Deep learning-based early warning method for new variants of respiratory viruses

Abstract

The invention relates to the technical field of virus data processing, in particular to a deep learning-based early warning method for new variants of respiratory viruses, which comprises the following steps of: extracting variation features by taking three nodes of the respiratory viruses, which are most easily varied, as the variation features; the method comprises the following steps of: quantifying variation activeness of a virus on a nucleotide level and functional change of a protein level, evaluating the risk of generating a new variant caused by gene recombination, and determining the variation of the virus on the protein level, namely variation generated by RNA instability, variation generated by mistake when nucleotide is translated into amino acid and variation generated by gene fragment recombination. The increase of the recombination event means that the probability of increasing the variants is increased; the propagation characteristics are added into deep learning, and the model can automatically learn a variation-propagation association mode of a novel variant, so that the epidemic outbreak trend of the respiratory virus variant is predicted, and early warning is realized; and various data are updated in real time, and rapid early warning can be realized according to the latest prediction effect.

Description

Technical Field

[0001] The present invention relates to the technical field of virus data processing, and in particular to an early warning method for new variants of respiratory viruses based on deep learning. Background Art

[0002] The frequent emergence of new variants of respiratory viruses poses a significant challenge to global public health. Providing timely and accurate early warning of new variant outbreaks is crucial for implementing effective prevention and control measures and mitigating the impact of epidemics. However, existing technologies have numerous shortcomings in addressing this issue. Traditional surveillance methods still rely primarily on laboratory testing and case reporting, which exhibit significant lags. For example, in the early stages of a new viral variant's emergence, limited detection methods may prevent timely detection, allowing the epidemic to spread undetected. Furthermore, prediction methods based on search engine data or simple models struggle to fully capture the complex characteristics and transmission patterns of viral mutations. While some influenza prediction models based on search engine data can leverage big data, they are passive and still exhibit lags. Therefore, a deep learning-based early warning method for new respiratory virus variants is needed that provides a degree of foresight and reduces the passive nature of these methods. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide an early warning method for new variants of respiratory viruses based on deep learning that has a certain degree of foresight and reduces passivity.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: A deep learning-based early warning method for new respiratory virus variants, including: Data collection and preprocessing: collecting full genome sequence data of respiratory viruses, the number of new cases of respiratory viruses per day, and the scope of their spread; Extract variation features and calculate nucleotide variation index , = ; Translate the nucleotide sequence into an amino acid sequence and calculate the amino acid variation coefficient , =α× ; Calculate the risk value of gene segment recombination , =β×(1- ); define the mutation propagation acceleration factor =γ×( + + ); in, is the length of the viral sequence, To take the maximum variation frequency for each site, n is the number of nucleotide types, is the number of occurrences of the jth nucleotide at the i-th site; represents the total number of occurrences of all nucleotide types at the i-th site in the same viral sequence population; α is the adjustment coefficient, is the length of the amino acid sequence, for The number of sites replaced; β is the adjustment coefficient, For the same number of bases, is the length of the recombinant fragment; Extract transmission characteristics and calculate the transmission speed based on the number of new cases per day and the scope of transmission and range extension rate , constructing a range expansion index = × ; Build a deep learning model, using a CNN and LSTM architecture to 、 、 、 、 、 As input, whether a new variant epidemic outbreak occurs is used as the output label; the constructed model inputs new data and outputs a predicted probability value.

[0005] Preferably, the propagation speed = ,in is the number of new cases on day t, is the number of new cases on day t-1.

[0006] Preferably, the range expansion rate = ,in is the number of areas with respiratory virus transmission on day t, is the number of areas with respiratory virus transmission on day t-1.

[0007] Preferably, a combined architecture of CNN and LSTM is adopted, where CNN extracts local patterns of virus sequence variation characteristics, and LSTM processes the dynamic change trend of the time series of propagation data.

[0008] Preferably, when building a deep learning model, the cross entropy loss function L is used to quantify the difference between the model's predicted probability and the true label, L = - ; Where N is the number of samples, is the true label of the k-th sample; is the predicted probability of the model for k samples, ranging from 0 to 1.

[0009] Preferably, when building a deep learning model, model optimization is also included, using L1 and L2 regularization and Dropout technology to prevent overfitting, and the optimized loss function =L+ + ; in, 、 is the regularization parameter, are model parameters, is the number of parameters, which improves the generalization ability of the model.

[0010] Preferably, the cross entropy loss function L is adjusted through the back propagation algorithm to minimize the loss function and improve the prediction accuracy.

[0011] Preferably, the constructed model inputs new data and outputs a predicted probability value, and sets a warning threshold T based on historical data and / or expert experience. When the predicted probability is greater than T, a warning signal is issued.

[0012] Preferably, α is a weight coefficient obtained based on regression analysis of historical data, which is used to quantify the impact of amino acid mutations on the ability of the virus to spread.

[0013] Preferably, β is an adjustment coefficient set according to different virus types.

[0014] The beneficial effects of the present invention are as follows: through the extraction of variation features, the three nodes where respiratory viruses are most likely to mutate are extracted as variation features, namely, variations caused by RNA instability, variations caused by errors when nucleotides are translated into amino acids, and variations caused by gene segment recombination. Quantify the mutation activity of the virus at the nucleotide level, It is a functional change at the protein level. The greater the probability of change, the more likely the protein produced will increase its transmission ability. Assess the risk of new variants caused by gene recombination. The increase in recombination events means an increase in the probability of additional variants. By adding transmission characteristics to deep learning, the model can automatically learn the mutation-transmission association pattern of new variants, and then predict the epidemic outbreak trend of respiratory virus variants, thereby achieving early warning. Various data are updated in real time, so the prediction model can quickly warn and respond based on the latest prediction results, reducing passivity. DETAILED DESCRIPTION

[0015] In order to explain the technical content, achieved objectives and effects of the present invention in detail, the following describes them in conjunction with the implementation methods.

[0016] A deep learning-based early warning method for new respiratory virus variants, including: Data collection and preprocessing: collecting full genome sequence data of respiratory viruses, the number of new cases of respiratory viruses per day, and the scope of their spread; Extract variation features and calculate nucleotide variation index , = ; Translate the nucleotide sequence into an amino acid sequence and calculate the amino acid variation coefficient , =α× ; Calculate the risk value of gene segment recombination , =β×(1- ); define the mutation propagation acceleration factor =γ×( + + ); in, is the length of the viral sequence, To take the maximum variation frequency for each site, n is the number of nucleotide types, is the number of occurrences of the jth nucleotide at the i-th site; represents the total number of occurrences of all nucleotide types at the i-th site in the same viral sequence population; α is the adjustment coefficient, is the length of the amino acid sequence, for The number of sites replaced; β is the adjustment coefficient, For the same number of bases, is the length of the recombinant fragment; Extract transmission characteristics and calculate the transmission speed based on the number of new cases per day and the scope of transmission and range extension rate , constructing a range expansion index = × ; Build a deep learning model, using a CNN and LSTM architecture to 、 、 、 、 、 As input, whether a new variant epidemic outbreak occurs is used as the output label; the constructed model inputs new data and outputs a predicted probability value.

[0017] From the above description, we can see that through the extraction of mutation features, the three nodes where respiratory viruses are most likely to mutate are extracted as mutation features, namely, mutations caused by the instability of RNA itself, mutations caused by errors when nucleotides are translated into amino acids, and mutations caused by gene segment recombination. Quantify the mutation activity of the virus at the nucleotide level, It is a functional change at the protein level. The greater the probability of change, the more likely the protein produced will increase its transmission ability. Assess the risk of new variants caused by gene recombination. The increase in recombination events means an increase in the probability of additional variants. By adding transmission characteristics to deep learning, the model can automatically learn the mutation-transmission association pattern of new variants, and then predict the epidemic outbreak trend of respiratory virus variants, thereby achieving early warning. Various data are updated in real time, so the prediction model can quickly warn and respond based on the latest prediction results, reducing passivity.

[0018] Furthermore, the propagation speed = ,in is the number of new cases on day t, is the number of new cases on day t-1.

[0019] Furthermore, the range expansion rate = ,in is the number of areas with respiratory virus transmission on day t, is the number of areas with respiratory virus transmission on day t-1.

[0020] Furthermore, a combined architecture of CNN and LSTM is adopted, where CNN extracts local patterns of virus sequence mutation characteristics, and LSTM processes the dynamic change trends of the time series of propagation data.

[0021] Furthermore, when building a deep learning model, the cross entropy loss function L is used to quantify the difference between the model's predicted probability and the true label, L=- ; Where N is the number of samples, is the true label of the k-th sample; is the predicted probability of the model for k samples, ranging from 0 to 1.

[0022] From the above description, we can see that by minimizing L, the model output probability is forced to be close to or focused on the true label, that is, the prediction probability of positive samples (epidemic outbreak) approaches 1, and the prediction probability of negative samples approaches 0.

[0023] Furthermore, when building a deep learning model, model optimization is also included, using L1 and L2 regularization and Dropout technology to prevent overfitting, and the optimized loss function =L+ + ; in, 、 is the regularization parameter, are model parameters, is the number of parameters, which improves the generalization ability of the model.

[0024] Furthermore, the cross entropy loss function L is adjusted through the back propagation algorithm to minimize the loss function and improve the prediction accuracy.

[0025] Furthermore, the constructed model inputs new data and outputs a predicted probability value. The warning threshold T is set based on historical data and / or expert experience, and a warning signal is issued when the predicted probability is greater than T.

[0026] Furthermore, α is a weight coefficient obtained from the regression analysis of historical data, which is used to quantify the impact of amino acid mutations on the virus's transmissibility.

[0027] Furthermore, β is an adjustment coefficient set according to different virus types.

[0028] Example 1 A deep learning-based early warning method for new respiratory virus variants, including: Data Collection: Whole-genome sequence data: Whole-genome sequence data for respiratory viruses can be collected from sources such as the China National Genome Science Data Center, GISAID (Global Initiative on Sharing All Influenza Data), and NCBI (National Center for Biotechnology Information). This data contains the genetic information of the virus and serves as the foundation for analyzing viral mutations. For each viral sequence, metadata such as the collection time, location, and host are recorded. Transmission data: Collect transmission data related to respiratory viruses, including number of cases, incidence, geographic spread, etc. This data can be obtained from official sources such as national and municipal centers for disease control and prevention, and the World Health Organization. Preprocessing: Collected viral sequence data is compared using sequence alignment algorithms, such as BLAST (Basic Local Alignment Search Tool), to remove duplicate and low-quality sequences. For transmission data, data cleaning is performed, missing values ​​are filled, and outliers are addressed. For example, if an unusually high number of cases suddenly appears in a particular region, corrections can be made by verifying the data source or using statistical methods. If the data has already been processed, further processing is unnecessary.

[0029] Variation feature extraction: Nucleotide variation comprehensive index: Calculate the nucleotide variation frequency at each site in the viral sequence , = , and establish a comprehensive nucleotide variation index , = ; in, is the length of the viral sequence, To obtain the maximum mutation frequency for each site, n is the number of nucleotide types (RNA viruses commonly have four types (AGCU, adenine (A), guanine (G), cytosine (C) and uracil (U) four bases)), is the number of occurrences of the jth nucleotide at the i-th site; represents the total number of occurrences of all nucleotide types at the i-th site in the same viral sequence population; Amino acid variation driving coefficient: Translate the nucleotide sequence into the amino acid sequence and calculate the amino acid substitution rate , = ; is the length of the amino acid sequence, for The number of sites where substitutions occur; this substitution rate reflects the degree of variation at the protein level. Since amino acid variation may change the function of viral proteins and affect transmission, the amino acid variation coefficient is defined , =α× ; α is the adjustment coefficient, which is the weight coefficient obtained based on the regression analysis of historical data, and is used to quantify the impact of amino acid mutations on the virus's ability to spread.

[0030] Gene segment recombination risk value: detect gene segment recombination and calculate the similarity s between the suspected recombinant segment and the reference sequence, s= ; For the same number of bases, is the length of the recombinant fragment; construct the gene fragment recombination risk value , =β×(1- ); β is an adjustment coefficient, which is set according to different virus types. The lower the similarity s, the higher the recombination risk value, which means that the virus is more likely to develop new characteristics due to recombination.

[0031] Propagation feature extraction: Variant transmission acceleration factor: Calculates the speed of transmission based on the number of new cases per day and the extent of transmission , = ,in is the number of new cases on day t, is the number of new cases on day t-1.

[0032] Mutation Range Expansion Index: Calculates the range expansion rate , = ,in is the number of areas with respiratory virus transmission on day t, is the number of areas with respiratory virus transmission on day t-1.

[0033] Defining mutation propagation acceleration factors =γ×( + + This factor reflects the effect of the degree of viral sequence mutation on the speed of transmission (i.e., the easier it is for the virus to mutate, the easier it is to produce a contagious strain, which in turn is more contagious and spreads faster). The more significant the mutation, the larger the acceleration factor value, and the more obvious the increase in the transmission speed.

[0034] Constructing a range expansion index = × ; Combining changes in transmission speed with range expansion reflects the virus's comprehensive expansion capacity in transmission scale due to mutations, providing a more comprehensive basis for early warning.

[0035] Model construction: A CNN and LSTM architecture is used. CNN extracts local patterns of viral sequence mutation characteristics, and LSTM processes the dynamic change trends of the time series of transmission data.

[0036] Model training: Extract the virus sequence features ( 、 、 ), propagation characteristics ( 、 ) and comprehensive characteristics ( ) as input, and whether a new variant outbreak occurs as the output label (1 if it occurs, 0 if it does not occur). The cross entropy loss function L is used to quantify the difference between the model's predicted probability and the true label, L=- ; Adjust parameters through the back propagation algorithm to minimize the loss function and improve prediction accuracy.

[0037] Where N is the number of samples, is the true label of the k-th sample; is the predicted probability of the model for k samples, ranging from 0 to 1.

[0038] Model optimization: L1 and L2 regularization and Dropout technology are used to prevent overfitting, and the optimized loss function =L+ + ; in, 、 is the regularization parameter, are model parameters, is the number of parameters, which improves the generalization ability of the model.

[0039] Early warning judgment: The trained model inputs new data and outputs a predicted probability value. The early warning threshold T is set based on historical data and / or expert experience. When the predicted probability is greater than T, an early warning signal is issued.

[0040] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention specification, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A deep learning-based early warning method for new respiratory virus variants, characterized in that: include: Data collection and preprocessing: collecting full genome sequence data of respiratory viruses, the number of new cases of respiratory viruses per day, and the scope of their spread; Extract variation features and calculate nucleotide variation index , = ; Translate nucleotide sequences into amino acid sequences and calculate amino acid variation coefficients , =α× ; Calculate the risk value of gene segment recombination , =β×(1- ); define the mutation propagation acceleration factor =γ×( + + ); in, is the length of the viral sequence, To obtain the maximum variation frequency for each site, n is the number of nucleotide types, is the number of occurrences of the jth nucleotide at the i-th site; represents the total number of occurrences of all nucleotide types at the i-th site in the same viral sequence population; α is the adjustment coefficient, is the length of the amino acid sequence, for The number of sites where substitutions occurred; β is the adjustment coefficient, For the same number of bases, is the length of the recombinant fragment; Extract transmission characteristics and calculate the transmission speed based on the number of new cases per day and the scope of transmission and range extension rate , constructing a range expansion index = × ; Build a deep learning model, using a CNN and LSTM architecture to 、 、 、 、 、 As input, whether a new variant epidemic outbreak occurs is used as the output label; the constructed model inputs new data and outputs a predicted probability value.

2. The early warning method for new variants of respiratory viruses based on deep learning according to claim 1, characterized in that: The propagation speed = ,in is the number of new cases on day t, is the number of new cases on day t-1.

3. The early warning method for new respiratory virus variants based on deep learning according to claim 1, characterized in that: The range extension rate = ,in is the number of areas with respiratory virus transmission on day t, is the number of areas with respiratory virus transmission on day t-1.

4. The early warning method for new respiratory virus variants based on deep learning according to claim 1, characterized in that: The architecture combines CNN and LSTM. CNN extracts local patterns of virus sequence mutation characteristics, and LSTM processes the dynamic change trends of the time series of transmission data.

5. The early warning method for new respiratory virus variants based on deep learning according to claim 1, characterized in that: When building a deep learning model, the cross entropy loss function L is used to quantify the difference between the model's predicted probability and the true label, L=- ; Where N is the number of samples, is the true label of the k-th sample; is the predicted probability of the model for k samples, ranging from 0 to 1.

6. The early warning method for new respiratory virus variants based on deep learning according to claim 5, characterized in that: When building a deep learning model, it also includes model optimization, using L1 and L2 regularization and Dropout technology to prevent overfitting, and the optimized loss function =L+ + ; in, 、 is the regularization parameter, are model parameters, is the number of parameters, which improves the generalization ability of the model.

7. The early warning method for new respiratory virus variants based on deep learning according to claim 5, characterized in that: The cross entropy loss function L adjusts parameters through the back propagation algorithm to minimize the loss function and improve prediction accuracy.

8. The early warning method for new respiratory virus variants based on deep learning according to claim 1, characterized in that: The constructed model inputs new data and outputs a predicted probability value. The warning threshold T is set based on historical data and / or expert experience, and a warning signal is issued when the predicted probability is greater than T.

9. The early warning method for new respiratory virus variants based on deep learning according to claim 1, characterized in that: α is the weight coefficient obtained from the regression analysis of historical data, which is used to quantify the impact of amino acid mutations on the virus's ability to spread.

10. The early warning method for new respiratory virus variants based on deep learning according to claim 1, characterized in that: β is the adjustment coefficient set according to different virus types.

Images