Crowd antibody concentration dynamic evaluation method based on neural network simulation calculation

Abstract

The invention discloses a population antibody concentration dynamic evaluation method based on neural network simulation calculation. The method comprises the following steps: S1, acquiring a multi-modal immunology related data set; s2, generating a complete multi-modal feature data set; s3, constructing a deep learning model based on the multi-modal feature data set; s4, training the segmented neural network model; s5, inputting the multi-modal immunology related data of the to-be-evaluated crowd into the trained sectional neural network model, and generating a prediction result of the individual antibody concentration changing along with time; s6, summarizing prediction data of a plurality of individuals according to the prediction result of the individual antibody concentration changing along with time, and establishing a population overall antibody concentration dynamic distribution model; and S7, providing vaccination optimization strategy suggestions and crowd immune intervention schemes in combination with data output. The invention provides key technical support for immune monitoring and vaccination strategy optimization.

Description

Technical Field

[0001] The present invention relates to the field of immunology, and particularly to a method for dynamically evaluating the antibody concentration of a population based on neural network simulation calculation. Background Art

[0002] With the rapid development of artificial intelligence and deep learning technologies, intelligent tools have been gradually introduced into the fields of medicine and public health to assist in disease prediction, individual health monitoring, and population immunity assessment. Among them, the dynamic evaluation of population antibody concentration is an important link in vaccine inoculation effect monitoring and epidemic control. By analyzing the dynamic changes of individual and population antibody concentrations over time, the protective efficacy of vaccines can be scientifically evaluated and reasonable immunization strategies can be formulated.

[0003] In the prior art, most of the methods for evaluating antibody concentration are based on static detection means, including single measurement and analysis after collecting serum antibody concentration through laboratory detection. Existing methods usually rely on sampling point data to record the antibody levels of individuals or populations at stages, but it is difficult to capture the time dynamic change trend of antibody concentration. In addition, such methods mostly describe the change law of antibody concentration based on linear statistical models or simple curve fitting, and cannot effectively cope with the complex non-linear characteristics in the immune response. The accuracy and adaptability of static methods are further reduced in the case of significant differences in individual immune systems or combined vaccination with multiple vaccines.

[0004] In population immunity assessment, most of the prior art adopts simple statistical analysis means, including calculating the population antibody level based on the mean or standard deviation. Existing methods are difficult to comprehensively reflect the dynamic change laws among different populations, and it is difficult to provide targeted dynamic evaluation results when the population immune status is heterogeneous. For the immune monitoring of large-scale populations, the calculation efficiency of existing methods is low, and it is difficult to effectively integrate multi-modal data, resulting in insufficient data utilization and limiting the practical application scope of the technology.

[0005] In summary, the prior art has significant deficiencies in the modeling accuracy of the dynamic changes of antibody concentration, the adaptability of population immunity assessment, and the high-efficiency computing ability for large-scale populations, and it is difficult to meet the requirements of dynamic evaluation, accurate prediction, and efficient decision-making in modern public health management. The existing technical defects directly affect the optimization of vaccine inoculation strategies and the prevention and control effects of epidemics. Summary of the Invention

[0006] An object of the present invention is to propose a method for dynamically evaluating the antibody concentration of a population based on neural network simulation calculation, which provides key technical support for immune monitoring and the optimization of vaccine inoculation strategies.

[0007] A method for dynamically evaluating the antibody concentration of a population based on neural network simulation calculation according to an embodiment of the present invention includes the following steps:

[0008] S1. Obtain a multi-modal immunology-related data set;

[0009] S2. Preprocess the multi-modal immunology-related data set to generate a complete multi-modal feature data set;

[0010] S3. Build a deep learning model based on the multi-modal feature data set, design a segmented neural network architecture, which is divided into a sub-network for modeling the rapid growth stage of antibody concentration, a sub-network for modeling the stable stage of antibody concentration, and a sub-network for modeling the decay stage of antibody concentration;

[0011] S4. Train the segmented neural network model, and optimize the parameters of the segmented neural network model through the backpropagation algorithm during the training process;

[0012] S5. Input the multi-modal immunology-related data of the population to be evaluated into the trained segmented neural network model, and perform segmented dynamic simulation calculations through the sub-network for modeling the rapid growth stage of antibody concentration, the sub-network for modeling the stable stage of antibody concentration, and the sub-network for modeling the decay stage of antibody concentration respectively, to generate a prediction result of the change of individual antibody concentration over time;

[0013] S6. According to the prediction result of the change of the individual antibody concentration over time, summarize the prediction data of multiple individuals, establish a dynamic distribution model of the overall antibody concentration of the population, and generate dynamic immune distribution characteristics evolving over time;

[0014] S7. Input the result of the dynamic distribution model of the overall antibody concentration of the population into the visualization module, display the stage change trend of the antibody concentration by generating a dynamic change chart and an analysis report, and provide suggestions for optimizing the vaccination strategy and a population immunity intervention plan in combination with the data output.

[0015] Optionally, the S1 includes the following steps:

[0016] S11. Obtain population serum antibody concentration data, obtain vaccination records, obtain individual immune history data, and obtain biological experiment data;

[0017] S12. Combine the population serum antibody concentration data, vaccination records, individual immune history data, and biological experiment data to construct a multi-modal immunology-related data set D m :

[0018] D m ={(d ai ,d vi ,d hi ,d bi)∣i = 1, 2, …, n};

[0019] where, d ai represents the serum antibody concentration data of the i-th individual, d vi represents the vaccination information of the i-th individual, including the type, dose, vaccination time, and vaccination interval of the vaccine, d hi represents the immunological history data of the i-th individual, including past infection history, autoimmune disease records, and related genetic information, d bi represents the immune-related data obtained by experimental analysis of the i-th individual, including neutralizing antibody level, T cell activity index, and antigen-specific reaction intensity;

[0020] S13. Store the multimodal immunology-related dataset D m in a structured data form, and mark each modal data with a timestamp t i , where the timestamp indicates the specific time point of data collection, to obtain the timestamped multimodal immunology-related dataset:

[0021]

[0022] Optionally, S2 includes the following steps:

[0023] S21. Clean the multimodal immunology-related dataset D′ m by removing the data with null values, duplicate values, or obvious error values in the incomplete data entries. The error values include serum antibody concentration values outside the biologically reasonable range, invalid vaccination records, or invalid timestamp data. After data cleaning, a cleaned multimodal immunology-related dataset is generated;

[0024] S22. Detect outliers in the cleaned multimodal immunology-related dataset. By calculating the mean and standard deviation of each modal feature data, remove the outlier data outside the range [μ f - kσ f , μ f + kσ f , where k is the deviation coefficient for outlier judgment, and the feature data includes antibody concentration, vaccination time, immunological history, and biological experiment data, to obtain the multimodal immunology-related dataset after removing outliers;

[0025] S23. Standardize the multimodal immunology-related dataset after removing outliers, and normalize all modal feature data to the unified range [0, 1];

[0026] S24. Extract features from the standardized multi-modal immunology-related dataset using an autoencoder, perform feature dimensionality reduction on the high-dimensional modal features, and obtain the multi-modal feature dataset F after dimensionality reduction. m :

[0027]

[0028] Among them, and represent the antibody concentration feature, vaccination feature, immune history feature, and biological experiment feature after dimensionality reduction respectively, and N a represents the number of valid individuals finally retained after data cleaning, outlier removal, and feature standardization.

[0029] Optionally, the S3 includes the following steps:

[0030] S31. Based on the multi-modal feature dataset F m Allocate each feature entry to the corresponding time period according to the timestamp to generate a phased input feature set

[0031]

[0032] Among them, t∈T represents the time label of different time periods, corresponding to the dynamic change stage of antibody concentration;

[0033] To integrate the phased input feature set, define the comprehensive input feature:

[0034]

[0035] Among them, is the comprehensive input feature for dynamic modeling, directly representing the dynamic characteristics of the antibody concentration of the i-th individual in the time period t, w a , w v , w h , w b are the weights of the antibody concentration feature, vaccination feature, immune history feature, and biological experiment feature respectively, reflecting the importance of different features for dynamic evaluation;

[0036] S32. Construct a segmented neural network model M, which is divided into the following three sub-networks:

[0037] Rapid growth stage modeling sub-network M g , fitting the non-linear dynamic characteristics of the rapid increase in antibody concentration in the early stage after vaccination;

[0038] Stable stage modeling sub-network M s , fitting the fluctuation law of antibody concentration after reaching a stable level in the middle stage;

[0039] Decay phase modeling sub-network M d , fitting the changing trend of the antibody concentration gradually decreasing over time in the long term.

[0040] Optionally, S4 includes the following steps:

[0041] S41. Based on the complete multi-modal feature dataset and the comprehensive input features take the comprehensive input features corresponding to the time period t as the training input of the segmented neural network model, combined with the historical antibody concentration change data as the supervised learning label to form a training dataset:

[0042]

[0043] where represents the actual antibody concentration measurement value of the i-th individual in the time period t;

[0044] S42. Train the segmented neural network model through supervised learning, using the weighted mean squared error as the loss function:

[0045]

[0046] where is the actually measured antibody concentration value, compared with the predicted value to optimize the model. λ is the regularization coefficient, used to prevent the model from overfitting and ensure the generality of the segmented simulation calculation. and is the second norm of the weight matrix;

[0047] S43. Optimize the parameters of the segmented neural network model through the backpropagation algorithm, including the weight matrix and the bias vector The gradient calculation formula for weight update is:

[0048]

[0049]

[0050] where and respectively represent the gradients of the input layer and hidden layer weights, represents the gradient of the input layer bias, is the intermediate feature of the hidden layer, and the gradient is used to update the model parameters to minimize the loss function L;

[0051] S44. Update the parameters of the segmented neural network model using the gradient descent algorithm;

[0052] Repeat the steps from S41 to S44 until the loss function L converges to a preset threshold or reaches the set maximum number of iterations. When it converges, the parameters of the piecewise neural network model can fit the dynamic changes of historical antibody concentrations, and finally an optimized piecewise neural network model is formed.

[0053] Optionally, S5 includes the following steps:

[0054] S51. Input the multi-modal immunology-related data set of the population to be evaluated into the trained piecewise neural network model;

[0055] S52. Use the piecewise neural network model to perform piecewise dynamic simulation calculations on the input features of the population to be evaluated, and generate the predicted antibody concentration values for time period t through the rapid growth phase modeling sub-network, the stable phase modeling sub-network, and the decay phase modeling sub-network respectively

[0056]

[0057] Among them, represents the predicted antibody concentration value of the i-th individual in time period t, T g ,T s ,T d are the time ranges corresponding to the rapid growth phase, the stable phase, and the decay phase respectively, M g ,M s ,M d represent the rapid growth phase, the stable phase, and the decay phase modeling sub-networks respectively;

[0058] The calculation result of each sub-network is determined by the comprehensive input features and the trained network parameters:

[0059]

[0060] Among them, is the intermediate feature of the hidden layer, and σ is the activation function;

[0061] S53. Aggregate the predicted values for each individual in the time period to generate the overall dynamic prediction result

[0062]

[0063] Among them, is the overall predicted antibody concentration result of the i-th individual, w t is the time period weight, which is used to balance the prediction contributions of different phases.

[0064] Optionally, S6 includes the following steps:

[0065] S61. Based on the dynamic prediction results of individual antibody concentrations Calculate the average value P(t) of the overall antibody concentration of the population corresponding to the time period t:

[0066]

[0067] where P(t) represents the average antibody concentration of the population within the time period t, reflecting the overall immune level, N p is the total number of individuals in the population to be evaluated, is the individual weight;

[0068] S62. Use the time series P(t) within the time period set T to construct a dynamic distribution model M P (T) of the population antibody concentration, and define the distribution function f(P,t) of the antibody concentration within the time period t:

[0069] M P (T) = {P(t) | t ∈ T};

[0070]

[0071] where M P (T) represents the dynamic distribution model of the overall antibody concentration of the population over time, f(P,t) is the normal distribution function of the antibody concentration within the time period t, used to characterize the distribution characteristics of the population antibody concentration at time t, P(t) is the mean antibody concentration at time t, σ t is the standard deviation of the antibody concentration at time t, reflecting the degree of dispersion of the immune levels among individuals;

[0072] S63. Calculate the change rate and change trend of the population antibody concentration within the time period t, and at the same time define the antibody concentration change acceleration A(P,t) to evaluate the dynamic trend;

[0073] S64. Define visual indicators of dynamic immune characteristics based on the dynamic distribution model of the population antibody concentration and the change rate of the population antibody concentration, including the antibody concentration coverage rate C(t,P th ):

[0074]

[0075] where C(t,P th ) represents the proportion of the population with an antibody concentration higher than the threshold P th at time t, the threshold P th is used to measure the number of people who have reached an effective immune barrier, f(P,t) is the normal distribution function of the antibody concentration, and P th represents the minimum antibody concentration level required for effective immunity;

[0076] S65. Generate an optimized vaccination strategy using the dynamic distribution model of population antibody concentration and the visualization analysis results.

[0077] The beneficial effects of the present invention are as follows:

[0078] (1) The present invention adopts a segmented neural network model, and respectively designs a modeling sub-network for the rapid growth stage of antibody concentration, a modeling sub-network for the stable stage, and a modeling sub-network for the decay stage. It accurately models the non-linear dynamic characteristics of antibody concentration in different stages. Through segmented dynamic simulation calculation combined with time series data, it can accurately capture the non-linear change trend of antibody concentration after vaccination, including the complex laws of rapid generation, plateau fluctuations, and long-term decay.

[0079] (2) Through the multi-modal data fusion technology, the present invention effectively integrates serum antibody concentration data, vaccination records, individual immunization history, and biological experiment data into a unified multi-modal feature data set, and realizes dynamic data processing by combining timestamp markers. It has significant advantages in the processing of multi-modal data, not only improving the compatibility with different data sources, but also realizing the dynamic evaluation of large-scale population data through the efficient calculation of the deep neural network model.

[0080] (3) By constructing a dynamic distribution model, the present invention comprehensively generates the dynamic distribution characteristics of population antibody concentration within a time period based on the dynamic prediction results of individual antibody concentration, which can accurately reflect the evolution law of the overall immune level of the population. By introducing a dynamic distribution function and the change rate of antibody concentration, it realizes the quantitative analysis and trend prediction of the immune state of the population in different time periods, and can intuitively display the dynamic changes of population antibody concentration and the maintenance time of the immune barrier, providing a scientific basis for optimizing the vaccination strategy. Description of the Drawings

[0081] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, but do not constitute a limitation to the present invention. In the drawings:

[0082] Figure 1 is a flowchart of a method for dynamically evaluating population antibody concentration based on neural network simulation calculation proposed by the present invention. Detailed Embodiments

[0083] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only showing the basic structure of the present invention in a schematic way, so they only show the components related to the present invention.

[0084] Refer to Figure 1, A method for dynamically evaluating the antibody concentration of a population based on neural network simulation calculation, comprising the following steps:

[0085] S1. Obtain a multi-modal immunology-related data set;

[0086] S2. Preprocess the multi-modal immunology-related data set to generate a complete multi-modal feature data set;

[0087] S3. Build a deep learning model based on the multi-modal feature data set, design a segmented neural network architecture, which is divided into a modeling sub-network for the rapid growth stage of antibody concentration, a modeling sub-network for the stable stage of antibody concentration, and a modeling sub-network for the decay stage of antibody concentration;

[0088] S4. Train the segmented neural network model, and optimize the parameters of the segmented neural network model through the backpropagation algorithm during the training process;

[0089] S5. Input the multi-modal immunology-related data of the population to be evaluated into the trained segmented neural network model, and perform segmented dynamic simulation calculations through the modeling sub-network for the rapid growth stage of antibody concentration, the modeling sub-network for the stable stage of antibody concentration, and the modeling sub-network for the decay stage of antibody concentration respectively, to generate a prediction result of the individual antibody concentration changing with time;

[0090] S6. According to the prediction result of the individual antibody concentration changing with time, summarize the prediction data of multiple individuals, establish a dynamic distribution model of the overall antibody concentration of the population, and generate dynamic immune distribution characteristics evolving with time;

[0091] S7. Input the result of the dynamic distribution model of the overall antibody concentration of the population into the visualization module, display the stage change trend of the antibody concentration by generating dynamic change charts and analysis reports, and provide suggestions for optimizing the vaccination strategy and the population immunity intervention plan in combination with the data output.

[0092] In this embodiment, S1 includes the following steps:

[0093] S11. Obtain the population serum antibody concentration data, obtain the vaccination records, obtain the individual immune history data, and obtain the biological experiment data;

[0094] S12. Combine the population serum antibody concentration data, vaccination records, individual immune history data, and biological experiment data to construct a multi-modal immunology-related data set D m :

[0095]

[0096] where d ai represents the serum antibody concentration data of the i-th individual, d virepresents the vaccination information of the ith individual, including the type, dose, time and interval of vaccination, d hi represents the immunological history data of the ith individual, including previous infection history, autoimmune disease records, and related genetic information, d bi represents the immune-related data obtained through experimental analysis of the i-th individual, including neutralizing antibody levels, T cell activity indicators, and antigen-specific reaction intensity;

[0097] S13. Multimodal immunology-related dataset D m Stored in structured data format, each modal data is timestamped i , the timestamp indicates the specific time point of data collection, and the multimodal immunology-related data set after time stamping is obtained:

[0098]

[0099] In this implementation, S2 includes the following steps:

[0100] S21. Multimodal immunology-related dataset D′ m Perform data cleaning to remove incomplete data entries with empty values, duplicate values, or obvious erroneous values. The erroneous values include serum antibody concentration values that exceed the biologically reasonable range, invalid vaccination records, or invalid timestamp data. After data cleaning, a cleaned multimodal immunology-related dataset is generated.

[0101] S22. Perform outlier detection on the cleaned multimodal immunology-related datasets. Calculate the mean and standard deviation of each modality feature data and remove outliers that exceed the range [μ f -kσ f ,μ f +kσ f ], where k is the deviation coefficient of outlier judgment, and the characteristic data include antibody concentration, vaccination time, immunization history and biological experimental data, to obtain a multimodal immunology-related data set after removing outliers;

[0102] S23. Standardize the multimodal immunology-related data set after removing outliers, and normalize all modality feature data to a unified range of [0,1];

[0103] S24. Use an automatic encoder to extract features from the standardized multimodal immunology-related dataset, perform feature dimensionality reduction on the high-dimensional modal features, and obtain the multimodal feature dataset F after dimensionality reduction. m :

[0104]

[0105] in, and respectively represent the antibody concentration characteristics, vaccination characteristics, immune history characteristics, and biological experiment characteristics after dimensionality reduction. N a represents the number of effective individuals finally retained after data cleaning, outlier removal, and feature standardization processing.

[0106] In this embodiment, S3 includes the following steps:

[0107] S31. Based on the multimodal feature dataset F m Allocate each feature entry to the corresponding time period according to the timestamp to generate a phased input feature set

[0108]

[0109] where t ∈ T represents the time label of different time periods, corresponding to the dynamic change stage of antibody concentration;

[0110] To integrate the phased input feature set, define the comprehensive input feature:

[0111]

[0112] where is the comprehensive input feature for dynamic modeling, directly representing the dynamic characteristics of the antibody concentration of the i-th individual in the time period t. w a , w v , w h , w b are the weights of the antibody concentration characteristics, vaccination characteristics, immune history characteristics, and biological experiment characteristics respectively, reflecting the importance of different characteristics for dynamic evaluation;

[0113] S32. Construct a segmented neural network model M, which is divided into the following three sub-networks:

[0114] The rapid growth stage modeling sub-network M g , fitting the non-linear dynamic characteristics of the rapid increase in antibody concentration in the initial stage after vaccination;

[0115] The stable stage modeling sub-network M s , fitting the fluctuation law of antibody concentration after reaching a stable level in the middle stage;

[0116] The decay stage modeling sub-network M d , fitting the changing trend of antibody concentration gradually decreasing over time in the long term.

[0117] In this embodiment, S4 includes the following steps:

[0118] S41. Based on the complete multimodal feature dataset and the comprehensive input features Use the comprehensive input features corresponding to the time period t as the training input of the segmented neural network model, and combine the historical antibody concentration change data as the supervised learning label to form a training dataset:

[0119]

[0120] where represents the actual antibody concentration measurement value of the i-th individual in the time period t;

[0121] S42. Train the segmented neural network model through supervised learning, and use the weighted mean square error as the loss function:

[0122]

[0123] where is the actually measured antibody concentration value, compared with the predicted value to optimize the model. λ is the regularization coefficient, which is used to prevent the model from overfitting and ensure the generality of the piecewise simulation calculation. and is the second norm of the weight matrix;

[0124] S43. Optimize the parameters of the segmented neural network model through the backpropagation algorithm, including the weight matrices of the input layer and the hidden layer and the bias vector The gradient calculation formula for weight update is:

[0125]

[0126] where and respectively represent the gradients of the weights of the input layer and the hidden layer. represents the gradient of the input layer bias. is the intermediate feature of the hidden layer, and the gradient is used to update the model parameters to minimize the loss function L;

[0127] S44. Use the gradient descent algorithm to update the parameters of the segmented neural network model;

[0128] S45. Repeat the steps of S41 to S44 until the loss function L converges to a preset threshold or reaches the set maximum number of iterations. When converging, the parameters of the segmented neural network model can fit the dynamic change of the historical antibody concentration, and finally an optimized segmented neural network model is formed.

[0129] In this embodiment, S5 includes the following steps:

[0130] S51. Input the multi-modal immunology-related data set of the population to be evaluated into the trained segmented neural network model;

[0131] S52. Use the segmented neural network model to perform segmented dynamic simulation calculations on the input features of the population to be evaluated, and generate the predicted antibody concentration values at time period t through the rapid growth stage modeling sub-network, the stable stage modeling sub-network, and the decay stage modeling sub-network respectively

[0132]

[0133] where, represents the predicted antibody concentration value of the i-th individual at time period t, T g , T s , T d are the time ranges corresponding to the rapid growth stage, the stable stage, and the decay stage respectively, M g , M s , M d represent the rapid growth stage, the stable stage, and the decay stage modeling sub-networks respectively;

[0134] The calculation result of each sub-network is determined by the comprehensive input features and the trained network parameters:

[0135]

[0136] where, is the intermediate feature of the hidden layer, and σ is the activation function;

[0137] S53. Aggregate the predicted values at time periods of each individual to generate the overall dynamic prediction result of the i-th individual

[0138]

[0139] where, is the overall predicted antibody concentration result of the i-th individual, w t is the time period weight, which is used to balance the prediction contributions of different stages.

[0140] In this embodiment, S6 includes the following steps:

[0141] S61. Calculate the average value P(t) of the overall antibody concentration of the population corresponding to time period t based on the individual antibody concentration dynamic prediction result :

[0142]

[0143] Among them, P(t) represents the average antibody concentration of the population within the time period t, reflecting the overall immune level, and N p is the total number of individuals in the population to be evaluated, and is the individual weight;

[0144] S62. Use the time series P(t) within the time period set T to construct the dynamic distribution model M P (T) of the population antibody concentration, and define the distribution function f(P, t) of the antibody concentration within the time period t:

[0145] M P (T) = {P(t) | t ∈ T};

[0146]

[0147] Among them, M P (T) represents the dynamic distribution model of the overall antibody concentration of the population over time. f(P, t) is the normal distribution function of the antibody concentration within the time period t, used to characterize the distribution characteristics of the population antibody concentration at time t. P(t) is the mean antibody concentration at time t, and σ t is the standard deviation of the antibody concentration at time t, reflecting the degree of dispersion of the immune levels among individuals;

[0148] S63. Calculate the change rate and change trend of the population antibody concentration within the time period t, and at the same time define the antibody concentration change acceleration A(P, t) to evaluate the dynamic trend;

[0149] S64. Define the visualization indicators of the dynamic immune characteristics based on the dynamic distribution model of the population antibody concentration and the change rate of the population antibody concentration, including the antibody concentration coverage rate C(t, P th ):

[0150]

[0151] Among them, C(t, P th ) represents the proportion of the population with an antibody concentration higher than the threshold P th at time t. The threshold P th is used to measure the number of people who have reached an effective immune barrier. f(P, t) is the normal distribution function of the antibody concentration, and P th represents the minimum antibody concentration level required for effective immunity;

[0152] S65. Use the dynamic distribution model of the population antibody concentration and the visualization analysis results to generate an optimized vaccination strategy.

[0153] Example 1:

[0154] To verify the feasibility and effectiveness of the method for dynamically evaluating the antibody concentration of the population based on neural network simulation calculation in the present invention, an actual scenario of dynamically evaluating the antibodies of the population after vaccination is selected in this embodiment. The ability of the present invention to solve the problems of the prior art is fully demonstrated by detailed descriptions of the application scenario, data processing flow, implementation process of the dynamic evaluation method, and comparative experiment results below.

[0155] In a certain vaccination program in 2024, the health department of a certain city monitored the changes in antibody concentration after vaccination of 300,000 residents. The residents covered multiple age groups (18 to 65 years old), the vaccine type was two doses of inactivated vaccine, and the vaccination interval was 28 days. The health department hopes to model and evaluate the dynamic changes in antibody concentration of the population after vaccination through the method of the present invention, so as to predict the maintenance time of the herd immunity barrier and determine the optimal time node for booster vaccination.

[0156] To verify the effectiveness of the present invention, two randomly selected groups of samples were compared: one group was the samples analyzed by traditional static statistical methods (about 3,000 people), and the other group was the samples dynamically evaluated by the method of the present invention (about 3,000 people). Both groups of samples covered vaccination records, serum antibody concentration data, individual immune history, and biological experiment data.

[0157] In the embodiment, the basic data of the participating residents was first collected, including vaccination records (vaccination time, vaccine type), serum antibody concentration detection data (sampling and detection 7 days, 14 days, 30 days, 60 days, 90 days, and 180 days after each person's vaccination), individual immune history (previous disease history, autoimmune disease records), and biological experiment data (neutralizing antibody level and T cell activity). The data was recorded and sorted through multiple samplings and detection systems of the health department.

[0158] In the data preprocessing, the original data set was cleaned, and the entries with null values or outliers (records of individuals who interrupted vaccination or did not complete all detections) were removed. For the antibody concentration data, a biologically reasonable range (0 - 5000 AU / ml) was defined, and the outliers outside this range were removed. After cleaning, a total of 296,000 valid data were retained, and 3,000 of them were used for the training of the method of the present invention.

[0159] Subsequently, the method of the present invention was used to perform dynamic modeling on the above data. The multi-modal feature dataset (including antibody concentration features, vaccination features, individual immune history features, and biological experiment features) was input into a segmented neural network model. The model was divided into sub-networks for the rapid growth stage, stable stage, and decay stage. Each sub-network was used to model and simulate the antibody concentration change characteristics in different stages. Historical data of antibody concentration was used as the label for supervised learning during model training. The weighted mean square error was used as the loss function, and the model parameters were optimized through the backpropagation algorithm.

[0160] In the prediction stage, the multi-modal data of the population to be evaluated was input into the trained segmented neural network model to generate a dynamic prediction result of the individual antibody concentration over time. By aggregating the individual prediction results, a dynamic distribution model of the population antibody concentration was further constructed, and the average value and change rate of the population antibody concentration in different time periods were calculated.

[0161] To verify the superiority of the method of the present invention, a comprehensive comparison was made with traditional static statistical methods. The following Table 1 shows the key data and comparison results of the two methods in the dynamic evaluation of antibody concentration:

[0162] Table 1 Comparison Results of Traditional Method and the Method of the Present Invention in the Dynamic Evaluation of Antibody Concentration

[0163]

[0164] As can be seen from the data in Table 1 above, the traditional method mainly relies on single detection data for statistical analysis of antibody concentration, making it difficult to capture the dynamic changes of antibody concentration. The prediction time range is limited to within 90 days after vaccination and the mean absolute error is relatively high. In contrast, the method of the present invention, through dynamic modeling, fully integrates multi-modal data, significantly improves the prediction accuracy and time range of the model, and also greatly enhances the calculation efficiency.

[0165] In addition, through the dynamic distribution model, the present invention further reveals the temporal evolution law of the population immune level. 30 days after vaccination, the average value of the predicted population antibody concentration is 2200 AU / ml. 180 days after vaccination, it drops to 850 AU / ml, and the change rate is -20 AU / ml / day. Through dynamic visualization charts, the critical point for maintaining the immune barrier is determined to be 150 days after vaccination, and it is recommended that the booster vaccination be administered between 150 and 180 days after vaccination.

[0166] Through this embodiment, the present invention effectively solves the problems of inaccurate modeling of the dynamic changes of antibody concentration, insufficient adaptability of population immune assessment, and low calculation efficiency in the prior art. The above experimental results and data verify the feasibility and superiority of the present invention in practical scenarios, and it has significant application value and promotion prospects.

[0167] The present invention adopts a segmented neural network model, and respectively designs a sub-network for modeling the rapid growth stage of antibody concentration, a sub-network for modeling the stable stage, and a sub-network for modeling the decay stage, to accurately model the non-linear dynamic characteristics of antibody concentration in different stages. Through segmented dynamic simulation calculation combined with time series data, the non-linear change trend of antibody concentration after vaccination can be accurately captured, including the complex laws of rapid generation, plateau fluctuation, and long-term decay.

[0168] Through the multi-modal data fusion technology, the present invention effectively integrates serum antibody concentration data, vaccination records, individual immune history, and biological experiment data into a unified multi-modal feature data set, and realizes dynamic data processing by combining timestamp marking. It has significant advantages in the processing of multi-modal data, not only improving the compatibility with different data sources, but also realizing the dynamic evaluation of large-scale population data through the efficient calculation of the deep neural network model.

[0169] By constructing a dynamic distribution model, the present invention generates the dynamic distribution characteristics of population antibody concentration within a time period by synthesizing the dynamic prediction results of individual antibody concentration, which can accurately reflect the evolution law of the overall immune level of the population. By introducing a dynamic distribution function and the change rate of antibody concentration, the quantitative analysis and trend prediction of the immune state of the population in different time periods are realized, and the dynamic change of population antibody concentration and the maintenance time of the immune barrier can be intuitively displayed, providing a scientific basis for optimizing the vaccination strategy.

[0170] The above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitution or change, and should be covered by the protection scope of the present invention.

Claims

1. A dynamic evaluation method for the antibody concentration of a population based on neural network simulation calculation, characterized in that, It includes the following steps: S1. Obtain a multi-modal immunology-related data set; S2. Preprocess the multi-modal immunology-related data set to generate a complete multi-modal feature data set; S3. Based on the multi-modal feature data set, construct a deep learning model, design a segmented neural network architecture, which is divided into a sub-network for modeling the rapid growth stage of antibody concentration, a sub-network for modeling the stable stage of antibody concentration, and a sub-network for modeling the decay stage of antibody concentration; S4. Train the segmented neural network model, and optimize the parameters of the segmented neural network model through the backpropagation algorithm during the training process; S5. Input the multi-modal immunology-related data of the population to be evaluated into the trained segmented neural network model, and perform segmented dynamic simulation calculations through the sub-network for modeling the rapid growth stage of antibody concentration, the sub-network for modeling the stable stage of antibody concentration, and the sub-network for modeling the decay stage of antibody concentration respectively, to generate the prediction results of the individual antibody concentration changing with time; S6. According to the prediction results of the individual antibody concentration changing with time, summarize the prediction data of multiple individuals, establish a dynamic distribution model of the overall population antibody concentration, and generate the dynamic immune distribution characteristics evolving with time; S7. Input the results of the dynamic distribution model of the overall population antibody concentration into the visualization module, display the stage change trend of the antibody concentration by generating dynamic change charts and analysis reports, and provide vaccine inoculation optimization strategy suggestions and population immune intervention plans in combination with the data output.

2. The dynamic evaluation method for the crowd antibody concentration based on neural network simulation calculation according to claim 1, wherein The S1 includes the following steps: S11. Obtain population serum antibody concentration data, obtain vaccination records, obtain individual immune history data, and obtain biological experiment data; S12. Combine population serum antibody concentration data, vaccination records, individual immune history data, and biological experiment data to construct a multimodal immunology-related dataset D m : D m = {(d ai , d vi , d hi , d bi ) | i = 1, 2, …, n}; Among them, d ai represents the serum antibody concentration data of the i-th individual, and d vi represents the vaccination information of the i-th individual, including the type, dose, vaccination time, and vaccination interval of the vaccine, and d hi represents the immunological history data of the i-th individual, including past infection history, autoimmune disease records, and related genetic information, and d bi represents the immune-related data obtained by experimental analysis of the i-th individual, including neutralizing antibody level, T cell activity index, and antigen-specific reaction intensity; S13. Store the multi-modal immunology-related data set D m in a structured data form, and perform timestamp marking t i on each modal data respectively. The timestamp marking represents the specific time point of data collection, and the multi-modal immunology-related data set after time marking is obtained: D′ m = {(d ai , d vi , d hi , d bi , t i ) | i = 1, 2, …, n}}。 3. A method for dynamically evaluating the antibody concentration of a population based on neural network simulation calculation according to claim 1, characterized in that, The S2 includes the following steps: S21. Clean the multi-modal immunology-related dataset D' m by removing data with null values, duplicate values, or obvious error values in incomplete data entries. Error values include serum antibody concentration values outside the biologically reasonable range, invalid vaccination records, or invalid timestamp data. After data cleaning, a cleaned multi-modal immunology-related dataset is generated; S22. Perform outlier detection on the cleaned multimodal immunology-related datasets. Calculate the mean and standard deviation of each modality feature data and remove outliers that exceed the range [μ f -kσ f ,μ f +kσ f ], where k is the deviation coefficient of outlier judgment, and the characteristic data include antibody concentration, vaccination time, immunization history and biological experimental data, to obtain a multimodal immunology-related data set after removing outliers; S23. Perform standardization processing on the multi-modal immunology-related data set after removing outliers, and normalize all modal feature data to a unified range [0,1]; S24. Extract features from the standardized multi-modal immunology-related dataset using an autoencoder, perform feature dimensionality reduction on the high-dimensional modal features, and obtain the multi-modal feature dataset F after dimensionality reduction m : F m = {(f ai , f vi , f hi , f bi ) | i ∈ N a}; Among them, f ai , f vi , f hi and f bi respectively represent the antibody concentration characteristics, vaccination characteristics, immune history characteristics, and biological experiment characteristics after dimensionality reduction, and N a represents the number of valid individuals finally retained after data cleaning, outlier removal, and feature standardization processing.

4. A dynamic evaluation method for crowd antibody concentration based on neural network simulation calculation according to claim 1, characterized in that The S3 includes the following steps: S31. Based on the multimodal feature dataset F m Assign each feature entry to the corresponding time period according to the timestamp to generate a phased input feature set F t m : Among them, t∈T represents the time label of different time periods, corresponding to the dynamic change stage of antibody concentration; To integrate the stage input feature set, define the comprehensive input feature: Among them, is the comprehensive input feature for dynamic modeling, directly representing the dynamic characteristics of the antibody concentration of the i-th individual at time period t, w a , w v , w h , w b are the weights of the antibody concentration feature, vaccination feature, immune history feature, and biological experiment feature respectively, reflecting the importance of different features for dynamic evaluation; S32. Construct a segmented neural network model M, which is divided into the following three sub-networks: Rapid growth phase modeling sub-network M g , fitting the non-linear dynamic characteristics of the rapid increase in antibody concentration in the initial stage after vaccination; Stable-phase Modeling Sub-network M s , fitting the fluctuation law of antibody concentration after reaching a stable level in the middle stage; Decay phase modeling sub-network M d , fitting the changing trend of the antibody concentration gradually decreasing over time in the long term.

5. A method for dynamically evaluating the antibody concentration of a population based on neural network simulation calculation according to claim 1, characterized in that, The S4 includes the following steps: S41. Based on the complete multi-modal feature dataset and the comprehensive input features Use the comprehensive input features corresponding to time period t as the training input of the segmented neural network model, and combine the historical antibody concentration change data as the supervised learning label to form a training dataset: Among them, represents the measured value of the actual antibody concentration of the i-th individual at time period t; S42. Train the segmented neural network model through supervised learning, and use the weighted mean square error as the loss function: Among them, is the actually measured antibody concentration value, which is compared with the predicted value to optimize the model. λ is the regularization coefficient, which is used to prevent the model from overfitting and ensure the generality of the piecewise simulation calculation. and is the second norm of the weight matrix; S43. Optimize the parameters of the segmented neural network model through the backpropagation algorithm, including the weight matrices of the input layer and the hidden layer and the bias vectors The gradient calculation formula for weight update is as follows: Among them, and represent the gradients of the weights of the input layer and the hidden layer respectively, represents the gradient of the bias of the input layer, is the intermediate feature of the hidden layer, and the gradient is used to update the model parameters to minimize the loss function L; S44. Update the parameters of the segmented neural network model using the gradient descent algorithm; S45. Repeat the steps of S41 to S44 until the loss function L converges to a preset threshold or reaches the set maximum number of iterations. When converging, the parameters of the segmented neural network model can fit the historical dynamic change of antibody concentration, and finally form an optimized segmented neural network model.

6. The dynamic evaluation method for the crowd antibody concentration based on neural network simulation calculation according to claim 1, wherein The S5 includes the following steps: S51. Input the multi-modal immunology-related data set of the population to be evaluated into the trained segmented neural network model; S52. Use a segmented neural network model to perform segmented dynamic simulation calculations on the input features of the population to be evaluated, and generate predicted antibody concentration values for time period t through a rapid growth phase modeling sub-network, a stable phase modeling sub-network, and a decay phase modeling sub-network respectively Among them, represents the predicted antibody concentration value of the i-th individual at time period t, T g ,T s ,T d are the time ranges corresponding to the rapid growth stage, stable stage, and decay stage respectively, M g ,M s ,M d represent the modeling sub-networks for the rapid growth stage, stable stage, and decay stage respectively; The calculation result of each sub-network is determined by the comprehensive input feature and the trained network parameters: Among them, is the intermediate feature of the hidden layer, and σ is the activation function; Predictive value of time period for each individual Summarize to generate the overall dynamic prediction result of the i-th individual Among them, is the predicted result of the overall antibody concentration of the i-th individual, w t is the time period weight, which is used to balance the prediction contributions of different stages.

7. A method for dynamically evaluating the antibody concentration of a population based on neural network simulation calculation according to claim 1, characterized in that The S6 includes the following steps: S61. Based on the dynamic prediction results of individual antibody concentrations Calculate the average value P(t) of the overall antibody concentration of the population corresponding to the time period t: Among them, P(t) represents the average antibody concentration of the population within the time period t, reflecting the overall immune level, N p is the total number of individuals in the population to be evaluated, is the individual weight; S62. Construct a dynamic distribution model M of population antibody concentration using the time series P(t) within the set of time periods T P (T), and define the distribution function f(P, t) of antibody concentration within the time period t: M P (T) = {P(t) | t ∈ T}; Among them, M P (T) represents the dynamic distribution model of the overall antibody concentration of the population over time. The normal distribution function f(P,t) of the antibody concentration within the time period t is used to characterize the distribution characteristics of the population antibody concentration at time t. P(t) is the mean antibody concentration at time t, and σ t is the standard deviation of the antibody concentration at time t, reflecting the degree of dispersion of the immune levels among individuals; S63. Calculate the change rate and change trend of the population antibody concentration within the time period t, and at the same time define the antibody concentration change acceleration A(P,t) to evaluate the dynamic trend; S64. Define a visualization index of dynamic immunity characteristics based on the dynamic distribution model of population antibody concentration and the change rate of population antibody concentration, including antibody concentration coverage C(t, P th ): Among them, C(t, P th ) represents the proportion of the population with antibody concentration higher than the threshold P at time t th , and the threshold P th is used to measure the number of people reaching the effective immune barrier. f(P, t) is the normal distribution function of antibody concentration, and P th represents the minimum antibody concentration level required for effective immunity; S65. Generate an optimized vaccination strategy using the dynamic distribution model of population antibody concentration and the visualization analysis results.

Images