KR102530114B1 - Method, System, and Computer-Readable Medium for Predicting Side Effects of Drugs based on Similarity Measurement - Google Patents

Abstract

The present invention relates to a method, system, and computer-readable medium for predicting side effects of drugs based on similarity measurement, and more particularly, in order to predict the possibility of occurrence of a predicted side effect for a predicted drug, a predicted side effect is Derive the similarity between a plurality of existing drugs known to be It relates to a method, system, and computer-readable medium for predicting side effects of a drug based on similarity measurement, predicting the possibility of occurrence of a predicted target side effect for a drug.

Description

Method, System, and Computer-Readable Medium for Predicting Side Effects of Drugs based on Similarity Measurement

The present invention relates to a method, system, and computer-readable medium for predicting side effects of drugs based on similarity measurement, and more particularly, in order to predict the possibility of occurrence of a predicted side effect for a predicted drug, a predicted side effect is Derive the similarity between a plurality of existing drugs known to be It relates to a method, system, and computer-readable medium for predicting side effects of a drug based on similarity measurement, predicting the possibility of occurrence of a predicted target side effect for a drug.

Most of the new drug candidates developed in the pharmaceutical industry over the past decades have not been approved by the US Food and Drug Administration (FDA) or other countries because of the side effects that the drugs can cause, so R&D performance has been poor. In the drug development process, it is difficult to find activities in which a new drug candidate drug that may cause side effects and further cause drug development failure affects factors other than the target. As such, most of the fatal side effects are identified in preclinical or clinical trials, but some are also discovered in the monitoring process after drug approval. Uncertainty about the potential side effects of these new drugs is an important issue not only for pharmaceutical companies, but also for patients who may be at risk for their health.

On the other hand, conventional methods used to predict side effects of drugs are based on the assumption that similar drugs have similar properties in terms of chemical and biological properties, such as drug structures and drug targets. Therefore, the structure and target similarity between a specific drug and other drugs is calculated, and based on this, the side effects of a specific drug are predicted, or the side effects of a drug are predicted by considering the similarity of phenotypic information of the drug. methods have been presented.

However, conventional methods for predicting drug side effects by focusing on the chemical and biological properties of drugs, when administered to the human body, are subject to complex effects such as metabolic transformation or pharmacokinetic changes, so side effects of drugs Simply considering only the chemical and biological properties of a drug to predict the drug is difficult to predict and the accuracy of the prediction is inevitably low. Therefore, in order to predict side effects caused by drugs, in addition to considering only the structure of conventional drugs or drug targets, the similarity of drugs is measured in various aspects to further facilitate the prediction of the occurrence of side effects and improve the prediction accuracy. There is a need for research on new methods that can do this.
Specifically, among the conventional technologies in the above technical field, Korean Patent Publication No. 10-2022-0117729 discloses a configuration for detecting unknown side effects of drugs using a machine learning algorithm, but the corresponding In the case of the technology, as described above, it corresponds to the configuration of detecting the side effects of drugs using the minimum information included in the spontaneous adverse event report data, and measures the similarity of drugs in various aspects including the structure and target of drugs. Therefore, there is no conventional technique for predicting the side effects of drugs based on similarity measurement in various aspects, since it does not describe a configuration that facilitates the prediction of the possibility of side effects.

delete

Republic of Korea Patent Publication No. 10-2022-0117729 (2022.08.24.)

The present invention relates to a method, system, and computer-readable medium for predicting side effects of drugs based on similarity measurement, and more particularly, in order to predict the possibility of occurrence of a predicted side effect for a predicted drug, a predicted side effect is Derive the similarity between a plurality of existing drugs known to be It relates to a method, system, and computer-readable medium for predicting side effects of a drug based on similarity measurement, predicting the possibility of occurrence of a predicted target side effect for a drug.

In order to solve the above problems, the present invention is a method for predicting drug side effects, implemented by a computing device, for each of two or more categories between a drug to be predicted and a plurality of existing drugs known to have side effects to be predicted. Deriving a first similarity value; deriving first similarity representative information between the predicted target drug and the predicted target side effect by extracting a maximum value from the plurality of first similarity values for each category; deriving each second similarity value between the predicted target side effect and a plurality of existing side effects known for the predicted target drug; deriving second similarity representative information between the predicted target drug and the predicted target side effect by extracting a maximum value from the plurality of second similarity values; And inputting the first similarity representative information and the second similarity representative information to the learned inference model unit to derive information on the possibility of occurrence of the predicted side effect for the predicted drug; including, drug Provides a method for predicting side effects of

In one embodiment of the present invention, the step of deriving the first similarity value includes, as a first category, a correlation between one or more other drugs that the drug has an effect on or is affected by, and the drug to be predicted has an effect. The first similarity value for the first category between the correlation between one or more other drugs that give or are affected and the correlation between one or more other drugs that each of the plurality of existing drugs has an effect on or is affected by can be derived

In one embodiment of the present invention, the step of deriving the first similarity value includes, as a second category, node information about genes and proteins included in an interaction path between a plurality of target genes targeted by a drug, , A first similarity value for the second category between the node information on the drug to be predicted and the node information on each of the plurality of existing drugs can be derived.

In one embodiment of the present invention, in the step of deriving the first similarity value, expression gene information for one or more genes in which SNPs (Single Nucleotide Polymorphisms) related to each of one or more drug-related diseases are expressed as a third category. In addition, a first similarity value for the third category between the expression gene information for the drug to be predicted and the expression gene information for each of the plurality of existing drugs may be derived.

In one embodiment of the present invention, the step of deriving the first similarity value includes symptom information for one or more symptoms or diseases related to the drug as a fourth category and target gene information for a plurality of target genes targeted by the drug. as a fifth category, and the first similarity value for the fourth category between the symptom information for the drug to be predicted and the symptom information for each of the plurality of existing drugs and the target gene information for the drug to be predicted, and A first similarity value for the fifth category between target gene information for each of the plurality of existing drugs may be derived.

In one embodiment of the present invention, the step of deriving the first similarity value may include, for each category, an element of the intersection between the information on the corresponding category of the drug to be predicted and the information on the corresponding category of each existing drug. A value obtained by dividing the number by the number of elements of the union may be calculated as the first similarity value.

In one embodiment of the present invention, the step of deriving the second similarity value may include, for each of a plurality of anatomically divided layers, the type of layer element in which the predicted side effect is expressed and the layer in which each of the existing side effects is expressed. The second similarity value may be derived based on the type of element.

In one embodiment of the present invention, the step of deriving the second similarity value may include a set of layer elements in which the predicted target side effect is expressed and a layer in which each of the existing side effects is expressed for each of a plurality of anatomically divided layers. Calculating a hierarchical similarity value by dividing the number of elements of the intersection between sets of elements by the number of elements of the union; and calculating, as the second similarity value, a value obtained by dividing the sum of the layer similarity values calculated for each of the plurality of layers by the number of the plurality of layers.

In one embodiment of the present invention, the inference model unit, based on the first similarity representative information and the second similarity representative information to derive information on the first occurrence possibility of the predicted target side effect for the predicted target drug a plurality of first inference models; And a second inference model for deriving information on the second expression possibility of the predicted target side effect for the predicted target drug based on the information on the plurality of first expression possibilities derived by each of the plurality of first inference models. can include

In one embodiment of the present invention, each of the plurality of first inference models includes different inference algorithms, and the inference algorithms include random forest, logistic regression, and naive Bayesian. and XGBoost.

In one embodiment of the present invention, the plurality of first inference models include a plurality of first samples paired with any one first drug and any one known side effect with respect to the first drug, and any one It is learned by the first similarity representative learning information and the second similarity representative learning information for each of a plurality of second samples pairing two drugs and any one side effect unknown to the second drug, and the first similarity Deriving a first similarity value for each of two or more categories between the representative learning information and the second similarity representative learning information, between a drug to be predicted and a plurality of existing drugs known to have side effects to be predicted; deriving first similarity representative information between the predicted target drug and the predicted target side effect by extracting a maximum value from the plurality of first similarity values for each category; deriving each second similarity value between the predicted target side effect and a plurality of existing side effects known for the predicted target drug; Deriving the second similarity representative information between the drug to be predicted and the side effect to be predicted by extracting the maximum value from the plurality of second similarity values;

In order to solve the above problems, the present invention is a system for predicting drug side effects, and derives a first similarity value for each of two or more categories between a drug to be predicted and a plurality of existing drugs known to have side effects to be predicted. a first similarity derivation unit to do; a first similarity representative information derivation unit extracting a maximum value from a plurality of first similarity values for each category and deriving first similarity representative information between the predicted target drug and the predicted target side effect; a second similarity derivation unit for deriving respective second similarity values between the predicted target side effect and a plurality of existing side effects known for the predicted target drug; a second similarity representative information derivation unit extracting a maximum value from a plurality of second similarity values and deriving second similarity representative information between the prediction target drug and the prediction target side effect; and an occurrence possibility prediction unit inputting the first similarity representative information and the second similarity representative information to the learned inference model unit to derive information on the occurrence possibility of the prediction target side effect for the prediction target drug. To provide a system for predicting side effects of drugs.

In order to solve the above problems, the present invention is a computer-readable medium for implementing a method for predicting side effects of a drug performed in a computing device having one or more processors and one or more memories, comprising a drug to be predicted and prediction deriving a first similarity value for each of two or more categories among a plurality of existing drugs known to have target side effects; deriving first similarity representative information between the predicted target drug and the predicted target side effect by extracting a maximum value from the plurality of first similarity values for each category; deriving each second similarity value between the predicted target side effect and a plurality of existing side effects known for the predicted target drug; deriving second similarity representative information between the predicted target drug and the predicted target side effect by extracting a maximum value from the plurality of second similarity values; And inputting the first similarity representative information and the second similarity representative information to the learned inference model unit to derive information on the possibility of occurrence of the predicted target side effect for the predicted target drug; computer comprising; -Provides a readable medium.

According to one embodiment of the present invention, since the first similarity value is derived by categorizing the drug-to-drug relationship for each of the drug to be predicted and a plurality of existing drugs, the predicted side effects of the drug to be predicted more accurately It can exert the effect of predicting the possibility of expression.

According to one embodiment of the present invention, since the first similarity value is derived by categorizing the node information between the drug to be predicted and a plurality of target genes targeted by the drug for each of the plurality of existing drugs, the drug to be predicted more accurately It can exert the effect of predicting the possibility of occurrence of the predicted target side effect for

According to an embodiment of the present invention, since the first similarity value is derived by categorizing expression gene information in which SNPs for each of one or more drug-related diseases are expressed for each of the drug to be predicted and a plurality of existing drugs, the first similarity value is derived more accurately It can exert the effect of predicting the possibility of occurrence of the predicted target side effects for the predicted target drug.

According to one embodiment of the present invention, since a second similarity value is derived between a predicted side effect and a plurality of existing side effects known for the predicted drug, the possibility of occurrence of the predicted side effect for the predicted drug can be more accurately predicted. effect can be exerted.

According to an embodiment of the present invention, the learned inference model unit for deriving information on the possibility of occurrence includes a second inference model implemented by ensemble learning that takes as input result values derived from a plurality of first inference models. , can exert the effect of improving the performance of the inference model unit.

1 schematically illustrates a process of predicting the possibility of occurrence of a predicted side effect for a predicted drug based on a plurality of databases according to an embodiment of the present invention.
2 schematically illustrates a computing device implementing a method for predicting side effects of drugs according to an embodiment of the present invention.
Figure 3 schematically shows detailed steps for a method for predicting side effects of drugs according to an embodiment of the present invention.
Figure 4 schematically shows a drug-to-drug relationship according to an embodiment of the present invention.
5 schematically shows an interaction pathway between a plurality of target genes targeted by a drug according to an embodiment of the present invention.
6 schematically shows expression gene information for a drug according to an embodiment of the present invention.
7 schematically illustrates symptom information related to drugs according to an embodiment of the present invention.
8 schematically illustrates hierarchical elements in which side effects are expressed according to a plurality of anatomically divided hierarchies according to an embodiment of the present invention.
9 schematically illustrates a process of deriving first similarity representative information and second similarity representative information from a plurality of first similarity values and a plurality of second similarity values for each category according to an embodiment of the present invention.
10 schematically illustrates the internal configuration of an inference model unit according to an embodiment of the present invention.
11 schematically illustrates a process in which a first inference model learns by learning data according to an embodiment of the present invention.
12 schematically illustrates a process of deriving similarity representative learning information according to an embodiment of the present invention.
13 schematically illustrates Area Under the ROC Curve (AUC) values for various configurations of inference model units according to an embodiment of the present invention.
14 schematically illustrates the internal configuration of a computing device according to an embodiment of the present invention.

In the following, various embodiments and/or aspects are disclosed with reference now to the drawings. In the following description, for purposes of explanation, numerous specific details are set forth in order to facilitate a general understanding of one or more aspects. However, it will also be appreciated by those skilled in the art that such aspect(s) may be practiced without these specific details. The following description and accompanying drawings describe in detail certain illustrative aspects of one or more aspects. However, these aspects are exemplary and some of the various methods in principle of the various aspects may be used, and the described descriptions are intended to include all such aspects and their equivalents.

Moreover, various aspects and features will be presented by a system that may include a number of devices, components and/or modules, and the like. It should also be noted that various systems may include additional devices, components and/or modules, and/or may not include all of the devices, components, modules, etc. discussed in connection with the figures. It must be understood and recognized.

"Example", "example", "aspect", "exemplary", etc., used herein should not be construed as preferring or advantageous to any aspect or design being described over other aspects or designs. . The terms '~unit', 'component', 'module', 'system', 'interface', etc. used below generally mean a computer-related entity, and for example, hardware, hardware It may mean a combination of and software, software.

Also, the terms "comprises" and/or "comprising" mean that the feature and/or element is present, but excludes the presence or addition of one or more other features, elements and/or groups thereof. It should be understood that it does not.

In addition, terms including ordinal numbers, such as first and second, may be used to describe various components, but the components are not limited by the terms. These terms are only used for the purpose of distinguishing one component from another. For example, a first element may be termed a second element, and similarly, a second element may be termed a first element, without departing from the scope of the present invention. The terms and/or include any combination of a plurality of related recited items or any of a plurality of related recited items.

In addition, in the embodiments of the present invention, unless otherwise defined, all terms used herein, including technical or scientific terms, are generally understood by those of ordinary skill in the art to which the present invention belongs. has the same meaning as Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and unless explicitly defined in the embodiments of the present invention, an ideal or excessively formal meaning not be interpreted as

1 schematically illustrates a process of predicting the possibility of occurrence of a predicted side effect for a predicted drug based on a plurality of databases according to an embodiment of the present invention.

As shown in Figure 1, a plurality of databases for drugs or side effects are used to predict the possibility of occurrence of predicted side effects for a predicted drug, and the plurality of databases are considered as a conventional method for predicting side effects. In addition to the database for information on the structure of drugs and target genes of drugs, a database on drug-drug interactions and a protein-protein interaction (PPI) network between multiple target genes of drugs By additionally using databases, databases on genes expressing Single Nucleotide Polymorphism (SNP) related to diseases, databases on drug indications, and databases on anatomical characteristics of side effects, predictive target side effects for predicted drugs possibility of occurrence can be predicted.

The databases shown in FIG. 1 may correspond to separate external databases including the above information, such as DrugBank, Search Tool for the Retrieval of Interacting Genes/Proteins (STRING), PubChem, TTD, CTD, and repoDB. The computing device 1000 that implements the method for predicting side effects of drugs may directly receive information from the external databases or may receive information about the external database from a user or manager of the computing device 1000.

Meanwhile, the computing device 1000 derives similarity representative information between a drug to be predicted and a side effect to be predicted by using the information of the plurality of databases described above. Specifically, the similarity representative information includes first similarity representative information and second similarity representative information, and the first similarity representative information corresponds to the degree of similarity between the predicted target drug and a plurality of existing drugs known to have the predicted target side effect. And, the second similarity representative information corresponds to the degree of similarity between the predicted target side effect and a plurality of existing side effects known for the predicted target drug.

As such, the present invention does not simply consider the similarity between drugs, but also considers the similarity between the side effect to be predicted and the known side effect of the drug to be predicted, so the expression of the side effect to be predicted for the drug to be predicted It can exert the effect of predicting the possibility more accurately.

The drug to be predicted in the present invention means a drug that is a target for predicting what kind of side effects may occur, such as a new drug candidate, and the side effect to be predicted means a side effect to be predicted for the drug to be predicted. In addition, the existing drug refers to a drug that is already known to cause the predicted target side effect, and the existing side effect refers to a side effect that is already known to be caused by the predicted target drug.

The computing device 1000 derives information on the possibility of occurrence of the predicted side effect for the predicted drug based on the similarity representative information between the predicted target drug and the predicted side effect. Specifically, the computing device 1000 inputs the similarity representative information into the machine-learned reasoning model unit 1700 to derive information about the possibility of occurrence, and the information about the possibility of occurrence determines whether the predicted side effect occurs. , or quantified information about the likelihood of occurrence.

Hereinafter, a method of predicting drug side effects implemented in the computing device 1000 will be described in detail.

2 schematically illustrates a computing device 1000 implementing a method for predicting side effects of drugs according to an embodiment of the present invention.

As shown in FIG. 2 , the computing device 1000 implementing the method for predicting side effects of other drugs according to an embodiment of the present invention may communicate with an external reference database 2000 .

The reference database 2000 may include a plurality of databases, and the plurality of databases may include a first reference database 2100, a second reference database 2200, and an Nth reference database 2300, As described above, each of the reference databases 2100 to 2300 includes information related to drugs or side effects.

In this way, the reference database 2000 can communicate with the computing device 1000 implementing a method for predicting side effects of drugs, and the computing device 1000 can communicate with the reference database 2000. have access to information.

Meanwhile, the computing device 1000 includes a database processing unit 1100, a first similarity derivation unit 1200, a first similarity representative information derivation unit 1300, a second similarity derivation unit 1400, and a second similarity representative It may include an information derivation unit 1500, an occurrence possibility prediction unit 1600, an inference model unit 1700, and a DB 1800.

Specifically, the database processing unit 1100 accesses the reference database 2000 and receives the stored information or, in the computing device 1000, the first similarity value between the predicted target drug and the existing drug and the predicted target side effect and the existing drug Received information may be processed to easily derive a second similarity value between side effects. Meanwhile, the database processing unit 1100 may directly receive information about the reference database 2000 from a user and process the corresponding information, or may directly receive processed information from a user.

The first similarity deriving unit 1200 derives a first similarity value corresponding to the degree of similarity between the prediction target drug and a plurality of existing drugs known to have the prediction target side effects. Specifically, the first similarity deriving unit 1200 derives a first similarity value for each of the prediction target drug and the plurality of existing drugs according to a plurality of preset categories, and derives a first similarity value according to the category A detailed method will be described later.

The first similarity representative information derivation unit 1300 obtains first similarity representative information based on a plurality of first similarity values between the drug to be predicted and the plurality of existing drugs derived by the first similarity derivation unit 1200. derive Specifically, the first similarity representative information derivation unit 1300 derives the first similarity representative information by extracting the maximum value among a plurality of first similarity values derived for each category of the prediction target drug and the plurality of existing drugs. Therefore, the first similarity representative information may include first similarity values corresponding to the maximum value for each category.

The second similarity deriving unit 1400 derives a second similarity value corresponding to the degree of similarity between the predicted target side effect and a plurality of existing side effects known for the predicted target drug. Specifically, the second similarity deriving unit 1400 derives a second similarity value for each of the predicted side effect and the plurality of existing side effects according to a preset category, and a detailed method of deriving the second similarity value will be described later. let it do

The second similarity representative information derivation unit 1500 obtains second similarity representative information based on a plurality of second similarity values between the prediction target side effect derived by the second similarity derivation unit 1400 and the plurality of existing side effects. derive Specifically, the second similarity representative information derivation unit 1500 derives the second similarity representative information by extracting the maximum value among a plurality of second similarity values derived for each of the predicted target side effect and the plurality of existing side effects. Therefore, the second similarity representative information may include a second similarity value corresponding to the maximum value of the preset category.

Meanwhile, the above-described first similarity derivation unit 1200, the first similarity representative information derivation unit 1300, the second similarity derivation unit 1400, and the second similarity representative information derivation unit 1500, In addition to deriving the information input to the inference model unit 1700 to derive the possibility of occurrence of the predicted target side effect for the target drug, the first similarity representative corresponding to the learning data for the inference model unit 1700 to learn Learning information and second similarity representative learning information may be derived.

The occurrence possibility prediction unit 1600 is the expression of the predicted target side effect with respect to the predicted target drug based on the first similarity representative information and the second similarity representative information derived for the predicted target drug and the predicted target side effect. Derive information about possibilities. Specifically, the occurrence possibility prediction unit 1600 may derive information on the occurrence possibility by inputting the first similarity representative information and the second similarity representative information to the inference model unit 1700 to be described later.

The reasoning model unit 1700 receives the first similarity representative information and the second similarity representative information about the predicted target drug and the predicted target side effect from the occurrence possibility prediction unit 1600, and derives information on the occurrence possibility. do. Specifically, the reasoning model unit 1700 may derive information on the possibility of occurrence based on a preset rule, but preferably, the reasoning model unit 1700 is a machine learning algorithm that performs learning in a predetermined method. Based on , information on the expression possibility is derived.

Meanwhile, in FIG. 2 , the inference model unit 1700 is illustrated as being included in the computing device 1000, but is not limited thereto, and in another embodiment of the present invention, the inference model unit 1700 is included in the computing device It may be included in another server or other computing device separate from (1000), and the occurrence possibility prediction unit 1600 calls the inference model unit 1700 included in the other server or other computing device to call the occurrence possibility prediction unit 1700. Information about the possibility can be derived.

The DB 1800 may store information necessary to predict the occurrence possibility of a predicted target drug and a predicted target side effect, and specifically, a plurality of first similarity values and a plurality of second predicted drug and predicted target side effects. Similarity representative information including a similarity value, first similarity representative information, and second similarity representative information may be stored, and learning data for the reasoning model unit 1700 to learn, specifically, the first similarity representative learning information and Learning data including the second similarity representative learning information may be stored. Additionally, the DB 1800 may store prediction results derived from the expression possibility prediction unit 1600, that is, information about expression possibility.

Figure 3 schematically shows detailed steps for a method for predicting side effects of drugs according to an embodiment of the present invention.

As shown in FIG. 3, a method for predicting drug side effects implemented by a computing device 1000 is provided for each of two or more categories between a drug to be predicted and a plurality of existing drugs known to have side effects to be predicted. 1 Deriving a similarity value (S100); Deriving first similarity representative information between the predicted drug and the predicted side effect by extracting a maximum value from the plurality of first similarity values for each category (S110); deriving each second similarity value between the predicted target side effect and a plurality of existing side effects known for the predicted target drug (S120); deriving second similarity representative information between the predicted target drug and the predicted target side effect by extracting a maximum value from the plurality of second similarity values (S130); And inputting the first similarity representative information and the second similarity representative information to the learned inference model unit 1700 to derive information on the possibility of occurrence of the predicted target side effect for the predicted target drug (S140) ; can be included.

Specifically, the computing device 1000 implementing a method for predicting side effects of a drug is a specific drug (target drug to be predicted) to predict side effects from a user and a specific side effect to be predicted for the drug to be predicted (target side effects to be predicted). information can be entered.

Thereafter, the first similarity deriving unit 1200 of the computing device 1000 derives a first similarity value for each of the predicted target drug and a plurality of existing drugs known to have the predicted target side effect ( S100) is performed.

Specifically, in the step of deriving the first similarity value (S100), a first similarity value for each category is derived according to preset two or more categories for each of the drug to be predicted and the plurality of existing drugs. Therefore, the type of first similarity value for the drug to be predicted and a specific existing drug may be derived for each of two or more predetermined categories, and a plurality of first similarity values corresponding to the number of a plurality of existing drugs are derived for each category. It can be.

That is, in the step of deriving the first similarity value (S100), the similarity between the predicted target drug and other drugs may be measured in order to derive information on the possibility of occurrence of the predicted target drug for the predicted target drug.

On the other hand, information on existing drugs known to have the predicted side effects may be input in advance to the computing device 1000 by the user or accessed or received from an external database such as the reference database 2000. .

The first similarity representative information derivation unit 1300 performs a step (S110) of deriving first similarity representative information based on a plurality of first similarity values derived by the first similarity derivation unit 1200.

Specifically, in the step of deriving the first similarity representative information (S110), a first similarity value having a maximum value among a plurality of first similarity values derived for each category is derived as the first similarity representative information. Accordingly, the first similarity representative information may include first similarity values having a maximum value for each category.

In this way, the first similarity representative information derived through the first similarity representative information derivation unit 1300 may be included in input information to be input to the inference model unit 1700, and specifically, prediction among the input information. It may correspond to an input element for the degree of similarity between the target drug and a plurality of existing drugs known to have predicted target side effects.

Meanwhile, the second similarity deriving unit 1400 performs a step (S120) of deriving a second similarity value for each of a plurality of existing side effects known to be caused by the predicted target side effect and the predicted target drug. do.

Specifically, in the step of deriving the second similarity value (S120), a second similarity value for a specific category is derived for each of the predicted target side effect and the plurality of existing side effects. Accordingly, a plurality of second similarity values corresponding to the number of a plurality of existing side effects may be derived for the specific category.

That is, in the step of deriving the second similarity value (S120), the degree of similarity between the predicted side effect and other side effects may be measured in order to derive information on the possibility of occurrence of the predicted side effect for the predicted drug.

On the other hand, information on a plurality of known side effects of the drug to be predicted may be entered into the computing device 1000 in advance by the user or accessed or received from an external database such as the reference database 2000. can

The second similarity representative information derivation unit 1500 performs a step of deriving second similarity representative information based on a plurality of second similarity values derived by the second similarity derivation unit 1400 (S130).

Specifically, in the step of deriving the second similarity representative information (S130), a second similarity value having the maximum value among a plurality of second similarity values is derived as the second similarity representative information, and thus the second similarity representative information. The second similarity representative information derived through the derivation unit 1500 may be included in input information to be input to the inference model unit 1700, and specifically, it is expressed by predicted target side effects and predicted target drugs among the input information. It may correspond to an input element for similarity with a plurality of existing side effects known to be possible.

In this way, by measuring the similarity between the predicted target drug and the existing drugs and the similarity between the predicted target side effect and the existing side effects through the steps S100 to S130, the first similarity representative information and the second similarity representative information And based on this, information on the possibility of occurrence of the side effect to be predicted for the drug to be predicted is derived from the occurrence possibility prediction unit. It is possible to derive information on the possibility of occurrence.

Meanwhile, through the steps S100 to S130, the computing device 1000 does not derive only the first similarity representative information and the second similarity representative information, but the learning data and the learning data for the reasoning model unit 1700 to learn. Verification data for verifying the reasoning model unit 1700 learned by the may be derived.

Specifically, the learning data may include first similarity representative learning information and second similarity representative learning information for a plurality of samples pairing a specific drug with a specific side effect, and in this case, the above-described steps S100 to S100. In S110, first similarity representative information may be derived by using a specific drug of the sample as the predicted target drug and a specific side effect of the sample as the predicted target side effect, and the derived first similarity representative information is the first similarity representative information. It may correspond to similarity representative learning information. Similarly, in the above steps S120 to S130, the second similarity representative learning information for the specific drug and the specific side effect may be derived, and the derived second similarity representative learning information may correspond to the second similarity representative learning information. can

In addition, as described above, the verification data may also include first similarity representative verification information and second similarity representative verification information for a plurality of samples in which a specific drug and a specific side effect are paired. It can be derived through step S130.

Finally, the occurrence possibility prediction unit 1600 inputs the derived first similarity representative information and the second similarity representative information to the learned reasoning model unit 1700, and the occurrence possibility of the predicted target side effect for the predicted target drug A step (S140) of deriving information about is performed.

In the step of deriving information on the possibility of occurrence (S140), the result value derived through the learned reasoning model unit 1700 is used as information on the possibility of occurrence, or a separate possibility of occurrence is determined based on the result value. information can also be derived.

Figure 4 schematically shows a drug-to-drug relationship according to an embodiment of the present invention.

As shown in FIG. 4, the step of deriving the first similarity value (S100) includes, as a first category, a correlation between one or more other drugs that the drug affects or is affected by, and the prediction target The first for the first category between the mutual relationship between one or more other drugs that the drug affects or is affected and the mutual relationship between one or more other drugs that each of the plurality of existing drugs affects or is affected by. A similarity value can be derived.

Specifically, in the step of deriving the first similarity value (S100), a drug affects one or more other drugs, or a correlation influenced by one or more other drugs is set as a first category, and the first category A plurality of first similarity values for each of the prediction target drug and the plurality of existing drugs for the category are derived.

The first category for the interaction between drugs and other drugs (DDIs-D) corresponds to a category considering the tendency that each similar drug shows a similar reaction to other drugs. As shown in FIG. 4, the correlation between each drug can be expressed in the form of (x, y, z), where x means another drug having a correlation, y means the type of correlation, and z indicates whether it affects other drugs or is affected by other drugs.

For example, among the correlations for drug A, (B,1,←) means that drug A is affected in type 1 by drug B. As such, the computing device 1000 stores information on the mutual relationship of each drug based on the information of the external database, and the step of deriving the first similarity value (S100) includes the relationship between the drug to be predicted and It is possible to derive a plurality of first similarity values for the first category corresponding to the degree of similarity of the mutual relationship for each of a plurality of existing drugs known to cause predicted target side effects.

5 schematically shows an interaction pathway between a plurality of target genes targeted by a drug according to an embodiment of the present invention.

As shown in FIG. 5, in the step of deriving the first similarity value (S100), node information on genes and proteins included in the interaction pathway between a plurality of target genes targeted by the drug is classified as a second category. , and a first similarity value for the second category between node information on the drug to be predicted and node information on each of the plurality of existing drugs can be derived.

Specifically, in the step of deriving the first similarity value (S100), node information on proteins, genes, or metabolites included in the interaction path between a plurality of target genes targeted in the human body after the drug is administered. as a second category, and a plurality of first similarity values for each of the drug to be predicted and a plurality of existing drugs for the second category are derived.

The second category for the node information (DDIs-N) of the interaction pathway between drug target genes corresponds to a category considering the functional similarity between two drugs in the human body. As shown in FIG. 5, each drug has one or more target genes (TG _A and T _B ) that are initially affected, and from the first target genes (TG _A and T B) to the finally affected target genes (TG A and T _B ). TG _C and TG _D ) can propagate the effect of the drug along a predetermined network pathway.

Meanwhile, proteins, genes, or metabolites (P1 to P7) corresponding to nodes exist in the network path between target genes, and the first similarity value for the second category is the first target gene of each predicted drug and existing drug ( It may correspond to similarity of information about nodes (P1 to P7) included in an interaction path between T _A and T _B ) and the final target genes (TG _C and TG _D ).

For example, drug A shown in FIG. 5 has _TGA and _TGA as initial target genes, and the initial target gene, _TGA , can interact with the final target gene, TG _C , via nodes P1, P4, and P6, , the first target gene T _B can interact with the final target gene TG _D via nodes P3, P4, P5, and P7. Therefore, the node information of drug A includes information on proteins, genes, or metabolites constituting the pathway between the initial target gene and the final target gene as described above.

Therefore, node information for each drug is stored in the computing device 1000 based on information of an external database, and in the step of deriving the first similarity value (S100), the node information of the predicted drug and the predicted side effect are expressed. A plurality of first similarity values for the second category corresponding to the similarity of node information for each of a plurality of existing drugs known to be available may be derived.

In another embodiment of the present invention, the node information further includes information on proteins, genes, or metabolites corresponding to the pathway between the first target gene and the final target gene, and information on the first target gene or the final target gene. can do.

In another embodiment of the present invention, a plurality of paths may exist from the first target gene to the final target gene, and in this case, among the paths from the first target gene to the final target gene, the shortest path may be used as node information.

The target gene in the present invention may refer to a specific gene in the human body, but is not limited thereto, and may include a specific protein or a specific metabolite.

On the other hand, as shown in FIG. 5, the step of deriving the first similarity value (S100) includes target gene information for a plurality of target genes targeted by the drug as a fifth category, and A first similarity value for the fifth category between the target gene information for the drug and the target gene information for each of the plurality of existing drugs may be derived.

Specifically, in the step of deriving the first similarity value (S100), the target gene information (Target) for a plurality of target genes targeted by the drug is set as the fifth category, and the predicted target drug for the fifth category and a plurality of first similarity values for each of the plurality of existing drugs are derived.

The fifth category for a plurality of target genes of a drug corresponds to a category that considers the biological characteristics of the drug, that is, when two drugs are related to the same target gene, the two drugs undergo similar biological processes and thus have the same side effects. .

Therefore, information on a plurality of target genes for each drug is stored in the computing device 1000 based on information of an external database, and the step of deriving the first similarity value (S100) is the target gene information of the drug to be predicted. It is possible to derive a plurality of first similarity values for the fifth category corresponding to the similarity of target gene information for each of a plurality of existing drugs known to have predicted target side effects.

On the other hand, as described above, the target gene information may include information on the first target genes (TG _A and T _B ) and the final target genes (TG _C and TG _D ), but preferably the target gene information It may also include only information on the first target genes (TG _A and T _B ).

6 schematically shows expression gene information for a drug according to an embodiment of the present invention.

As shown in FIG. 6, the step of deriving the first similarity value (S100) provides expression gene information for one or more genes in which a single nucleotide polymorphism (SNP) associated with each of one or more drug-related diseases is expressed. 3 categories, and a first similarity value for the third category between the expression gene information for the predicted drug and the expression gene information for each of the plurality of existing drugs can be derived.

Specifically, in the step of deriving the first similarity value (S100), expression gene information for one or more genes in which SNPs related to each of one or more diseases related to drugs is expressed as a third category, and A plurality of first similarity values for each of a prediction target drug and a plurality of existing drugs are derived.

The third category for drug expression gene information (SNP) corresponds to a category considering that similar side effects may occur when the genetic characteristics of diseases related to two drugs are similar. SNP, that is, single nucleotide polymorphism (SNP) corresponds to a factor that determines the individual diversity of humans, and SNPs appear in about one out of every 1000 bases in the human genome, whereby individual genetic diversity is expressed. Specifically, there may be people who have C (cytosine) and people who have A (adenine) at specific positions in the DNA chain, and individual diversity can be expressed by these differences. The genetic diversity expressed by SNPs can also be utilized in terms of susceptibility to disease. The susceptibility to a specific disease may appear differently depending on the genetic diversity of each person.

Accordingly, as shown in FIG. 6, for each drug (Drug A), one or more diseases (Disease 1 to Disease N) associated with the corresponding drug (Drug A) exist, and each disease (Disease 1 to Disease N) SNPs related to sensitivity may exist, and genes (Gene1 to Gene2) in which each SNP can be expressed, that is, expression genes for each drug may exist.

Therefore, information on a plurality of expression genes for each drug is stored in the computing device 1000 based on the information of the external database, and the step of deriving the first similarity value (S100) is the expression gene of the predicted drug. It is possible to derive a plurality of first similarity values for the third category corresponding to the similarity of expression gene information for each of a plurality of existing drugs known to have the information and predicted target side effects.

7 schematically illustrates symptom information related to drugs according to an embodiment of the present invention.

As shown in FIG. 7, the step of deriving the first similarity value (S100) includes symptom information on one or more symptoms or diseases related to a drug as a fourth category, and symptom information on the predicted drug. And a first similarity value for the fourth category between symptom information for each of the plurality of existing drugs may be derived.

Specifically, in the step of deriving the first similarity value (S100), symptom information on an indication corresponding to one or more symptoms or diseases related to a drug is set as a fourth category, and prediction for the fourth category is made. A plurality of first similarity values for each of the target drug and a plurality of existing drugs are derived.

The fourth category for symptom information of drugs corresponds to a category considering the similarity of phenotypic characteristics between drugs, and indication refers to a disease or symptom for which a therapeutic effect is expected by a drug.

Accordingly, as shown in FIG. 7 , symptom information on one or more symptoms or diseases related to each drug may exist for each drug. For example, the symptom information for Drug A shown in FIG. 7 may include disease A, disease B, symptom A, and symptom D. As described above, symptom information for each drug is stored in the computing device 1000 based on information of an external database, and the step of deriving the first similarity value (S100) includes symptom information about the predicted drug and the predicted target side effect. It is possible to derive a plurality of first similarity values for the fourth category corresponding to the similarity of symptom information for each of a plurality of existing drugs known to be expressed.

As in the above-described FIGS. 4 to 7, the step of deriving the first similarity value (S100) includes the above-described first to fifth categories in the two or more categories, and the first to fifth categories A plurality of first similarity values may be derived for each. In this way, the first similarity representative information can be derived based on the plurality of first similarity values derived for each category, and the step (S110) of deriving the first similarity representative information will be described in detail with reference to FIG. 9 .

Meanwhile, in another embodiment of the present invention, the step of deriving the first similarity value (S100) includes information on the chemical structure of the drug as a sixth category, information on the chemical structure of the predicted drug, and A first similarity value for the sixth category between information on the chemical structure of each of the plurality of existing drugs may be additionally derived.

Therefore, in one embodiment of the present invention, since the degree of similarity between drugs is measured for each of a plurality of categories, side effects of drugs can be more accurately predicted.

8 schematically illustrates hierarchical elements in which side effects are expressed according to a plurality of anatomically divided hierarchies according to an embodiment of the present invention.

As shown in FIG. 8, in the step of deriving the second similarity value (S120), for each of a plurality of anatomically divided layers, the type of layer element in which the predicted target side effect is expressed and each of the existing side effects The second similarity value may be derived based on the type of hierarchical elements that are expressed.

Specifically, as shown in (A) of FIG. 8 , each side effect may be classified according to a predetermined hierarchy classified anatomically. In (A) of FIG. 8, it is divided into three classes, Organs, Subsystems, and Systems, and in (B) of FIG. 8, side effects are expressed for each anatomically classified class. The types of hierarchical elements that are used are shown.

A plurality of hierarchical elements may exist according to a predetermined anatomically classified hierarchy. For example, hierarchical elements such as the heart, liver, and lungs may exist in the Organs hierarchy, and hierarchical elements such as the respiratory system, muscular system, and digestive system may exist in the Systems hierarchy. Meanwhile, each side effect may affect a specific hierarchical element for each anatomically classified hierarchy. For example, as shown in (B) of FIG. 8, side effect 1 may affect A1, A3, and A4, which are hierarchical elements of the corresponding hierarchy, in the Organs hierarchy, and may affect the Subsystems hierarchy. can affect the hierarchical elements of the corresponding layer, B2 and B3, and can affect the hierarchical elements of the corresponding layer, C1, C4, and C5, in the Systems layer. In this way, the classification of side effects into hierarchical elements for each level that is anatomically classified takes into consideration that similar side effects are associated with similar anatomical levels.

On the other hand, for example, if a specific side effect affects the lungs, which is a hierarchical element of the Organs hierarchy, and also affects the respiratory system to which the lungs belong, it is not classified as affecting only the respiratory system, but each is classified independently. Therefore, specific side effects can be classified as those affecting the lungs in the Organs hierarchy and the respiratory system in the Systems hierarchy, respectively.

Therefore, the computing device 1000 stores the types of hierarchical elements in which the corresponding side effect can be expressed for each of a plurality of anatomically divided hierarchies for each side effect based on the information of the external database, and the second similarity value. The step of deriving (S120) is a plurality of second similarity values corresponding to the similarity between the types of hierarchical elements of the predicted side effects and the types of hierarchical elements for each of a plurality of existing side effects known to be caused by the predicted drug. can be derived.

Meanwhile, although three anatomically separated layers are shown in FIG. 8, the present invention is not limited thereto, and two or more anatomically divided layers may be used in another embodiment of the present invention.

9 schematically illustrates a process of deriving first similarity representative information and second similarity representative information from a plurality of first similarity values and a plurality of second similarity values for each category according to an embodiment of the present invention.

9(A) corresponds to a diagram showing a first similarity value for each of two or more categories with respect to a drug to be predicted and each of the plurality of existing drugs. Specifically, as shown in the upper left of FIG. 9 (A), in order to predict whether the predicted target side effect (green diamond) can be expressed by the predicted target drug (red drug), as shown at the bottom, A first similarity value for each of two or more categories may be derived between the four existing drugs known to have the expected side effects (green diamonds) and the predicted target drug (red-type drug).

To this end, in the step of deriving the first similarity value (S100), for each category, the number of elements of the intersection between the information on the corresponding category of the drug to be predicted and the information on the corresponding category of each existing drug is determined. A value divided by the number of elements in the union may be calculated as the first similarity value.

Specifically, in the step of deriving the first similarity value (S100), the first similarity value for each category between the drug to be predicted and the existing drug may be calculated using Equation (1) below.

- 식 (1)

- formula (1)

Equation (1) above is for calculating the first similarity value for drug D _A and drug D _B , and the set of drug D _A information set S _A and drug DB information set _{S B for} a specific category The value obtained by dividing the number of elements in the intersection of sets S _A and S _B by the number of elements in the union of sets S _A and S _B corresponds to the first similarity value between drugs D _A and drugs D _B for that particular category. The range of the first similarity value has a value of 0 when the two drugs are not completely similar, and has a value of 1 when the two drugs are completely similar.

Preferably, the above formula (1) can be used to calculate the first similarity value for the above-mentioned first to fifth categories, and the Tanimoto coefficient for the above-mentioned sixth category for the chemical structure of the drug. ) can also be calculated through a separate formula for calculating .

Therefore, as shown in (A) of FIG. 9, the first similarity values of the drug to be predicted and existing drugs for the first category (DDIs-D) are 0.23, 0.87, 0.79, and 0.44, in order from the top. and the first similarity values of the predicted drug and existing drugs for the fifth category (Target) were calculated as 0.45, 0.71, 0.24, and 0.95 in order from the top.

Meanwhile, in the step of deriving the first similarity representative information (S110), the first similarity representative information is derived by extracting the maximum value among the plurality of first similarity values derived for each category as described above. In the first similarity representative information shown in (A) of FIG. 9, the maximum value of 0.87 among the plurality of first similarity values for the first category (DDIs-D) and the plurality of first similarity values for the fifth category (Target) The maximum value of 0.95 among the similarity values and the maximum value of 0.86 among the plurality of first similarity values for the sixth category may be included.

9(B) corresponds to a diagram showing a second similarity value for each of a predicted target side effect and a plurality of existing side effects. Specifically, as shown on the left side of FIG. 9 (B), in order to predict whether the predicted target side effect (green diamond) can be expressed by the predicted target drug (red series drug), the predicted target side effect (green diamond) ) and a second similarity value between a plurality of existing side effects (blue, purple, yellow, light yellow diamonds) known to be expressed by the predicted target drug (red drug) can be derived.

To this end, in the step of deriving the second similarity value (S120), for each of a plurality of anatomically divided layers, a set of layer elements in which the predicted target side effect is expressed and each of the layer elements in which the existing side effect is expressed Calculating a hierarchical similarity value by dividing the number of elements of the intersection between sets by the number of elements of the union; and calculating, as the second similarity value, a value obtained by dividing the sum of the layer similarity values calculated for each of the plurality of layers by the number of the plurality of layers.

Specifically, in the step of deriving the second similarity value (S120), a second similarity value between the predicted side effect and the existing side effect may be calculated using Equation (2) below.

- 식 (2)

- Eq. (2)

S _l (SE _a , SE _b ) in Equation (2) can be calculated by Equation (1), and l in Equation (2) corresponds to a specific layer among a plurality of anatomically separated layers. And, n corresponds to the number of anatomically divided plural layers. In the case of calculating the second similarity value for FIG. 8 described above, n may correspond to 3. Similarly, the range of the second similarity value calculated by Equation (2) may have a value of 0 or more and 1 or less.

Therefore, as shown in (B) of FIG. 9, the second similarity values of the predicted side effects and existing side effects were calculated as 0.33, 0.44, 0.5, and 0.11 in order from the top.

Meanwhile, in the step of deriving the second similarity representative information (S130), the second similarity representative information is derived by extracting the maximum value among the plurality of second similarity values derived as described above. The second similarity representative information shown in (B) of FIG. 9 may include a maximum value of 0.5 among a plurality of second similarity values.

Accordingly, as described in FIG. 9 (A) and FIG. 9 (B), the first similarity representative information and the second similarity representative between the predicted target drug (red series drug) and the predicted target side effect (green diamond) Similarity representative information including information can be derived as shown in FIG. 9(C). The similarity representative information may be input to the learned inference model unit 1700, and the inference model unit 1700 determines the possibility of occurrence of the predicted side effect with respect to the predicted drug based on the input similarity representative information. information can be derived.

The 'Class' item described in (C) of FIG. 9 corresponds to a label assigned when the similarity representative information is used as learning or verification data for learning or verification of the reasoning model unit 1700, and 'Class' The 'False' value of the ' item may correspond to a value labeled when the relationship between the predicted drug and the predicted side effect is not known, and the 'True' value of the 'Class' item It may correspond to the value to be labeled when the relationship between the expected side effects is known.

In this way, the similarity representative information derived through steps S100 to S130 may be used as input information for predicting the possibility of occurrence of the predicted side effect for the predicted drug, and derived through steps S100 to S130. The similarity representative information to be used may correspond to similarity representative learning information or similarity representative verification information for learning or verification of the reasoning model unit 1700, and the similarity representative learning information and the similarity representative verification information include the corresponding drug and the corresponding side effect. The relationship between them can be labeled.

10 schematically illustrates the internal configuration of an inference model unit 1700 according to an embodiment of the present invention.

As shown in FIG. 10, the inference model unit 1700 includes a plurality of first inference models, each of the plurality of first inference models includes a different inference algorithm, and the inference algorithm is random. It may include Random forest, Logistic regression, Naive Bayesian and XGBoost.

Specifically, FIG. 10(A) shows result values derived by inputting similarity representative information for a predicted drug and a predicted side effect to a plurality of first inference models. The plurality of first inference models are different inference algorithms, specifically, each of the plurality of first inference models is an algorithm for classifying input information, such as random forest, logistic regression, and Naive Bayesian ) and XGBoost's inference algorithm.

In (A) of FIG. 10, four first inference models are shown, and the algorithms of each first inference model are described as random forest, logistic regression, naive Bayesian, and XGBoost However, in another embodiment of the present invention, it is not limited thereto, and the inference model unit 1700 may include two or more first inference models, and the inference algorithm of the first inference model is a random forest , logistic regression, Naive Bayesian, and XGBoost, as well as various inference algorithms for classifying input information.

Meanwhile, as shown in (B) of FIG. 10 , the inference model unit 1700 determines the side effects to be predicted for the drug to be predicted based on the first similarity representative information and the second similarity representative information. a plurality of first inference models for deriving information on a first expression possibility; And a second inference model for deriving information on the second expression possibility of the predicted target side effect for the predicted target drug based on the information on the plurality of first expression possibilities derived by each of the plurality of first inference models. can include

Specifically, the inference model unit 1700 may be implemented by an ensemble learning method, and thus each of the plurality of first inference models may correspond to a classifier, and the second inference model may correspond to a meta classifier. That is, it includes a plurality of first inference models and a plurality of second inference models, and each of the plurality of first inference models receives similarity representative information about the drug to be predicted and the side effect to be predicted, and information on the first expression possibility, respectively. derived, and the second inference model may receive information on a plurality of first expression possibilities derived by each of the plurality of first inference models and derive information on a second expression possibility.

The information on the second expression possibility may correspond to the information on the expression possibility derived from the expression possibility prediction unit 1600 described above, or the information on the second expression possibility in the expression possibility prediction unit 1600. Based on, it is also possible to derive separate information on the possibility of occurrence.

Meanwhile, the inference model unit 1700 of the present invention may apply various conventional methods such as majority voting and majority voting to perform ensemble learning.

11 schematically illustrates a process in which a first inference model learns by learning data according to an embodiment of the present invention.

As shown in FIG. 11, the plurality of first inference models include a plurality of first samples paired with any one first drug and any one known side effect with respect to the first drug, and any one first inference model. It can be learned by the first similarity representative learning information and the second similarity representative learning information for each of a plurality of second samples pairing two drugs and any one side effect unknown to the second drug.

Specifically, the first inference model includes similarity representative learning information including first similarity representative learning information and second similarity representative learning information for a plurality of first samples, first similarity representative learning information for a plurality of second samples, and Learning may be performed using similarity representative learning information including the second similarity representative learning information as learning data.

The first sample corresponds to a sample paired with a first drug, which is a specific drug, and a specific side effect known to the first drug, and the similarity representative learning information for the first sample is the step described above. It can be derived in the same way as the method of deriving the similarity representative information through steps S100 to S130. Meanwhile, the similarity representative learning information for the first sample may further include a labeling value such as 'Ture' when the first inference model learns in a supervised learning method.

On the other hand, the second sample corresponds to a sample in which a second drug, which is a specific drug, and a specific side effect unknown to the second drug are paired, and the similarity representative learning information for the second sample is as described above. It can be derived in the same way as the method of deriving similarity representative information through one step S100 to step S130. Similarly, the similarity representative learning information for the second sample may further include a labeling value such as 'False' when the first inference model learns in a supervised learning method.

In another embodiment of the present invention, the verification data for verifying the first inference model may include similarity representative verification information for the plurality of first samples and similarity representative verification information for the plurality of second samples, , The similarity representative verification information can be derived in the same way as the method of deriving the similarity representative information through steps S100 to S130 described above.

In addition, the test data for testing the first inference model may include similarity representative test information for the plurality of first samples and similarity representative test information for the plurality of second samples, and the similarity representative test information may be derived in the same way as the method of deriving similarity representative information through steps S100 to S130 described above.

In this way, when learning data, verification data, and test data are derived for the first inference model, each data may be derived at a predetermined ratio, and preferably, a plurality of data included in the learning data, verification data, and test data The ratio of the first sample and the plurality of second samples may be 8:1:1, respectively.

12 schematically illustrates a process of deriving similarity representative learning information according to an embodiment of the present invention.

Specifically, in (A) of FIG. 12, a plurality of drugs already known for a specific side effect (green triangle), a plurality of drugs paired with a corresponding specific side effect (green triangle) in the learning data, and a corresponding corresponding side effect in the test or verification data Multiple drugs paired with specific side effects (green triangles) are shown.

On the other hand, in (B) of FIG. 12, in order to derive similarity representative information between a specific drug (red-type drug) included in the learning data and the corresponding specific side effect (green triangle), information that is already known about the specific side effect (green triangle) A plurality of first similarity values are derived for each of a plurality of categories between a plurality of drugs and the specific drug (red-type drug), and the maximum value among the plurality of first similarity values for each category is extracted as the first similarity representative information of the corresponding category. do.

On the other hand, as shown in (B) of FIG. 12, if the relationship between the specific drug (red-type drug) and the specific side effect (green triangle) is already known, the first relationship between the specific drug (red-type drug) A similarity value may not be derived.

In addition, although not shown in (B) of FIG. 12, a plurality of second similarity values between the specific side effect (green triangle) and a plurality of known side effects for the specific drug (red series drug) can be derived, , The maximum value among the plurality of second similarity values may be extracted as the second similarity representative information. Similarly, when the relationship between the specific drug (red-type drug) and the specific side effect (green triangle) is already known, the second similarity value between the specific side effect (green triangle) may not be derived.

Similarly, as shown in (C) of FIG. 12, similarity representative information for test data or verification data can also be derived in a similar way.

13 schematically illustrates Area Under the ROC Curve (AUC) values for various configurations of the inference model unit 1700 according to an embodiment of the present invention.

In (A) of FIG. 13, each inference model unit 1700 according to the configuration of a plurality of categories in the step of deriving the first similarity value (S100) and whether or not the step of deriving the second similarity value (S120) is included. ) corresponds to a drawing showing Area Under the ROC Curve (AUC) values.

Specifically, 'Base' in (A) of FIG. 13 corresponds to the reasoning model unit 1700 considering only the similarity to the 5th category (Target) and 6th category (chemical structure of drugs) used in the conventional method. 'Base + SE-AH' corresponds to the reasoning model unit 1700 that additionally considers the similarity between the predicted side effect and the existing side effect in the configuration for 'Base', and 'Base + DDIs-N' corresponds to 'Base' Corresponds to the inference model unit 1700 in which the similarity to the second category described above is additionally considered in the configuration for 'Base', and 'Base + SNPs' is inference in which the similarity to the third category described above in the configuration for 'Base' is additionally considered. Corresponds to the model unit 1700, 'Base + DDIs-D' corresponds to the inference model unit 1700 that additionally considers the similarity to the first category described above in the configuration for 'Base', and 'All features' It corresponds to the reasoning model unit 1700 considering all the above-mentioned categories and the similarity between the predicted target side effect and the existing side effect.

As for the AUC value of the inference model unit 1700 according to each configuration shown in FIG. 13(A), the highest value was derived in the case of 'All features', and therefore, compared to the conventional method ('Base'), this It can be confirmed that predicting side effects by considering various items of similarity in the invention can predict side effects more effectively.

13(B) shows a case in which the inference model unit 1700 includes only a single first inference model (Proposed method (RF)) and the inference model unit 1700 includes a plurality of first inference models and second inference models. It corresponds to the figure showing the AUC value for the case of including the inference model (proposed method (stacking)).

As shown in (B) of FIG. 13, the inference model unit 1700 uses a plurality of first inference models and second inference models as an ensemble learning method rather than predicting side effects using only a single first inference model. It can be confirmed that side effects can be predicted more effectively when included.

14 schematically illustrates the internal configuration of a computing device according to an embodiment of the present invention.

The computing device 1000 illustrated in FIG. 2 may include components of the computing device 11000 illustrated in FIG. 14 .

As shown in FIG. 14, a computing device 11000 includes at least one processor 11100, a memory 11200, a peripheral interface 11300, an input/output subsystem ( It may include at least an I/O subsystem (11400), a power circuit (11500), and a communication circuit (11600). In this case, the computing device 11000 may correspond to the computing device 1000 shown in FIG. 2 .

The memory 11200 may include, for example, high-speed random access memory, magnetic disk, SRAM, DRAM, ROM, flash memory, or non-volatile memory. . The memory 11200 may include a software module, a command set, or other various data necessary for the operation of the computing device 11000.

In this case, access to the memory 11200 from other components, such as the processor 11100 or the peripheral device interface 11300, may be controlled by the processor 11100.

Peripheral interface 11300 may couple input and/or output peripherals of computing device 11000 to processor 11100 and memory 11200 . The processor 11100 may execute various functions for the computing device 11000 and process data by executing software modules or command sets stored in the memory 11200 .

The input/output subsystem can couple various input/output peripherals to peripheral interface 11300. For example, the input/output subsystem may include a controller for coupling a peripheral device such as a monitor, keyboard, mouse, printer, or touch screen or sensor to the peripheral device interface 11300 as needed. According to another aspect, input/output peripherals may be coupled to the peripheral interface 11300 without going through the input/output subsystem.

The power circuit 11500 may supply power to all or some of the terminal's components. For example, power circuit 11500 may include a power management system, one or more power sources such as a battery or alternating current (AC), a charging system, a power failure detection circuit, a power converter or inverter, a power status indicator or power It may contain any other components for creation, management and distribution.

The communication circuit 11600 may enable communication with another computing device using at least one external port.

Alternatively, as described above, the communication circuit 11600 may include an RF circuit and transmit/receive an RF signal, also known as an electromagnetic signal, to enable communication with other computing devices.

The embodiment of FIG. 14 is only an example of the computing device 11000, and the computing device 11000 may omit some components shown in FIG. 14, further include additional components not shown in FIG. It may have a configuration or arrangement combining two or more components. For example, a computing device for a communication terminal in a mobile environment may further include a touch screen or a sensor in addition to the components shown in FIG. , Bluetooth, NFC, Zigbee, etc.) may include a circuit for RF communication. Components that may be included in the computing device 11000 may be implemented as hardware including one or more signal processing or application-specific integrated circuits, software, or a combination of both hardware and software.

Methods according to embodiments of the present invention may be implemented in the form of program instructions that can be executed through various computing devices and recorded in computer readable media. In particular, the program according to the present embodiment may be composed of a PC-based program or a mobile terminal-specific application. An application to which the present invention is applied may be installed in the computing device 1000 through a file provided by a file distribution system. For example, the file distribution system may include a file transmission unit (not shown) that transmits the file according to a request of the computing device 1000 .

The device described above may be implemented as a hardware component, a software component, and/or a combination of hardware components and software components. For example, devices and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA) , a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. The processing device may run an operating system (OS) and one or more software applications running on the operating system. A processing device may also access, store, manipulate, process, and generate data in response to execution of software. For convenience of understanding, there are cases in which one processing device is used, but those skilled in the art will understand that the processing device includes a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it can include. For example, a processing device may include a plurality of processors or a processor and a controller. Other processing configurations are also possible, such as parallel processors.

Software may include a computer program, code, instructions, or a combination of one or more of the foregoing, which configures a processing device to operate as desired or processes independently or collectively. You can command the device. Software and/or data may be any tangible machine, component, physical device, virtual equipment, computer storage medium or device, intended to be interpreted by or to provide instructions or data to a processing device. , or may be permanently or temporarily embodied in a transmitted signal wave. Software may be distributed on networked computing devices and stored or executed in a distributed manner. Software and data may be stored on one or more computer readable media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program commands recorded on the medium may be specially designed and configured for the embodiment or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. - includes hardware devices specially configured to store and execute program instructions, such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter, as well as machine language codes such as those produced by a compiler. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

According to one embodiment of the present invention, since the first similarity value is derived by categorizing the drug-to-drug relationship for each of the drug to be predicted and a plurality of existing drugs, the predicted side effects of the drug to be predicted more accurately It can exert the effect of predicting the possibility of expression.

According to one embodiment of the present invention, since the first similarity value is derived by categorizing the node information between the drug to be predicted and a plurality of target genes targeted by the drug for each of the plurality of existing drugs, the drug to be predicted more accurately It can exert the effect of predicting the possibility of occurrence of the predicted target side effect for

According to an embodiment of the present invention, since the first similarity value is derived by categorizing expression gene information in which SNPs for each of one or more drug-related diseases are expressed for each of the drug to be predicted and a plurality of existing drugs, the first similarity value is derived more accurately It can exert the effect of predicting the possibility of occurrence of the predicted target side effects for the predicted target drug.

According to one embodiment of the present invention, since a second similarity value is derived between a predicted side effect and a plurality of existing side effects known for the predicted drug, the possibility of occurrence of the predicted side effect for the predicted drug can be more accurately predicted. effect can be exerted.

According to an embodiment of the present invention, the learned inference model unit for deriving information on the possibility of occurrence includes a second inference model implemented by ensemble learning that takes as input result values derived from a plurality of first inference models. , can exert the effect of improving the performance of the inference model unit.

As described above, although the embodiments have been described with limited examples and drawings, those skilled in the art can make various modifications and variations from the above description. For example, the described techniques may be performed in an order different from the method described, and/or components of the described system, structure, device, circuit, etc. may be combined or combined in a different form than the method described, or other components may be used. Or even if it is replaced or substituted by equivalents, appropriate results can be achieved.

Therefore, other implementations, other embodiments, and equivalents of the claims are within the scope of the following claims.

Claims (13)

As a method for predicting drug side effects, implemented as a computing device,
deriving a first similarity value for each of two or more categories between a predicted target drug and a plurality of existing drugs known to have predicted target side effects;
deriving first similarity representative information between the predicted target drug and the predicted target side effect by extracting a maximum value from the plurality of first similarity values for each category;
deriving each second similarity value between the predicted target side effect and a plurality of existing side effects known for the predicted target drug;
deriving second similarity representative information between the predicted target drug and the predicted target side effect by extracting a maximum value from the plurality of second similarity values; and
Entering the first similarity representative information and the second similarity representative information into the learned inference model unit to derive information about the possibility of occurrence of the predicted side effect for the predicted drug;
The categories include, interrelationships between one or more other drugs; node information for genes and proteins; Expression gene information for one or more genes; Symptom information for one or more symptoms or diseases; target gene information for a plurality of target genes; and information about the chemical structure of the drug; contains two or more of
The step of deriving the first similarity value,
As shown in Equation (1) below, for each category, the set of information on the corresponding category of the drug to be predicted (S _A ) and the set of intersection between the set of information on the corresponding category of each existing drug (S _B ) A value obtained by dividing the number of elements by the number of elements in the union is calculated as the first similarity value,

- formula (1)
The step of deriving the second similarity value,
For each of a plurality of anatomically separated hierarchies, the number of elements of the intersection between the set of hierarchical elements in which the predicted side effect is expressed (SE _a ) and the set of hierarchical elements in which each of the existing side effects is expressed (SE _b ) is the union Calculating a layer similarity value (S _l (SE _a , SE _b )) divided by the number of elements in; and
A value obtained by dividing the sum of the layer similarity values (S _l (SE _a , SE _b )) calculated for each of the plurality of layers by the number of the plurality of layers (n, where n is a natural number) is calculated as the second similarity value. Including;
The second similarity value for the set of hierarchical elements in which the predicted side effect is expressed (SE _a ) and the set of hierarchical elements in which each of the existing side effects is expressed (SE _b ) is calculated through the following equation (2),

- Eq. (2)
The layer similarity value (S _l (SE _a , SE _b )) can be calculated by Equation (1),
The l corresponds to a specific layer among a plurality of anatomically divided layers,
Wherein n corresponds to the number of a plurality of anatomically separated layers, a method for predicting side effects of a drug.
The method of claim 1,
The step of deriving the first similarity value,
Including as a first category the interaction between one or more other drugs that the drug affects or is affected by,
In the first category between the correlation between one or more other drugs that the predicted drug affects or is affected by and the correlation between one or more other drugs that each of the plurality of existing drugs affects or is affected by A method for predicting side effects of drugs, deriving a first similarity value for.
The method of claim 1,
The step of deriving the first similarity value,
Includes, as a second category, node information on genes and proteins included in the interaction pathway between a plurality of target genes targeted by the drug;
A method for predicting a side effect of a drug, wherein a first similarity value for the second category is derived between node information on the predicted drug and node information on each of the plurality of existing drugs.
The method of claim 1,
The step of deriving the first similarity value,
A third category includes expression gene information for one or more genes in which SNPs (Single Nucleotide Polymorphisms) related to each of one or more drug-related diseases are expressed,
A method for predicting side effects of drugs, wherein a first similarity value for the third category is derived between expression gene information for the predicted drug and expression gene information for each of the plurality of existing drugs.
The method of claim 1,
The step of deriving the first similarity value,
Symptom information on one or more symptoms or diseases related to drugs is classified into the 4th category and
Includes target gene information for a plurality of target genes targeted by the drug as a fifth category,
A first similarity value for the fourth category between the symptom information for the prediction target drug and the symptom information for each of the plurality of existing drugs, and
Deriving a first similarity value for the fifth category between the target gene information for the drug to be predicted and the target gene information for each of the plurality of existing drugs.
delete The method of claim 1,
The step of deriving the second similarity value,
For each of a plurality of anatomically divided classes, the side effects of drugs for deriving the second similarity value based on the type of layer elements in which the predicted side effect is expressed and the type of layer elements in which each of the existing side effects is expressed How to predict.
delete The method of claim 1,
The reasoning model unit,
a plurality of first inference models for deriving information on a first occurrence possibility of the predicted target drug for the predicted target drug based on the first similarity representative information and the second similarity representative information; and
A second inference model for deriving information on the second expression possibility of the predicted side effect for the predicted drug based on the information on the plurality of first expression possibilities derived by each of the plurality of first inference models Including, a method for predicting side effects of drugs.
The method of claim 9,
Each of the plurality of first inference models includes different inference algorithms,
The inference algorithm,
Methods for predicting side effects of drugs, including Random forest, Logistic regression, Naive Bayesian and XGBoost.
The method of claim 9,
The plurality of first inference models,
A plurality of first samples paired with any one first drug and any one side effect known to the first drug, and any one second drug and any one unknown side effect with respect to the second drug It is learned by the first similarity representative learning information and the second similarity representative learning information for each of a plurality of second samples paired,
The first similarity representative learning information and the second similarity representative learning information,
A method for predicting side effects of a drug, derived by the steps of claim 1 .
As a system for predicting drug side effects,
a first similarity derivation unit for deriving a first similarity value for each of two or more categories between a predicted target drug and a plurality of existing drugs known to have predicted target side effects;
a first similarity representative information derivation unit extracting a maximum value from a plurality of first similarity values for each category and deriving first similarity representative information between the predicted target drug and the predicted target side effect;
a second similarity derivation unit for deriving respective second similarity values between the predicted target side effect and a plurality of existing side effects known for the predicted target drug;
a second similarity representative information derivation unit extracting a maximum value from a plurality of second similarity values and deriving second similarity representative information between the prediction target drug and the prediction target side effect; and
An occurrence possibility prediction unit that inputs the first similarity representative information and the second similarity representative information to the learned inference model unit to derive information on the occurrence possibility of the predicted side effect for the predicted drug; and ,
The categories include, interrelationships between one or more other drugs; node information for genes and proteins; Expression gene information for one or more genes; Symptom information for one or more symptoms or diseases; target gene information for a plurality of target genes; and information about the chemical structure of the drug; contains two or more of
The first similarity derivation unit,
As shown in Equation (1) below, for each category, the set of information on the corresponding category of the drug to be predicted (S _A ) and the set of intersection between the set of information on the corresponding category of each existing drug (S _B ) A value obtained by dividing the number of elements by the number of elements in the union is calculated as the first similarity value,

- formula (1)
The second similarity derivation unit,
For each of a plurality of anatomically separated hierarchies, the number of elements of the intersection between the set of hierarchical elements in which the predicted side effect is expressed (SE _a ) and the set of hierarchical elements in which each of the existing side effects is expressed (SE _b ) is the union Calculating a layer similarity value (S _l (SE _a , SE _b )) divided by the number of elements in; and
A value obtained by dividing the sum of the layer similarity values (S _l (SE _a , SE _b )) calculated for each of the plurality of layers by the number of the plurality of layers (n, where n is a natural number) is calculated as the second similarity value. The step of doing;
The second similarity value for the set of hierarchical elements in which the predicted side effect is expressed (SE _a ) and the set of hierarchical elements in which each of the existing side effects is expressed (SE _b ) is calculated through the following equation (2),

- Eq. (2)
The layer similarity value (S _l (SE _a , SE _b )) can be calculated by Equation (1),
The l corresponds to a specific layer among a plurality of anatomically divided layers,
The system for predicting side effects of drugs, wherein n corresponds to the number of a plurality of anatomically separated layers.
As a computer-readable medium for implementing a method for predicting side effects of drugs performed on a computing device having one or more processors and one or more memories,
deriving a first similarity value for each of two or more categories between a predicted target drug and a plurality of existing drugs known to have predicted target side effects;
deriving first similarity representative information between the predicted target drug and the predicted target side effect by extracting a maximum value from the plurality of first similarity values for each category;
deriving each second similarity value between the predicted target side effect and a plurality of existing side effects known for the predicted target drug;
deriving second similarity representative information between the predicted target drug and the predicted target side effect by extracting a maximum value from the plurality of second similarity values; and
Entering the first similarity representative information and the second similarity representative information into the learned inference model unit to derive information about the possibility of occurrence of the predicted side effect for the predicted drug;
The categories include, interrelationships between one or more other drugs; node information for genes and proteins; Expression gene information for one or more genes; Symptom information for one or more symptoms or diseases; target gene information for a plurality of target genes; and information about the chemical structure of the drug; contains two or more of
The step of deriving the first similarity value,
As shown in Equation (1) below, for each category, the set of information on the corresponding category of the drug to be predicted (S _A ) and the set of intersection between the set of information on the corresponding category of each existing drug (S _B ) A value obtained by dividing the number of elements by the number of elements in the union is calculated as the first similarity value,

- formula (1)
The step of deriving the second similarity value,
For each of a plurality of anatomically separated hierarchies, the number of elements of the intersection between the set of hierarchical elements in which the predicted side effect is expressed (SE _a ) and the set of hierarchical elements in which each of the existing side effects is expressed (SE _b ) is the union Calculating a layer similarity value (S _l (SE _a , SE _b )) divided by the number of elements in; and
A value obtained by dividing the sum of the layer similarity values (S _l (SE _a , SE _b )) calculated for each of the plurality of layers by the number of the plurality of layers (n, where n is a natural number) is calculated as the second similarity value. Including;
The second similarity value for the set of hierarchical elements in which the predicted side effect is expressed (SE _a ) and the set of hierarchical elements in which each of the existing side effects is expressed (SE _b ) is calculated through the following equation (2),

- Eq. (2)
The layer similarity value (S _l (SE _a , SE _b )) can be calculated by Equation (1),
The l corresponds to a specific layer among a plurality of anatomically divided layers,
Wherein n corresponds to the number of a plurality of anatomically separated layers, a computer-readable medium.

Images