Early warning method and system for new respiratory infectious diseases

Abstract

The invention relates to the technical field of infectious disease detection, and discloses a new respiratory infectious disease early warning method, which comprises the following steps: S1, specimen collection and data collection: collecting biological samples of new respiratory infectious disease virus infected individuals and other respiratory infectious disease virus infected individuals, the biological sample comprises a blood sample and a swab sample; meanwhile, basic information of the individuals is collected, wherein the basic information comprises gender, age and respiratory symptom duration; s2, quantitative detection: the biological sample is subjected to quantitative detection by adopting reagents of various respiratory infectious disease viruses, the detection content comprises antigens and antibodies, and the antibodies comprise an I gM antibody, an I gG antibody and an I gA antibody; according to the invention, basic information of individuals and antigen-antibody detection reagent resources of various respiratory infectious disease viruses are deeply integrated, so that the recognition capability of new respiratory infectious disease virus infected individuals is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of infectious disease detection, and focuses on a method for identifying new respiratory infectious disease viruses by combining multiple antigen-antibody data. BACKGROUND

[0002] New respiratory infectious diseases are a major challenge in current disease prevention and control. Due to their uncertainty and lack of specific treatment and prevention measures, they have caused immeasurable harm to humans and posed a serious challenge to the prevention and control of medical and public health institutions. For new respiratory infectious diseases, early and accurate detection and identification of infected individuals and the implementation of prevention and control measures are not only an important means to prevent the spread of infectious diseases, but also a key process to promote health and save lives.

[0003] In the event of a new respiratory infectious disease outbreak, targeted antigen or antibody reagents have unique value in determining whether an individual is infected. However, for new infectious diseases, due to limited understanding, there is a lack of accurate identification means in the early prevention and control process. Even if a dedicated antigen or antibody detection reagent is developed, there are problems such as low recognition accuracy and the need for product iteration, making it difficult to meet the prevention and control needs in a timely manner. Therefore, there is an urgent need to combine multiple existing antigen and antibody detection results and establish innovative methods to integrate the results to enhance the detection efficiency of epidemic prevention and control. SUMMARY

[0004] The present application aims to make full use of the advantages of multiple antigen and antibody detection to significantly improve the accuracy and reliability of identifying new respiratory infectious disease viruses, and to provide more scientific and accurate basis for early diagnosis, prevention and control, and subsequent treatment of new infectious diseases.

[0005] To achieve the above-mentioned application purposes, the present application provides an early warning method for respiratory infectious diseases, comprising the following steps:

[0006] S1, specimen collection and data collection: collecting biological samples of individuals infected with new respiratory infectious disease viruses and other respiratory infectious disease viruses, the biological samples including blood samples and swab samples; collecting basic information of the individuals, the basic information including gender, age and duration of respiratory symptoms;

[0007] S2, quantitative detection: using multiple respiratory infectious disease virus reagents to quantitatively detect the biological samples, the detection content including antigens and antibodies, the antibodies including IgM antibodies, IgG antibodies and IgA antibodies;

[0008] S3, constructing an ensemble learning model: selecting multiple base learners, taking the basic information, the antigen and antibody values of the sample quantitative detection as features, and taking whether it is a positive infection of the new respiratory infectious disease virus as a label to construct a data set; dividing the data set into a training set and a validation set in proportion, training each base learner based on the training set, combining the trained base learners using an ensemble strategy to construct an ensemble learning model, and evaluating and optimizing the model through the validation set;

[0009] S4, effective reagent screening: calculating the SHAP value of the reagent in each base learner, and determining the effective reagent based on the SHAP value;

[0010] S5, early warning application: collecting biological samples and basic information of individuals with unknown infection status, using the effective reagent for quantitative detection, inputting the detection results and basic information into the optimized ensemble learning model, and outputting early warning results.

[0011] Preferably, in step S1, the biological sample collection object includes individuals infected with the new respiratory infectious disease virus and individuals infected with other respiratory infectious disease viruses, and the other respiratory infectious disease viruses include at least one of influenza A virus, influenza B virus, novel coronavirus, other coronavirus, respiratory syncytial virus, and rhinovirus.

[0012] Preferably, in step S2, the specific way of quantitative detection is: using real-time fluorescent quantitative PCR technology to detect antigens in swab samples to obtain Ct values of each virus; using enzyme-linked immunosorbent assay to detect antibodies in blood samples to obtain OD values of each antibody.

[0013] Preferably, in step S3, the base learners include random forest, support vector machine, naive Bayes, neural network and K- nearest neighbor algorithm; the data set is randomly divided into a training set and a validation set in a ratio of 70% and 30%; the ensemble strategy is a weight method, and the weights of each base learner are equal in the initial state, and the weights are dynamically adjusted based on the F1 value of the validation set.

[0014] Preferably, in step S3, the model evaluation indicators include accuracy, precision, recall and F1 value, and the model is optimized based on the indicators.

[0015] A new respiratory infectious disease early warning system for implementing the above method, characterized in that it comprises:

[0016] A sample collection module for collecting biological samples of individuals infected with the new respiratory infectious disease virus and other respiratory infectious disease viruses, and collecting basic information of the individuals;

[0017] A quantitative detection module is configured to quantitatively detect antigens and antibodies of the biological sample by using reagents for multiple respiratory infectious disease viruses.

[0018] A model construction module is configured to select a base learner, construct a data set with the basic information and the quantitative detection result as features and the infection state as a label, train and construct an ensemble learning model based on the data set, and evaluate and optimize the model through a validation set.

[0019] A reagent screening module is configured to calculate SHAP values of each reagent and screen effective reagents.

[0020] An early warning module is configured to receive sample detection results and basic information of an individual with unknown infection state, and output early warning results through the optimized ensemble learning model.

[0021] Preferably, the basic information collected by the sample collection module includes gender, age and duration of respiratory symptoms; and the biological sample includes a blood sample and a swab sample.

[0022] Preferably, in the quantitative detection module, antigen detection uses real-time fluorescent quantitative PCR technology to output Ct values; and antibody detection uses enzyme-linked immunosorbent assay to detect antibody types including IgM antibody, IgG antibody and IgA antibody, and outputs OD values.

[0023] Preferably, in the model construction module, the base learner includes random forest, support vector machine, naive Bayes, neural network and K-neighbor algorithm; the data set is divided into 70% training set and 30% validation set, the ensemble strategy is weight method, and the weight of each base learner is dynamically adjusted based on the F1 value of the validation set.

[0024] Preferably, the reagent screening module screens effective reagents by the following method: TreeSHAP method is used to calculate the SHAP value of the random forest base learner, KernelSHAP method is used to calculate the SHAP value of the support vector machine, naive Bayes, neural network and K-neighbor algorithm base learner; the global SHAP value of each reagent is calculated based on the average value method, the threshold value is determined according to the error gradient change principle, and the reagent with a SHAP value greater than the threshold value is reserved as an effective reagent.

[0025] Compared with the prior art, the present application has the following unique advantages:

[0026] 1. The present application deeply integrates individual basic information and antigen and antibody detection reagent resources of multiple respiratory infectious disease viruses, effectively improving the recognition ability of individuals infected with new respiratory infectious disease viruses.

[0027] 2. This invention screens effective detection reagents, eliminates redundant information, reduces the workload of detection, and also reduces model complexity and computational cost, making it more conducive to efficient application. Attached Figure Description

[0028] Figure 1 This is a flowchart of the antigen-antibody detection and joint identification algorithm for emerging respiratory infectious disease viruses in this invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0030] Reference Figure 1 This invention discloses an early warning method for respiratory infectious diseases, comprising the following steps:

[0031] S1. Specimen Collection and Data Collection: Collect biological samples from individuals infected with emerging respiratory infectious disease viruses and other respiratory infectious disease viruses, including blood samples and swab samples; and collect basic information about the individuals, including gender, age, and duration of respiratory symptoms.

[0032] S2. Quantitative detection: The biological sample is quantitatively detected using reagents for various respiratory infectious disease viruses. The detection includes antigens and antibodies, including IgM antibodies, IgG antibodies, and IgA antibodies.

[0033] S3. Construct an ensemble learning model: Select multiple base learners, using the basic information, the antigen and antibody values ​​of quantitative sample detection as features, and whether the individual is a positive case of a newly emerging respiratory infectious disease virus as a label to construct a dataset; divide the dataset into a training set and a validation set according to the proportions, train each base learner based on the training set, combine the trained base learners using an ensemble strategy to construct an ensemble learning model, and evaluate and optimize the model using the validation set;

[0034] S4. Screening of effective reagents: Calculate the SHAP value of the reagents in each base learner, and determine the effective reagents based on the SHAP value;

[0035] S5. Early warning application: Collect biological samples and basic information from individuals with unknown infection status, perform quantitative detection using the effective reagents, input the detection results and basic information into the optimized ensemble learning model, and output early warning results.

[0036] In a preferred embodiment, in step S1, the biological samples are collected from individuals diagnosed with newly emerging respiratory infectious disease viruses and individuals clearly infected with other respiratory infectious disease viruses, including at least one of influenza A virus, influenza B virus, novel coronavirus, other coronaviruses, respiratory syncytial virus, and rhinovirus.

[0037] In a preferred embodiment, the specific method of quantitative detection in step S2 is as follows: real-time quantitative PCR technology is used to detect antigens in swab samples to obtain the Ct value of each virus; enzyme-linked immunosorbent assay (ELISA) is used to detect antibodies in blood samples to obtain the OD value of each antibody.

[0038] In a preferred implementation, in step S3, the base learners include random forest, support vector machine, Naive Bayes, neural network and K-nearest neighbor algorithm; the dataset is randomly divided into training set and validation set at a ratio of 70% and 30%; the ensemble strategy is a weighted method, in which the weights of each base learner are equal in the initial state, and the weights are dynamically adjusted based on the F1 value of the validation set.

[0039] In a preferred implementation, in step S3, the model evaluation metrics include accuracy, precision, recall, and F1 score, and the model is optimized based on these metrics.

[0040] A respiratory infectious disease early warning system that implements the above method, characterized in that it includes:

[0041] The sample collection module is used to collect biological samples from individuals infected with emerging respiratory infectious disease viruses and other respiratory infectious disease viruses, and to collect basic information about the individuals.

[0042] The quantitative detection module is used to quantitatively detect antigens and antibodies in the biological sample using reagents for various respiratory infectious disease viruses.

[0043] The model building module is used to select a base learner, construct a dataset with the basic information and quantitative detection results as features and the infection status as a label, train and build an ensemble learning model based on the dataset, and evaluate and optimize the model through a validation set.

[0044] The reagent screening module is used to calculate the SHAP value of each reagent and screen for effective reagents.

[0045] The early warning module is used to receive sample detection results and basic information of individuals with unknown infection status, and output early warning results through an optimized ensemble learning model.

[0046] In a preferred embodiment, the basic information collected by the sample collection module includes gender, age, and duration of respiratory symptoms; the biological samples include blood samples and swab samples.

[0047] In a preferred embodiment, the quantitative detection module employs real-time quantitative PCR technology for antigen detection, outputting a Ct value; and uses enzyme-linked immunosorbent assay (ELISA) for antibody detection, detecting IgM, IgG, and IgA antibodies, outputting an OD value.

[0048] In a preferred embodiment, the base learners in the model building module include random forest, support vector machine, naive Bayes, neural network and K-nearest neighbor algorithm; the dataset is divided into 70% training set and 30% validation set, and the ensemble strategy is weighted method, which dynamically adjusts the weights of each base learner based on the F1 value of the validation set.

[0049] In a preferred embodiment, the reagent screening module screens effective reagents in the following way: the SHAP value of the random forest base learner is calculated using the TreeSHAP method, and the SHAP value of the support vector machine, Naive Bayes, neural network, and K-nearest neighbor algorithm base learners is calculated using the KernelSHAP method; the global SHAP value of each reagent is calculated based on the average value method, a threshold is determined according to the error gradient change principle, and reagents with SHAP values ​​greater than the threshold are retained as effective reagents.

[0050] To enable those skilled in the art to fully understand and implement the present invention, the following specific embodiment will be used to further explain the specific implementation principle of the present invention.

[0051] 1. Specimen collection and data gathering

[0052] Specimen collection is a crucial foundation for algorithm development. This invention requires the systematic and comprehensive collection of biological samples from two key groups of individuals: those diagnosed with a novel respiratory infectious disease virus and those confirmed to be infected with other respiratory infectious disease viruses. Biological samples, including swabs and blood samples, are collected from these individuals. Simultaneously, basic information that may influence infection is gathered, including sex, age, and duration of respiratory symptoms. The number of samples should be as large as possible to ensure sufficient data for model construction.

[0053] 2. Quantitative detection

[0054] (1) Antigen detection

[0055] Real-time quantitative PCR technology was used to detect various respiratory infectious disease viruses in collected swab samples, including but not limited to emerging respiratory infectious disease viruses, influenza A virus, influenza B virus, novel coronavirus, other coronaviruses, respiratory syncytial virus, rhinovirus, etc., and to obtain the corresponding Ct values ​​of various respiratory pathogens.

[0056] (2) Antibody detection

[0057] Enzyme-linked immunosorbent assay (ELISA) was used to detect antibodies in collected blood samples, including but not limited to various types of viruses such as emerging respiratory infectious disease viruses, influenza A virus, influenza B virus, novel coronavirus, other coronaviruses, respiratory syncytial virus, and rhinovirus. Antibody types included IgM antibodies, IgG antibodies, and IgA antibodies. The OD values ​​of various antibodies in the ELISA detection were obtained.

[0058] 3. Construct an ensemble learning model

[0059] Ensemble algorithms, by combining the power of multiple base learners, have significant advantages in terms of model performance, stability, and generalization ability.

[0060] (1) Selecting a base learner

[0061] This invention is applied to determine whether an individual is infected with a novel respiratory infectious disease virus, and belongs to the category of classification algorithms. The machine learning model incorporated into the ensemble algorithm is called the base learner. The ensemble algorithm constructed in this invention may include random forests, support vector machines, Naive Bayes, neural networks, and the K-nearest neighbors algorithm as its base learners.

[0062] (2) Training base learner

[0063] First, a learning model dataset is constructed using the collected basic information, antigen and antibody values ​​from quantitative sample detection as features, and whether the individual is a positive case (0-1 variable) as a label. Then, 70% of the data is randomly selected as the training set, and the remaining 30% is used as the validation set. Each base learner is trained independently, and the output results of different base learners are obtained.

[0064] (3) Integration Model Construction

[0065] The output is a classification value, either positive or negative. Therefore, a weighted approach is used as the ensemble strategy to combine different base learners. Taking the application of 5 base learners as an example, the specific steps are as follows:

[0066] Assign weights ω to each base learner i The initial weights can be set to be equal, i.e., ω1=ω2=ω3=ω4=ω5=1.

[0067] For a sample x that is the input predictor variable, the output of the ensemble model is:

[0068]

[0069] h i (x) represents the prediction result (negative 0 or positive 1) of the i-th base learner for sample x, ω i This represents the weight of the i-th base learner, reflecting the confidence level. The initial value for each base learner is 1, with equal weights. The ensemble model calculates a weighted sum of the predictions from all base learners. If the weighted sum is greater than or equal to half the total weights, the ensemble model predicts a positive result; otherwise, it predicts a negative result.

[0070] (4) Evaluation and Optimization

[0071] Based on the validation dataset, the ensemble model is evaluated. Accuracy, Precision, Recall, and the harmonic mean (F1) of Precision and Recall are calculated based on the true values ​​and model output values ​​on the validation set to evaluate the model and serve as metrics for model optimization. Based on the performance on the validation set, the weights of the base learners are dynamically adjusted using the comprehensive metric, the harmonic mean (F1). The weight adjustment formula for the i-th base learner is:

[0072]

[0073] 4. Screening of effective reagents

[0074] Ranking reagent results by their contribution and screening for effective reagents can eliminate redundant information, reduce testing workload, and lower model complexity and computational costs, thus facilitating efficient application. This invention uses SHAP values ​​for screening. SHAP value calculation is based on the concept of Shapley values, which are theoretically derived from cooperative game theory and aim to measure the contribution of each participant to the overall payoff in a cooperative alliance.

[0075] (1) Calculate the SHAP of each base learner

[0076] For a random forest base learner, the TreeSHAP method is used to calculate the SHAP value. For an input sample x, the SHAP value of its j-th feature is:

[0077] φ j (x)=∑ n∈N ω(n)·[v(S n ∪{j})-vs n ]

[0078] N represents the set of all paths in the random forest model. ω(n) is the weight of path n, which is related to the probability that a sample traverses that path. Sn It is the set of features already used in path n. n This indicates the marginal contribution of the feature subset S.

[0079] For base learners such as Support Vector Machines, Naive Bayes, Neural Networks, and K-Nearest Neighbors, the KernelSHAP method is used to calculate the SHAP value. Its core idea is to solve the following weighted linear regression problem:

[0080]

[0081] F is the set of all features, S is a subset of all features, f(S) is the predicted output of the model using only the feature subset S, and π(S) is the kernel weight.

[0082] (2) Determine the contribution of the reagents

[0083] For each feature, there are 5 SHAP values ​​from 5 base learners. The average SHAP value is calculated using the averaging method and is used as the SHAP value for that feature in the ensemble learning:

[0084]

[0085] Then, to assess the contribution of each reagent, all samples were included in the calculation of the global SHAP value:

[0086]

[0087] The absolute values ​​of the SHAP values ​​for different samples x are taken, because the sign of the SHAP value only indicates the direction of influence, while the absolute value indicates the intensity of the influence. n represents the total number of samples. Based on this formula, the global SHAP value for each reagent can be obtained, representing the contribution of that feature to identifying whether a sample is positive or negative. The samples are then sorted from largest to smallest global SHAP value.

[0088] (3) Screening effective reagents

[0089] Based on the changes in SHAP, the SHAP value threshold is determined using the error gradient variation principle, and reagents with SHAP values ​​greater than the threshold are retained as valid reagents. The specific steps are as follows:

[0090] Define a cumulative contribution function to calculate the cumulative contribution of all reagents:

[0091]

[0092] All reagents are sorted according to their SHAP values, and the gradient rate of change of adjacent features can be calculated:

[0093] ΔC(k)=C(k)-C(k-1)

[0094] Based on the gradient rate of change, the inflection point (Thr) of gradient change is monitored using the second derivative:

[0095]

[0096] Finally, all reagents with SHAP values ​​greater than the threshold are retained. In short, even including more results from other reagents will not significantly improve the model's recognition performance.

[0097] Optimization and Validation

[0098] The stability of the feature subset was evaluated using k-fold cross-validation. First, the SHAP value calculation and feature selection were repeated at each fold, and the frequency of each selected feature was counted. High-frequency features (reagents) were then retained. Second, the performance before and after feature selection was compared using metrics such as accuracy, precision, recall, and the harmonic mean (F1) of precision and recall.

[0099] 5. Predictive Applications

[0100] Sampling and basic information collection are performed on individuals with newly emerging respiratory infectious disease pathogens whose infection status is unknown. The collected samples are then tested using effective reagents to obtain corresponding quantitative values. The obtained quantitative values ​​and basic information are input into a reconstructed ensemble learning model. During this process, the model integrates the test results from multiple reagents to make a comprehensive judgment.

[0101] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0102] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for early warning of emerging respiratory infectious diseases, characterized in that, Includes the following steps: S1. Specimen Collection and Data Collection: Collect biological samples from individuals infected with emerging respiratory infectious disease viruses and other respiratory infectious disease viruses, including blood samples and swab samples; and collect basic information about the individuals, including gender, age, and duration of respiratory symptoms. S2. Quantitative detection: The biological sample is quantitatively detected using reagents for various respiratory infectious disease viruses. The detection includes antigens and antibodies, including IgM antibodies, IgG antibodies, and IgA antibodies. S3. Construct an ensemble learning model: Select multiple base learners, using the basic information, the antigen and antibody values ​​of quantitative sample detection as features, and whether the individual is a positive case of a newly emerging respiratory infectious disease virus as a label to construct a dataset; divide the dataset into a training set and a validation set according to the proportions, train each base learner based on the training set, combine the trained base learners using an ensemble strategy to construct an ensemble learning model, and evaluate and optimize the model using the validation set; S4. Screening of effective reagents: Calculate the SHAP value of the reagents in each base learner, and determine the effective reagents based on the SHAP value; S5. Early warning application: Collect biological samples and basic information from individuals whose infection status of emerging infectious diseases is unknown, perform quantitative detection using the effective reagents, input the detection results and basic information into the optimized ensemble learning model, and output early warning results.

2. The method according to claim 1, characterized in that, In step S1, the biological samples are collected from individuals diagnosed with newly emerging respiratory infectious disease viruses and individuals clearly infected with other respiratory infectious disease viruses, including at least one of influenza A virus, influenza B virus, novel coronavirus, other coronaviruses, respiratory syncytial virus, and rhinovirus.

3. The method according to claim 1, characterized in that, In step S2, the specific method of quantitative detection is as follows: real-time fluorescence quantitative PCR technology is used to detect antigens in swab samples to obtain the Ct value of each virus; enzyme-linked immunosorbent assay (ELISA) is used to detect antibodies in blood samples to obtain the OD value of each antibody.

4. The method according to claim 1, characterized in that, In step S3, the base learners include random forest, support vector machine, naive Bayes, neural network and K-nearest neighbor algorithm; the dataset is randomly divided into training set and validation set at a ratio of 70% and 30%; the ensemble strategy is the weight method, in which the weights of each base learner are equal in the initial state, and the weights are dynamically adjusted based on the F1 value of the validation set.

5. The method according to claim 1, characterized in that, In step S3, the evaluation metrics for the model include accuracy, precision, recall, and F1 score, and the model is optimized based on these metrics.

6. A respiratory infectious disease early warning system implementing the method of any one of claims 1-5, characterized in that, include: The sample collection module is used to collect biological samples from individuals infected with emerging respiratory infectious disease viruses and other respiratory infectious disease viruses, and to collect basic information about the individuals. The quantitative detection module is used to quantitatively detect antigens and antibodies in the biological sample using reagents for various respiratory infectious disease viruses. The model building module is used to select a base learner, construct a dataset with the basic information and quantitative detection results as features and the infection status as a label, train and build an ensemble learning model based on the dataset, and evaluate and optimize the model through a validation set. The reagent screening module is used to calculate the SHAP value of each reagent and screen for effective reagents. The early warning module is used to receive sample detection results and basic information of individuals with unknown infection status, and output early warning results through an optimized ensemble learning model.

7. The system according to claim 6, characterized in that, The basic information collected by the sample collection module includes gender, age, and duration of respiratory symptoms; the biological samples include blood samples and swab samples.

8. The system according to claim 6, characterized in that, In the quantitative detection module, antigen detection uses real-time quantitative PCR technology and outputs Ct value; antibody detection uses enzyme-linked immunosorbent assay (ELISA), and the antibody types detected include IgM antibody, IgG antibody, and IgA antibody, outputting OD value.

9. The system according to claim 6, characterized in that, In the model building module, the base learners include random forest, support vector machine, naive Bayes, neural network and K-nearest neighbor algorithm; the dataset is divided into 70% training set and 30% validation set, and the ensemble strategy is weighted method, which dynamically adjusts the weights of each base learner based on the F1 value of the validation set.

10. The system according to claim 6, characterized in that, The reagent screening module screens effective reagents in the following ways: it uses the TreeSHAP method to calculate the SHAP value of the random forest base learner, and the KernelSHAP method to calculate the SHAP value of the support vector machine, Naive Bayes, neural network and K-nearest neighbor algorithm base learners. The global SHAP value of each reagent is calculated using the average value method. A threshold is determined according to the error gradient change principle, and reagents with SHAP values ​​greater than the threshold are retained as valid reagents.

Images