimensionality reduction processing on the first feature convolution matrix to obtain a first feature pooling matrix, and transmit it to the second convolutional layer;
[0084] The second convolutional layer is used to perform convolution operations on the first feature pooling matrix to obtain a second feature convolution matrix, and transmit it to the second pooling layer;
[0085] The second pooling layer is used to perform downsampling and dimensionality reduction processing on the second feature convolution matrix to obtain a second feature pooling matrix, and transmit it to the prediction output layer;
[0086] The prediction output layer is used to predict and output the virus category of the novel coronavirus according to the second feature pooling matrix;
[0087] S4: Use the weighted vector matrix of nucleotide eigenvalue as feature data, and use the virus category corresponding to the novel coronavirus genome as label data to obtain a sample data set. Then, train the virus detection and classification model based on the sample data set to obtain the optimal virus detection and classification model;
[0088] Specifically, it includes the following steps:
[0089] S41: Use the weighted vector matrix of nucleotide eigenvalue as feature data, and use the virus category corresponding to the novel coronavirus genome as label data to obtain a sample data set;
[0090] Randomly divide the sample data set into a training set and a test set;
[0091] S42: Train the constructed virus detection and classification model based on the training set to obtain the trained virus detection and classification model:
[0092] The model training includes:
[0093] Transmit the weighted vector matrix of nucleotide eigenvalue to the first convolutional layer through the input layer;
[0094] Perform feature convolution operation on the weighted vector matrix of nucleotide eigenvalue through the first convolutional layer to obtain the first feature convolution matrix, and the acquisition formula of the first feature convolution matrix is
[0095] C1(x,y) = Sigmoid(I(x,y) * K(x,y)) (2)
[0096] In the formula: C1(x,y) represents the convolution result of the first convolutional layer; x,y represent the coordinate of the feature point of the weighted vector matrix of nucleotide eigenvalue; Sigmoid() represents the activation function; I(x,y) represents the eigenvalue of the weighted vector matrix of nucleotide eigenvalue; K(x,y) represents the trainable convolution kernel; * represents the convolution operation;
[0097] In this embodiment, the input feature of the first convolutional layer is set to 784, arranged as a 28×28 weighted vector matrix of nucleotide eigenvalue, and convolution operations are respectively performed with 12 5×5 convolution kernels. After being processed by the activation function Sigmoid(), 12 24×24 convolution result matrices, that is, the first feature convolution matrix, can be obtained respectively;
[0098] Perform downsampling and dimensionality reduction processing on the first feature convolution matrix through the first pooling layer to obtain the first feature pooling matrix, and the acquisition formula of the first feature pooling matrix is
[0099] P2(x,y) = Max_C1(x*s,y*s) (3)
[0100] Where: P2(x, y) represents the element of the first feature pooling matrix; s represents the pooling window size; x*s, y*s represent the sliding positions of the pooling window;
[0101] In this embodiment, the second convolutional layer uses the max pooling method to implement downsampling and dimensionality reduction processing, reducing the 12 convolutional matrices of 24×24 output by the first convolutional layer to 12 pooling matrices of 12×12, that is, the first feature pooling matrix;
[0102] Through the second convolutional layer, a convolution operation is performed on the first feature pooling matrix to obtain the second feature convolution matrix, and the acquisition formula of the second feature convolution matrix is
[0103]
[0104] Where: represents the convolution result of the second convolutional layer; i represents the matrix serial number of the first feature pooling matrix and i = 1, 2,... 12; represents the input neuron from the first pooling layer; j represents the serial number of the input and output feature convolution matrices and j = 1, 2,... 12; W ij represents the fully connected weight parameter matrix;
[0105] In this embodiment, the second convolutional layer is a fully connected convolutional layer, and its input is 12 pooling matrices of 12×12. Each group uses a 5×5 convolutional kernel for convolution operation to obtain 12 groups of convolution intermediate results. Each group obtains 12 convolution intermediate matrices of 8×8. Then, the convolution intermediate matrices at the corresponding positions in each group are accumulated to obtain 12 matrices of 8×8. Finally, after being processed by the activation function Sigmoid(), 12 convolution result matrices of 8×8, that is, the second feature convolution matrix, are respectively obtained;
[0106] Through the second pooling layer, downsampling and dimensionality reduction processing are performed on the second feature convolution matrix to obtain the second feature pooling matrix; in this embodiment, the second pooling layer also uses the max pooling method to implement downsampling and dimensionality reduction processing, reducing the 12 convolution matrices of 8×8 output by the second convolutional layer to 12 pooling matrices of 4×4, that is, the second feature pooling matrix;
[0107] Through the prediction output layer, a one-dimensional vector expansion operation is performed on the second feature pooling matrix, and based on the second feature pooling matrix after the one-dimensional vector expansion operation, the virus category of the novel coronavirus is predicted and output; and the expression for predicting and outputting the virus category of the novel coronavirus is
[0108]
[0109] Where: represents the output neuron of the prediction output layer; Represents the input neurons from the second pooling layer; W ij Represents the fully connected weight parameter matrix;
[0110] In this embodiment, the 12 4×4 pooling matrices input by the second pooling layer are expanded into a one-dimensional vector as the input. There are 192 input neurons and 6 output neurons. Each output neuron forms a fully connected relationship with the 192 input neurons, W ij Represents the fully connected weight from the i-th input neuron to the j-th output neuron; for example, the 6 output neurons respectively correspond to 6 viruses: Omicron, Gamma, Alpha, Beta, Delta, and Lambda;
[0111] S43: Based on the mean square error function as the model loss function, and according to the test set, evaluate the trained virus detection and classification model to determine whether the output of the trained virus detection and classification model converges;
[0112] If so, the trained virus detection and classification model at this time is the optimal virus detection and classification model;
[0113] Otherwise, based on the gradient descent method, implement error backpropagation, adaptively adjust the convolution kernels and fully connected weight parameters of the trained virus detection and classification model with the goal of seeking the minimum error, and repeat step S42;
[0114] S5: Obtain the genome of the novel coronavirus to be classified and detected after token feature extraction preprocessing, and input it into the optimal virus detection and classification model to realize the classification prediction of the novel coronavirus.
[0115] This embodiment also includes a virus classification system based on a convolutional neural network, including a virus nucleotide sequence long character acquisition module, a feature extraction preprocessing module, a data set acquisition module, a virus detection and classification model construction module, and a training and deployment module;
[0116] The virus nucleotide sequence long character acquisition module is used to acquire the virus nucleotide sequence long characters of various novel coronavirus genomes;
[0117] The feature extraction preprocessing module is used to perform token feature extraction preprocessing on the virus nucleotide sequence long characters to obtain a nucleotide feature value weighted vector matrix;
[0118] The data set acquisition module is used to use the nucleotide feature value weighted vector matrix as the feature data and the virus category corresponding to the novel coronavirus genome as the label data to obtain a sample data set;
[0119] The virus detection and classification model construction module is used to construct a virus detection and classification model based on a convolutional neural network;
[0120] The training and deployment module is used to retrieve the sample data set and the constructed virus detection and classification model based on the convolutional neural network, and train the virus detection and classification model according to the sample data set to obtain the optimal virus detection and classification model, and then realize the classification prediction of the novel coronavirus through the optimal virus detection and classification model.
[0121] Based on step S5 of the method described in this embodiment, that is, the convolutional neural network for classifying the novel coronavirus has been constructed. When a human-infected novel coronavirus RNA sequence preprocessed through steps S1 to S2 is input, the membership degrees of the input virus belonging to various novel coronaviruses can be obtained through forward propagation calculation. The one with the highest membership degree is the classification result. For example, Figures 2 to 3 as shown, Figure 2 It is a feature matrix formed by arranging 784 RNA data into 28 rows and 28 columns and presented in different colors. Figure 3 Based on Figure 2 The feature matrix obtained from 784 data in is used as the input, and the classification result generated by the optimal virus detection and classification model is obtained, thereby realizing the classification prediction of the novel coronavirus.
[0122] Beneficial effects: In this embodiment, by performing token feature extraction preprocessing on the long characters of the virus nucleotide sequence, a nucleotide feature value weighted vector matrix is obtained, which solves the problems of the extremely large genomic sequence data of ribonucleotides, the low accuracy and large detection error of the detection rules formulated according to the experience of the detector, greatly improves the feature extraction accuracy and efficiency of the genomic sequence data of ribonucleotides, and through the constructed virus detection and classification model based on the convolutional neural network; using the nucleotide feature value weighted vector matrix as the feature data and the virus category corresponding to the novel coronavirus genome as the label data, training the model with the obtained sample data set to obtain the optimal virus detection and classification model; obtaining the novel coronavirus genome to be classified and detected after token feature extraction preprocessing, and inputting it into the optimal virus detection and classification model to realize the classification prediction of the novel coronavirus, which solves the problems of long cycle, large classification efficiency and low classification accuracy in the corresponding conversion, comparison, analysis and classification of large amounts of data, and can quickly and effectively make a judgment on the virus category in a timely manner, providing a timely and reliable basis for subsequent treatment.
[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present invention.
Claims
1. A virus classification method based on a convolutional neural network, characterized in that, Specifically, it includes the following steps: S1: Obtain the long character of the viral nucleotide sequence of various novel coronavirus genomes; S2: Perform token feature extraction preprocessing on the long character of the viral nucleotide sequence to obtain a nucleotide feature value weighted vector matrix; S3: Construct a viral detection and classification model based on a convolutional neural network; S4: Use the nucleotide feature value weighted vector matrix as feature data and the viral category corresponding to the novel coronavirus genome as label data to obtain a sample data set; And perform model training on the viral detection and classification model according to the sample data set to obtain an optimal viral detection and classification model; S5: Obtain the novel coronavirus genome to be classified and detected after token feature extraction preprocessing, and input it into the optimal viral detection and classification model, thereby realizing the classification prediction of the novel coronavirus.
2. The virus classification method based on a convolutional neural network according to claim 1, wherein The viral detection and classification model constructed in S3 includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, and a prediction output layer connected in sequence; The input layer is used to transmit the nucleotide feature value weighted vector matrix to the first convolutional layer; The first convolutional layer is used to perform feature convolution operation on the nucleotide feature value weighted vector matrix to obtain a first feature convolution matrix, and transmit it to the first pooling layer; The first pooling layer is used to perform downsampling and dimensionality reduction processing on the first feature convolution matrix to obtain a first feature pooling matrix, and transmit it to the second convolutional layer; The second convolutional layer is used to perform convolution operation on the first feature pooling matrix to obtain a second feature convolution matrix, and transmit it to the second pooling layer; The second pooling layer is used to perform downsampling and dimensionality reduction processing on the second feature convolution matrix to obtain a second feature pooling matrix, and transmit it to the prediction output layer; The prediction output layer is used to predict and output the viral category of the novel coronavirus according to the second feature pooling matrix.
3. A virus classification method based on a convolutional neural network according to claim 1, characterized in that, The specific steps of S4 are as follows: S41: Use the nucleotide feature value weighted vector matrix as feature data and the viral category corresponding to the novel coronavirus genome as label data to obtain a sample data set; Randomly divide the sample data set into a training set and a test set; S42: Perform model training on the constructed viral detection and classification model according to the training set to obtain a trained viral detection and classification model: The model training includes: Transmit the nucleotide feature value weighted vector matrix to the first convolutional layer through the input layer; Perform feature convolution operation on the nucleotide feature value weighted vector matrix through the first convolutional layer to obtain a first feature convolution matrix, and the acquisition formula of the first feature convolution matrix is C1(x,y) = Sigmoid(I(x,y)*K(x,y)) (1) In the formula: C1(x,y) represents the convolution result of the first convolutional layer; x,y represent the feature point coordinates of the nucleotide feature value weighted vector matrix; Sigmoid() represents the activation function; I(x,y) represents the eigenvalue of the nucleotide feature value weighted vector matrix; K(x,y) represents the trainable convolution kernel; * represents the convolution operation; The first pooling layer downsamples and reduces the dimension of the first feature convolution matrix to obtain the first feature pooling matrix, and the acquisition formula of the first feature pooling matrix is O2(x,y) = Max_C1(x*s,y*s) (2) Where: P2(x,y) represents the element of the first feature pooling matrix; s represents the pooling window size; x*s,y*s represent the sliding positions of the pooling window The second convolution layer performs a convolution operation on the first feature pooling matrix to obtain the second feature convolution matrix, and the acquisition formula of the second feature convolution matrix is In the formula: represents the convolution result of the second convolutional layer; i represents the matrix serial number of the first feature pooling matrix and i = 1, 2, …, 12; represents the input neuron from the first pooling layer; j represents the serial number of the input and output feature convolution matrices and j = 1, 2, …, 12; W ij represents the fully connected weight parameter matrix; The second pooling layer downsamples and reduces the dimension of the second feature convolution matrix to obtain the second feature pooling matrix The prediction output layer performs a one-dimensional vector expansion operation on the second feature pooling matrix, and based on the second feature pooling matrix after the one-dimensional vector expansion operation, predicts and outputs the virus category of the novel coronavirus; and the expression for predicting and outputting the virus category of the novel coronavirus is Wherein: represents the output neuron of the prediction output layer; represents the input neuron from the second pooling layer; W ij represents the fully connected weight parameter matrix; S43: Based on the mean square error function as the model loss function, and according to the test set, evaluate the trained virus detection and classification model to determine whether the output of the trained virus detection and classification model converges If so, the trained virus detection and classification model at this time is the optimal virus detection and classification model Otherwise, based on the gradient descent method, implement error backpropagation, adaptively adjust the convolution kernel and fully connected weight parameters of the trained virus detection and classification model with the goal of seeking the minimum error, and repeat step S42 4. A virus classification method based on a convolutional neural network according to claim 1, characterized in that The specific steps of S2 are as follows S21: Based on the K-mer segmentation algorithm, segment the long characters of the virus nucleotide sequence into overlapping character fragments with a continuous character length feature of k, and use each overlapping character fragment as a nucleotide feature sequence S22: Calculate the feature occurrence frequencies of each nucleotide feature sequence respectively, numerically process the nucleotide feature sequences according to the number of times the feature occurrence frequencies appear, and normalize the numerically processed virus frequency feature sequences to obtain the virus frequency eigenvalue sequence S23: Perform an IF-IDF transformation on the virus frequency eigenvalue sequence to obtain the nucleotide eigenvalue weighted vector matrix of the novel coronavirus genome 5. A virus classification method based on a convolutional neural network according to claim 4, characterized in that, The method for obtaining the nucleotide eigenvalue weighted vector matrix of the novel coronavirus genome in S23 specifically includes the following steps S231: Statistically obtain the frequency TF of each virus frequency eigenvalue sequence appearing in the long characters of a single virus nucleotide sequence, and calculate the inverse frequency IDF of each virus frequency eigenvalue sequence in the dataset of the entire novel coronavirus genome And the acquisition formula of the inverse frequency IDF is Where: N represents the total number of long characters of the viral nucleotide sequences in the entire dataset of the novel coronavirus genome; n t represents the number of long characters of the viral nucleotide sequence containing the nucleotide characteristic sequence t; S232: Multiply the frequency TF by the inverse frequency IDF to obtain the IF-IDF weighted feature vector, that is, the nucleotide eigenvalue weighted vector matrix of the novel coronavirus genome 6. A system for a virus classification method based on a convolutional neural network according to any one of claims 1 to 5, characterized in that, It includes a long character acquisition module for virus nucleotide sequences, a feature extraction and preprocessing module, a dataset acquisition module, a virus detection and classification model construction module, and a training and deployment module The long character acquisition module for virus nucleotide sequences is used to obtain the long characters of the virus nucleotide sequences of various novel coronavirus genomes The feature extraction preprocessing module is used to perform token feature extraction preprocessing on the long characters of the virus nucleotide sequence to obtain a nucleotide feature value weighted vector matrix; The dataset acquisition module is used to use the nucleotide feature value weighted vector matrix as feature data and the virus category corresponding to the novel coronavirus genome as label data to obtain a sample dataset; The virus detection and classification model construction module is used to construct a virus detection and classification model based on a convolutional neural network; The training and deployment module is used to electrically retrieve the sample dataset and the constructed virus detection and classification model based on a convolutional neural network, and perform model training on the virus detection and classification model according to the sample dataset to obtain an optimal virus detection and classification model, and then realize the classification prediction of the novel coronavirus through the optimal virus detection and classification model.