WO2023229239A1 - Method for predicting and analyzing side effects of vaccine by using artificial intelligence learning model based on vaccine subject variable information, and apparatus therefor - Google Patents

Abstract

An operation method for a vaccine side effect prediction and analysis apparatus, according to an embodiment of the present invention, comprises the steps of: acquiring subject variable information about vaccine side effect prediction and analysis subjects; acquiring side effect variable information corresponding to the subject variable information; acquiring an estimated vaccine side effect classification model and probability information by inputting the subject variable information and the side effect variable information into a pre-constructed vaccine side effect variable learning-based artificial intelligence model; and outputting vaccine side effect prediction and analysis information on the basis of the estimated vaccine side effect classification model and the probability information.

Description

Vaccine side effect prediction analysis method and device using an artificial intelligence learning model based on vaccine subject variable information

The present invention relates to a method and device for predicting and analyzing vaccine side effects. More specifically, the present invention relates to a method and device for predicting and analyzing vaccine side effects using an artificial intelligence learning model based on vaccine subject variable information.

Currently, new types of new viruses that are difficult to predict are emerging around the world, and various mutations are also emerging, resulting in a continuous shortage of preventive vaccine supply. As a result, rapid vaccine development is continuously required.

In particular, coronavirus disease 19 (COVID-19) is a respiratory infectious disease that first broke out in Wuhan, China in December 2019 and has since spread around the world. The World Health Organization (WHO) announced on January 9, 2020, The pathogen was confirmed on the 1st, revealing that the cause of the pneumonia was a new type of coronavirus (SARS-CoV-2, named on February 11 by the International Committee on Taxonomy of Viruses). It is known that COVID-19 is transmitted when an infected person's droplets penetrate the respiratory tract or the mucous membranes of the eyes, nose, and mouth. If infected, an incubation period of approximately 2 to 14 days (estimated) is followed by a fever (37.5 degrees Celsius). ) and respiratory symptoms such as coughing or difficulty breathing, and pneumonia are the main symptoms, but cases of asymptomatic infection are also rare.

However, the rapid development of vaccines is also accompanied by concerns about their side effects. In the case of the recent COVID-19 vaccine, chronic disease patients such as diabetics and hyperlipidemia patients, depending on the individual's underlying disease, are highly concerned about the risk of vaccine side effects.

In contrast, current technology remains at the level of testing biological responses to side effects after vaccination, which involves collecting biological samples and predicting vaccine side effects based on the correlation between the expression level of RNA and the expression level of inflammatory cytokines. As a method, it has the disadvantage of being very cumbersome and requiring a lot of time and money.

The present invention was designed to solve the problems described above. It can quickly predict the type and frequency of side effects of various vaccines according to the individual's disease or variable characteristics and further provide customized recommendations for the type of vaccine with a low risk of side effects for each individual. The purpose is to provide a vaccine side effect prediction analysis method and device using an artificial intelligence learning model based on vaccine subject variable information that can quickly and accurately predict vaccine side effects for each individual.

A method according to an embodiment of the present invention for solving the problems described above is a method of operating a vaccine side effect prediction and analysis device, comprising: acquiring subject variable information of a person subject to vaccine side effect prediction analysis; Obtaining side effect variable information corresponding to the subject variable information; Inputting the subject variable information and the side effect variable information into a pre-built vaccine side effect variable learning-based artificial intelligence model to obtain an estimated vaccine side effect classification model and probability information; and outputting vaccine side effect prediction analysis information based on the estimated vaccine side effect classification model and probability information.

An apparatus according to an embodiment of the present invention for solving the problems described above is a vaccine side effect prediction and analysis device, comprising: a subject variable information processor that acquires subject variable information of a subject of vaccine side effect prediction analysis; a side effect variable information processing unit that acquires side effect variable information corresponding to the subject variable information; An analysis processing unit that inputs the subject variable information and the side effect variable information into a pre-built vaccine side effect variable learning-based artificial intelligence model to obtain an estimated vaccine side effect classification model and probability information; and a prediction result output unit that outputs vaccine side effect prediction analysis information based on the estimated vaccine side effect classification model and probability information.

According to an embodiment of the present invention, the subject variable information and side effect variable information of the vaccine side effect prediction analysis subject are input into a pre-built vaccine side effect variable learning-based artificial intelligence model to obtain an estimated vaccine side effect classification model and probability information, It is possible to provide predictive analysis information for vaccine side effects based on an estimated vaccine side effect classification model and probability information, and thus, predictive analysis of vaccine side effects can be performed.

Accordingly, according to an embodiment of the present invention, it is possible to provide a means for testing vaccine side effects that enables simple and accessible initial diagnosis without a separate biological test, and the types and frequencies of side effects of various vaccines according to individual diseases or variable characteristics. It is possible to provide a vaccine side effect prediction and analysis method and device using an artificial intelligence learning model based on vaccine subject variable information that can quickly predict and further recommend the type of vaccine with a low risk of side effects for each individual.

Figure 1 is a block diagram specifically illustrating a vaccine side effect prediction and analysis device according to an embodiment of the present invention.

Figure 2 is a flowchart for explaining the operation method of the vaccine side effect prediction and analysis device according to an embodiment of the present invention.

Figures 3 and 4 are flowcharts to explain the process of building an artificial intelligence learning model according to an embodiment of the present invention.

5 to 10 are graphs showing performance test analysis results of an artificial intelligence learning model according to an embodiment of the present invention.

The following merely illustrates the principles of the invention. Therefore, those skilled in the art will be able to invent various devices and methods that embody the principles of the present invention and are included in the concept and scope of the present invention, although not explicitly described or shown herein. In addition, all conditional terms and examples listed herein are, in principle, expressly intended only for the purpose of enabling the concept of the invention to be understood, and should be understood not as limiting to the examples and states specifically listed as such. do.

Additionally, it is to be understood that any detailed description reciting the principles, aspects and embodiments of the invention, as well as specific embodiments, is intended to encompass structural and functional equivalents thereof. In addition, these equivalents should be understood to include not only currently known equivalents but also equivalents developed in the future, that is, all elements invented to perform the same function regardless of structure.

Accordingly, for example, the block diagrams herein should be understood as representing a conceptual view of an example circuit embodying the principles of the invention. Similarly, all flow diagrams, state transition diagrams, pseudo-code, etc. are understood to represent various processes that can be substantially represented on a computer-readable medium and are performed by a computer or processor, whether or not the computer or processor is explicitly shown. It has to be.

Additionally, the clear use of terms such as processor, control, or similar concepts should not be construed as exclusively referring to hardware capable of executing software, and should not be construed as referring exclusively to hardware capable of executing software, including, without limitation, digital signal processor (DSP) hardware, and ROM for storing software. It should be understood as implicitly including ROM, RAM, and non-volatile memory. Other hardware for public use may also be included.

The above-described purpose, features and advantages will become clearer through the following detailed description in conjunction with the accompanying drawings, and accordingly, those skilled in the art will be able to easily implement the technical idea of the present invention. There will be. Additionally, in carrying out the present invention, if it is determined that a detailed description of known techniques related to the present invention may unnecessarily obscure the gist of the present invention, the detailed description will be omitted.

The terms used in this application are only used to describe specific embodiments and are not intended to limit the invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this application, terms such as “comprise” or “have” are intended to designate the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, but are not intended to indicate the presence of one or more other features. It should be understood that this does not exclude in advance the possibility of the existence or addition of elements, numbers, steps, operations, components, parts, or combinations thereof.

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the attached drawings. In order to facilitate overall understanding when describing the present invention, the same reference numerals are used for the same components in the drawings, and duplicate descriptions for the same components are omitted.

Figure 1 is a block diagram specifically illustrating a vaccine side effect prediction and analysis device according to an embodiment of the present invention.

Referring to FIG. 1, the vaccine side effect prediction and analysis device 100 according to an embodiment of the present invention includes a vaccine subject variable information processing unit 110, a side effect variable information processing unit 120, an analysis processing unit 130, and an artificial intelligence learning-based It includes a model building unit 140 and a prediction result output unit 140.

First, the vaccine side effect prediction and analysis device 100 described in this specification includes personal computers (PCs), laptop computers, mobile phones, tablet PCs, personal digital assistants (PDAs), PMP (Portable Multimedia Player), etc. may be included. However, the present invention is not limited to the above device classification and may also include devices such as a server system that can enhance and expand data processing, storage, and management functions.

In addition, the vaccine side effect prediction and analysis device 100 according to an embodiment of the present invention acquires subject variable information and side effect variable information of the analysis subject from an external mobile terminal, a server, or directly input user input information, The subject variable information and the side effect variable information are input into a pre-built artificial intelligence model based on vaccine side effect variable learning to obtain a vaccine side effect classification prediction model and probability information, and a vaccine based on the vaccine side effect classification prediction model and probability information. It may be an analysis device that outputs side effect analysis information.

Here, the vaccine side effect analysis information output from the vaccine side effect prediction and analysis device 100 may include recommendation guidance information corresponding to the predicted vaccine side effect, and may be displayed through the above-mentioned mobile terminal, server, or separate display device. can be printed. For example, the recommended guidance information corresponding to the predicted vaccine side effects may be provided through a system such as a server provided by a health care center, or may be provided through a mobile terminal, and for this purpose, the vaccine side effect prediction analysis device 100 , can be connected to a mobile terminal or server system through a wired/wireless network.

Devices or terminals connected to the network can communicate with each other through a preset network channel and may be equipped with a communication module that supports each protocol for communication.

Here, the network includes Local Area Network (LAN), Wide Area Network (WAN), Value Added Network (VAN), Personal Area Network (PAN), and Mobile Network (Mobile Area Network). It can be implemented as any type of wired/wireless network, such as a radio communication network or satellite communication network.

By processing the vaccine side effect prediction analysis data of the vaccine side effect prediction and analysis device 100, it is possible to predict vaccine side effects just by inputting the subject's variable information without a separate biological test. To implement this, see Figure 1. , First, the vaccine subject variable information processing unit 110 can perform information acquisition processing to obtain the subject variables of the analysis subject, and the side effect variable information processing unit 120 can perform the acquisition processing of variable information related to side effects. there is.

Here, the subject variable information of the subject obtained from the vaccine subject variable information processing unit 110 is information that can be input in relation to the subject, for example, the subject's gender information, body information (e.g., age information, weight information , height information), disease information, disease history information, health status information, lifestyle information (e.g., drinking status information and drinking frequency information, smoking status information and smoking frequency information, eating habits information, exercise status information, and exercise frequency information etc.), medication information (e.g., information on the type of drug consumed, dosage information and frequency information, etc.), and biomarker information, etc., and preferably, gender information and age information are essential. may be included. This can be preprocessed as subject variable information for predicting a vaccine side effect model.

In addition, the side effect variable information obtained from the side effect variable information processing unit 120 includes vaccine manufacturer information, vaccine type information, vaccination order information, vaccination site information, vaccination route information, information on whether side effects that appear after vaccination have been completely cured, Information on the period until side effects occurred after vaccination, information on whether life was threatened due to side effects after vaccination, information on whether people visited the hospital due to side effects after vaccination, information on the period of stay in the hospital due to side effects after vaccination , It may include at least one of information on whether a disability occurred due to a side effect after vaccination, and preferably, vaccine manufacturer information, vaccine type information, and vaccination order information may be necessarily included. This can be preprocessed as side effect variable information for predicting a vaccine side effect model.

And, the analysis processing unit 130 inputs the vaccine subject variable information and the side effect variable information into an artificial intelligence model based on vaccine side effect variable learning pre-built by the learning process of the artificial intelligence learning-based model construction unit 140. , predicted vaccine side effect classification model information and probability information can be obtained.

Additionally, the prediction result output unit 140 may output vaccine side effect prediction analysis information based on the vaccine side effect classification model information and probability information. Depending on the prediction result, the output unit 140 may output various types of vaccine guidance interfaces including the vaccine side effect prediction analysis information using a mobile terminal, server, or user display device.

In addition, the artificial intelligence learning-based model construction unit 140 configures the vaccine subject variable information and the side effect variable information as input values when constructing an artificial intelligence model based on learning the vaccine subject variable information and the side effect variable information. , Artificial intelligence-based machine learning can be performed using a learning data set that consists of an appropriate vaccine side effect model and vaccine side effect model classification information and probability information representing the probability as output values.

To this end, the artificial intelligence learning-based model construction unit 140 may preprocess the vaccine subject variable information and the side effect variable information according to a standardization algorithm.

Here, the artificial intelligence learning-based model construction unit 140 uses the vaccine subject variable information and the side effect variable information using a tree-based pipeline optimization tool ( Optimization can be done using TPOT (Tree-based Pipeline Optimization Tool).

In addition, the artificial intelligence learning-based model construction unit 140 selectively combines one or more of the vaccine subject variable information and the side effect variable information and repeatedly processes the model performance test, thereby creating the vaccine subject variable information and the Among the side effect variable information, a plurality of learning variables without performance degradation can be determined as the final model input variables. For example, five variables selected from the vaccine subject variable information can be determined as model input variables, and depending on the combination, various input variable model combinations, such as six or seven, can be configured.

In addition, the artificial intelligence learning-based model construction unit 140 sets model input variables obtained from the vaccine subject variable information and the side effect variable information as input values, and the subject's vaccine side effect model classification information corresponding to the input values. By setting the probability information as the output value, artificial intelligence model learning based on gradient boosting learning can be performed.

More specifically, examples of input variables and output variables used to build a model in the artificial intelligence learning-based model building unit 140 according to an embodiment of the present invention are as follows.

input variable

type

gender age vaccine manufacturer Number of vaccinations vaccination site Vaccination route Whether or not side effects that occurred after vaccination have been cured Period until side effects occur after vaccination Whether you felt life-threatening due to side effects after vaccination Whether you visited the hospital due to side effects after vaccination Length of stay in hospital due to side effects after vaccination Whether disability occurred due to side effects after vaccination

categorical continuous type categorical continuous type categorical categorical categorical continuous type categorical categorical continuous type categorical

output variable	type
Serious side effects related to the central nervous system Serious respiratory side effects Serious heart-related side effects Serious blood-related side effects Other (non-central nervous/respiratory/heart/digestive/blood) serious side effects Dead or not	categorical

As disclosed in Tables 1 and 2 above, type information can be assigned to each variable item, and the output variable includes side effect model classification determined according to each input variable and probability information that the subject falls into each side effect model classification. It can be included. Here, probability information may be configured as probability level information.

.And, the artificial intelligence model based on gradient boosting learning, which is constructed according to these input and output variables, may be a model learned using a parallel processing-based extreme gradient boosting (XGBoost, eXtreme Gradient Boosting) algorithm.

In addition, in the artificial intelligence model based on gradient boosting learning, training scale weights (scale_pos_weight) for each level of vaccine side effect model classification information and probability information may be appropriately set to adjust classification imbalance (imbalance classification). More detailed learning model construction and operation methods will be described in more detail later.

Figure 2 is a flowchart for explaining the operation method of the vaccine side effect prediction and analysis device according to an embodiment of the present invention, and Figures 3 and 4 are flowcharts for explaining the construction process of an artificial intelligence learning model according to an embodiment of the present invention. am.

First, referring to FIG. 2, the vaccine side effect prediction and analysis device 100 according to an embodiment of the present invention preprocesses the vaccine side effect variables and vaccine subject variables as parameters of the learning model to obtain the vaccine side effect learning variables and vaccine subject learning variables. Obtain (S101).

In addition, the vaccine side effect prediction analysis device 100 sets the vaccine side effect learning variable and the subject learning variable as input values, sets the subject's estimated vaccine side effect classification model and probability corresponding thereto as output values, and adjusts the weight. Build an artificial intelligence model based on gradient boosting learning (S103).

Thereafter, the vaccine side effect prediction and analysis device 100 preprocesses the subject input information of the analysis subject and configures subject variable information and side effect variable information of the analysis subject (S105).

Then, the vaccine side effect prediction and analysis device 100 inputs the subject variable information and side effect variable information of the analysis subject into the constructed artificial intelligence model to obtain an estimated vaccine side effect classification model and probability information of the analysis subject (S107).

Thereafter, the vaccine side effect prediction and analysis device 100 provides vaccine side effect guidance information tailored to the analysis subject using a vaccine side effect classification model and probability information (S107).

The vaccine side effect prediction and analysis device 100 can provide various customized vaccine guidance services based on vaccine side effect analysis information to a mobile terminal or server system.

3 and 4 show in more detail the construction and analysis process of an artificial intelligence learning model according to an embodiment of the present invention. First, the learning-based artificial intelligence model according to an embodiment of the present invention is the above-mentioned As such, it can be pre-built using the parallel processing-based extreme gradient boosting (XGBoost, eXtreme Gradient Boosting) algorithm.

Here, the XGBoost algorithm used for learning according to an embodiment of the present invention is a machine learning algorithm that uses gradient boosting tree algorithm technology, and combines a plurality of tree models generated according to learning variables. This is an algorithm known to make the final decision by correcting the errors of the previous tree model when creating tree models (this is called boosting), and at this time, using the gradient descent algorithm to minimize the loss. In addition, XGBoost can handle missing values using a sparsity-aware split finding method and has the advantage of accelerating calculation speed using GPU.

And, in order to use this Accordingly, it may be composed of variable values classified for each level.

For example, the estimated vaccine side effect classification model and probability information may fall into any one of the first level, less than 30%, the second level, less than 60%, and the third level, less than 100%, Accordingly, any one of the three levels can be selectively assigned. And, the output result may be determined as a positive or negative result according to the estimated vaccine side effect classification model and probability information. In addition, the threshold for each level for a positive or negative result of the level may be set in advance, and the threshold may be set differently for each subgroup, which will be described later.

Additionally, for learning such estimated vaccine side effect classification model and probability information, at least some of the learning variables according to an embodiment of the present invention may be normalized.

More specifically, for example, among subject variable information according to an embodiment of the present invention, in the case of age information, weight information, etc., standardization processing may be performed to remove the mean and readjust to the unit variance. The standardization process for this can be explained as Equation 1.

Here, z represents the normalized input value, x is the original input value, u is the average value of the input values, and s represents the standard deviation of the input values.

Accordingly, the artificial intelligence learning-based model construction unit 140 according to an embodiment of the present invention sets the subject variable information and side effect variable information preprocessed according to standardization using Equation 1 as input values, and the above-mentioned estimated vaccine side effects It constructs a learning data set that sets the classification model and probability information prediction level as output values, performs a learning process based on the gradient boosting tree algorithm using actual clinical data, and repeatedly improves the performance of result prediction to arrive at the optimal model. can be built.

In addition, here, the artificial intelligence learning-based model construction unit 140 repeatedly processes model performance tests using the subject learning variables and side effect learning variables, and selects a plurality of learning variables that show the best performance among the subject learning variables and side effect learning variables. You can perform input variable selection optimization to determine these as model input variables.

For example, the artificial intelligence learning-based model building unit 140 according to an embodiment of the present invention can select feature variables from the subject learning variables and side effect learning variables, and for this purpose, the artificial intelligence learning-based model building unit ( 140) removes the least important variable among the model variables according to the feature importance value calculated from the You can remove any variables that are not present and perform the performance test again. Additionally, the artificial intelligence learning-based model construction unit 140 repeats this process until there is no performance degradation when compared to the initial model, thereby reducing the number of input variables required for input. According to a test according to an embodiment of the present invention, it was confirmed that there was no deterioration in model performance from the initial 12 input variables to 7.

Meanwhile, the artificial intelligence learning-based model building unit 140 according to an embodiment of the present invention converts the hyperparameters of XGBoost into the Tree-based Pipeline Optimization Tool (TPOT), an automated artificial intelligence tool. By optimizing, higher performance can be achieved. More specifically, TPOT is an automated machine learning tool for Python that uses genetic programming to optimize machine learning pipelines.

In particular, the hyperparameters on which optimization was performed according to an embodiment of the present invention are shown in Table 3 below.

Here, in particular, the scale positive weight (scale_pos_weight) variable is a variable that adjusts the imbalance when the output ratio is usually unbalanced in a binary classification model. It is a value adjusted by the output ratio of the training data, and is usually a large ratio. A value may be set by dividing the number of groups by the number of a small percentage of groups.

Accordingly, the artificial intelligence learning-based model building unit 140 according to an embodiment of the present invention can set the learning model for each level and its learning parameters differently, and in particular, adjust the scale positive weight (scale_pos_weight) variable to The imbalance between positive/negative output ratios for each level can be adjusted in advance.

In addition, the learning model according to an embodiment of the present invention can be classified into one or more subgroups according to the characteristics of the input variables, and a different estimated vaccine side effect classification model can be determined corresponding to each subgroup, and corresponding Probability information can be determined, and by assigning a prediction level threshold corresponding to each probability information, the predictability of side effects can be processed to more accurately predict by model variation. Here, the subgroup may be divided into at least four groups depending on the probability level for each side effect model, and this is used to correct problems that may cause model prediction performance to deteriorate due to individual differences.

Once the model is built in this way, the analysis process as shown in FIGS. 3 and 4 can be performed.

First, the analysis processing unit 130 determines the subgroup information classification of the estimated vaccine side effect classification model and sets a threshold for each model according to the subgroup information classification (S203).

Here, the subgroup information classification may be classified and processed in the analysis processing unit 130 according to the estimated vaccine side effect classification model and probability information, as described above.

Figure 4 illustrates the classification process processed in step S203 in more detail. As shown in Figure 4, at least six subgroup types can be set according to information about the input value. For example, referring to Figure 4, subgroup information classification can be set according to whether the age information is 65 years or older, gender is male, and vaccine manufacturer is Pfizer, and each of the six types accordingly. Subgroups may be formed.

In addition, the analysis processing unit 130 may further perform specialized service processes to guide each subgroup classified by each estimated vaccine side effect classification model and each reference value of probability information. For example, the analysis processing unit 130 performs a first service process that provides action information in the event of respiratory-related side effects in men over 65 years of age, depending on individual characteristics and side effect subgroup classification, or performs a first service process that provides action information when a respiratory-related side effect occurs in a man aged 65 or older, or the heart rate when receiving the Pfizer vaccine. When a related side effect occurs, a process such as performing a second service process that provides action information may be performed.

And, referring again to FIG. 3, the analysis processing unit 130 performs standardization processing of the input variables (S205).

Here, the standardization process can be performed through Equation 1 described above, and as a result, standardization of input information is possible using the data used for learning.

Then, the analysis processing unit 130 applies the standardized input information to the pre-trained artificial intelligence model, and according to the positive or negative information for each estimated vaccine side effect classification model and probability information level obtained as output information, the estimated vaccine Determine the side effect classification model and probability information (S209).

Thereafter, the analysis processing unit 130 generates information on the occurrence of vaccine side effects based on the threshold set corresponding to the previously determined subgroup type, the estimated vaccine side effect classification model, and the level probability information (S211).

Here, when predicting the occurrence of vaccine side effects, the analysis processing unit 130 calculates the SHAP (SHapley Additive exPlanations) value and generates explanatory information explaining the contribution of each input variable corresponding to the prediction of the occurrence of vaccine side effects ( S213).

Accordingly, the analysis processing unit 130 can configure vaccine side effect analysis result information, including an estimated vaccine side effect classification model and probability information, and explanatory information related to the degree to which the input value contributed, and output it to the prediction result output unit 140. There is (S215).

For example, when the occurrence of vaccine side effects is predicted according to the estimated vaccine side effect classification model and probability information, the analysis processing unit 130 calculates the contribution of each input variable corresponding to the subject variable information and side effect variable information to SHAP (SHapley Additive). exPlanations) algorithm calculation, generate analysis information based on the contribution of each input variable, and include it in the vaccine side effect analysis information.

5 to 10 are graphs showing performance test analysis results of an artificial intelligence learning model according to an embodiment of the present invention.

Figures 5 and 10 are learning data, which are the results of learning and internal verification using 9,267 cases using the Vaccine Adverse Event Reporting System (VAERS) data, and are the results of selection applicable to the general population. Indicates the results of inspection and verification.

In the test analysis, the estimated vaccine side effect classification model was constructed to select an analysis algorithm for each side effect model using an artificial intelligence learning model according to an embodiment of the present invention, and was processed to selectively determine a model with excellent performance. . The artificial intelligence model used in the test analysis was selectively constructed from Decision Tree, Random Foreset, Etra Trees, LightGBM, It has been done.

side effect model	algorithm	AUC	F1-score	accuracy
Serious side effects related to the central nervous system	CatBoost	0.77	0.74	0.72
Serious respiratory side effects	CatBoost	0.94	0.88	0.88
Serious heart-related side effects	CatBoost	0.90	0.85	0.84
Serious blood-related side effects	LightGBM	0.69	0.69	0.65
Other (non-central nervous/respiratory/heart/digestive/blood) serious side effects	CatBoost	0.84	0.78	0.77
Dead or not	Extra Trees	0.90	0.84	0.83

The graphs shown in FIGS. 5 to 10 are ROC graphs showing the prediction performance for each model in Table 3, and the final performance result is that the AUC (Area under the Curve) records a maximum of 0.94, and the accuracy records a maximum of 0.94. By recording 0.88, it can be confirmed that reliable model construction and operation is possible according to the accumulation of learning data for each model. According to the vaccine side effect prediction analysis device 100 and its operation method according to the embodiment of the present invention, separate Even without biological testing, vaccine side effects can be quickly and easily predicted by simply entering subject variable information and side effect variable information.

Accordingly, the present invention enables prediction and monitoring of customized vaccine side effects depending on the type of vaccine or the type of disease of chronically ill patients, and not only reduces socioeconomic costs due to vaccine side effects, but also enables analysis using artificial intelligence. Through this, as learning data accumulates, more accurate services and device operations can be provided.

The method according to the present invention described above can be produced as a program to be executed on a computer and stored in a computer-readable recording medium. Examples of computer-readable recording media include ROM, RAM, CD-ROM, and magnetic tape. , floppy disks, optical data storage devices, etc.

The computer-readable recording medium is distributed in a computer system connected to a network, so that computer-readable code can be stored and executed in a distributed manner. And, functional programs, codes, and code segments for implementing the method can be easily deduced by programmers in the technical field to which the present invention pertains.

In addition, although preferred embodiments of the present invention have been shown and described above, the present invention is not limited to the specific embodiments described above, and the technical field to which the invention pertains without departing from the gist of the present invention as claimed in the claims. Of course, various modifications can be made by those skilled in the art, and these modifications should not be understood individually from the technical idea or perspective of the present invention.

Claims (15)

1. In the method of operating the vaccine side effect prediction analysis device,

   Obtaining subject variable information of the subject of vaccine side effect prediction analysis;

   Obtaining side effect variable information corresponding to the subject variable information;

   Inputting the subject variable information and the side effect variable information into a pre-built vaccine side effect variable learning-based artificial intelligence model to obtain an estimated vaccine side effect classification model and probability information; and

   Including the step of outputting vaccine side effect prediction analysis information based on the estimated vaccine side effect classification model and probability information.

   Method of operation of vaccine side effect prediction analysis device.
2. According to paragraph 1,

   The subject variable information of the subject includes at least one of age information, gender information, and disease information,

   The subject's side effect variable information includes at least one of vaccine manufacturer information and vaccine type information.

   Method of operation of vaccine side effect prediction analysis device.
3. According to paragraph 2,

   The estimated vaccine side effect classification model is,

   Containing at least one of the central nervous system-related serious side effect model, respiratory-related serious side effect model, heart-related serious side effect model, and blood-related serious side effect model.

   Method of operation of vaccine side effect prediction analysis device.
4. According to paragraph 1,

   Further comprising the step of building an artificial intelligence model based on learning the vaccine side effect variables,

   The construction step is,

   Preprocessing subject learning variables and side effect learning variables according to a standardization algorithm; and

   The preprocessed subject learning variables and side effect learning variables are set as input values, and the subject's estimated vaccine side effect classification model and probability information corresponding to the input values are set as output values to perform artificial intelligence model learning based on gradient boosting learning. containing the steps of

   Method of operation of vaccine side effect prediction analysis device.
5. According to clause 4,

   The artificial intelligence model based on gradient boosting learning is learned using the parallel processing-based extreme gradient boosting (XGBoost, eXtreme Gradient Boosting) algorithm.

   Method of operation of vaccine side effect prediction analysis device.
6. According to clause 4,

   The artificial intelligence model based on gradient boosting learning is,

   The scale positive weights of the training parameters corresponding to the estimated vaccine side effect classification model and probability information to adjust classification imbalance are set, respectively.

   Method of operation of vaccine side effect prediction analysis device.
7. According to clause 4,

   The construction step is,

   Further comprising the step of applying hyperparameters optimized with a Tree-based Pipeline Optimization Tool (TPOT) to the gradient boosting learning-based artificial intelligence model.

   Method of operation of vaccine side effect prediction analysis device.
8. According to clause 4,

   The preprocessing step is,

   Performing input variable selection optimization by repeatedly processing model performance tests using the subject learning variables and side effect learning variables to determine a plurality of learning variables without performance degradation among the subject learning variables and side effect learning variables as model input variables. containing steps

   Method of operation of vaccine side effect prediction analysis device.
9. According to paragraph 1,

   The step of outputting the vaccine side effect prediction analysis information,

   When vaccine side effects are predicted according to the estimated vaccine side effect classification model and probability information, the contribution for each input variable corresponding to the subject variable and the side effect variable is obtained according to the SHAP (SHapley Additive exPlanations) algorithm operation, and for each input variable Generating analysis information based on contribution and including it in the vaccine side effect prediction analysis information.

   Method of operation of vaccine side effect prediction analysis device.
10. In the vaccine side effect prediction analysis device,

    a subject variable information processing unit that acquires subject variable information of the subject of vaccine side effect prediction analysis;

    a side effect variable information processing unit that acquires side effect variable information corresponding to the subject variable information;

    An analysis processing unit that inputs the subject variable information and the side effect variable information into a pre-built vaccine side effect variable learning-based artificial intelligence model to obtain an estimated vaccine side effect classification model and probability information; and

    Comprising a prediction result output unit that outputs vaccine side effect prediction analysis information based on the estimated vaccine side effect classification model and probability information.

    Vaccine side effect prediction analysis device.
11. According to clause 10,

    The subject variable information of the subject includes at least one of age information, gender information, physical information, disease information, medication information, biomarker information, and lifestyle information,

    The subject's side effect variable information includes at least one of vaccine manufacturer information and vaccine type information,

    The estimated vaccine side effect classification model is,

    Containing at least one of the central nervous system-related serious side effect model, respiratory-related serious side effect model, heart-related serious side effect model, and blood-related serious side effect model.

    Vaccine side effect prediction analysis device.
12. According to clause 10,

    It further includes an artificial intelligence learning-based model construction unit that builds an artificial intelligence model based on vaccine side effect variable learning,

    The artificial intelligence learning-based model construction unit,

    Subject learning variables and side effect learning variables are preprocessed according to a standardization algorithm, the preprocessed subject learning variables and side effect learning variables are set as input values, and the subject's estimated vaccine side effect classification model and probability information corresponding to the input values are output values. Set to perform artificial intelligence model learning based on gradient boosting learning.

    Vaccine side effect prediction analysis device.
13. According to clause 12,

    The artificial intelligence model based on gradient boosting learning is learned using the parallel processing-based extreme gradient boosting (XGBoost, eXtreme Gradient Boosting) algorithm,

    According to the extreme gradient boosting algorithm, the training scale weight (scale_pos_weight) for each vaccine side effect classification model to adjust classification imbalance is set, respectively.

    Vaccine side effect prediction analysis device.
14. According to clause 12,

    The artificial intelligence learning-based model construction unit,

    Hyperparameters optimized with the Tree-based Pipeline Optimization Tool (TPOT) are applied to the gradient boosting learning-based artificial intelligence model,

    Performing input variable selection optimization by repeatedly processing model performance tests using the subject learning variables and side effect learning variables to determine a plurality of learning variables without performance degradation among the subject learning variables and side effect learning variables as model input variables. containing steps

    Vaccine side effect prediction analysis device.
15. According to clause 12,

    When a vaccine side effect is predicted according to the estimated vaccine side effect classification model and probability information, the analysis processing unit obtains the contribution of each input variable corresponding to the subject variable and the side effect variable according to SHAP (SHapley Additive exPlanations) algorithm operation, , generating analysis information based on the contribution of each input variable and including it in the vaccine side effect prediction analysis information.

    Vaccine side effect prediction analysis device.

Images