WO2022265480A1 - 약물들 간 상호작용을 분석하기 위한 방법 및 장치 - Google Patents

약물들 간 상호작용을 분석하기 위한 방법 및 장치 Download PDF

Info

Publication number
WO2022265480A1
WO2022265480A1 PCT/KR2022/095058 KR2022095058W WO2022265480A1 WO 2022265480 A1 WO2022265480 A1 WO 2022265480A1 KR 2022095058 W KR2022095058 W KR 2022095058W WO 2022265480 A1 WO2022265480 A1 WO 2022265480A1
Authority
WO
WIPO (PCT)
Prior art keywords
drug
data set
drugs
data
side effect
Prior art date
Application number
PCT/KR2022/095058
Other languages
English (en)
French (fr)
Inventor
이지현
배상훈
최정규
이인섭
Original Assignee
닥터노아바이오텍 주식회사
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 닥터노아바이오텍 주식회사 filed Critical 닥터노아바이오텍 주식회사
Publication of WO2022265480A1 publication Critical patent/WO2022265480A1/ko

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00Machine learning
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16HHEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H20/00ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
    • G16H20/10ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to patients
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16HHEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16HHEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H70/00ICT specially adapted for the handling or processing of medical references
    • G16H70/40ICT specially adapted for the handling or processing of medical references relating to drugs, e.g. their side effects or intended usage

Definitions

  • the present invention relates to drug-drug interaction (DDI) analysis, and more particularly, to a method and apparatus for analyzing drug-drug interaction using an artificial intelligence algorithm.
  • DCI drug-drug interaction
  • Drug-Drug Interaction is one of the major considerations in drug development.
  • drug interaction refers to a phenomenon that occurs when the efficacy or toxicity of one drug is modified by another drug, food, or environmental chemicals.
  • the interaction between drugs means that they affect each other when two or more drugs are used together at the same time or with a short interval in clinical practice.
  • the domestic combination drug clinical trial standard in terms of safety, 'considering the pathological mechanism and treatment mechanism, the safety and effectiveness of combined administration for drugs with the possibility of pharmacokinetic and pharmacodynamic interactions when administered in combination with individual active ingredients.
  • a close evaluation is required'. That is, in evaluating drug interactions, both the direct effects of drugs on the body and the effects of reactions between drugs should be considered.
  • there are more than thousands of types of existing drugs, and even data that can determine whether a combination of drugs is appropriate is incomplete, as it is revealed that the approval requirements for interactions between drugs vary by country and institution. The situation is.
  • the present invention is to provide a method and apparatus for effectively analyzing drug-drug interaction (DDI) using an artificial intelligence algorithm.
  • DAI drug-drug interaction
  • An object of the present invention is to provide a method and apparatus for reducing the time, human, and material resources required for analyzing interactions between drugs using artificial intelligence algorithms.
  • the present invention is to provide a method and apparatus for quickly predicting an appropriate drug combination for a combination drug based on an artificial intelligence algorithm.
  • An object of the present invention is to provide a method and apparatus for quickly recommending an appropriate drug combination for combination drugs based on an artificial intelligence algorithm.
  • a method for analyzing drug-drug interaction includes, as data sets for learning, a first data set for chemical structures of drugs, the above Obtaining a second data set on the grade of side effects between drugs and a third data set on the type of side effects between drugs, pre-processing the first data set to obtain details of each of the drugs Generating attribute information, preprocessing the second data set, normalizing classes included in the second data set, and assigning a directionality to the second data set, preprocessing the third data set, extracting expressions representing side effect types included in the third data set, normalizing the expressions, and assigning directionality to the third data set; the preprocessed first data set and the preprocessed second data set , learning at least one artificial intelligence model using the preprocessed third data set, and level of side effects between the pair of drugs from information on the pair of drugs using the at least one artificial intelligence model and determining the type.
  • DAI drug-drug interaction
  • the step of learning the at least one artificial intelligence model may include matching the preprocessed first data set with the preprocessed second data set and the preprocessed third data set, It may include generating a learning data set in which the level and type of side effects are mapped with respect to the attribute combination of .
  • the step of determining the level and type of side effects between the pair of drugs from the information on the pair of drugs may include pre-processing the information on the pair of drugs, Generating detailed attribute information of each drug, and inputting the detailed attribute information as input data of the at least one artificial intelligence model.
  • the detailed attribute information includes BDSI (Binary data of Drug Structural Information), ISD (Index of Similarity between Drugs), IIPD (Index of Interaction between Protein and Drug), IISD ( Index of Interaction Similarity between Drugs), Absortion Distribution Metabolism Excretion Toxicity (ADMET).
  • BDSI Binary data of Drug Structural Information
  • ISD Index of Similarity between Drugs
  • IIPD Index of Interaction between Protein and Drug
  • IISD Index of Interaction Similarity between Drugs
  • ADMET Absortion Distribution Metabolism Excretion Toxicity
  • the second data set includes side effect level data between first drugs collected from a first source and side effect level data between second drugs collected from a second source, wherein the The side effect level data between the first drugs and the side effect level data between the second drugs indicate the same class with different expressions, and the different expressions indicating the same class may be normalized through the preprocessing.
  • the third data set includes a first sentence expressing the type of side effect of the first drug pair and a second sentence expressing the type of side effect of the second drug pair, wherein the Each of the first sentence and the second sentence includes an expression indicating at least one type, and the first sentence and the second sentence include different expressions indicating a type with the same meaning, and have the same meaning. Different expressions indicating types may be replaced with one term through the preprocessing.
  • the second data set includes an item including side effect level information for a drug pair combined in the order of a first drug and a second drug
  • the preprocessed second data set may be processed to further include side effect level information on the drug pair combined in the order of the second drug and the first drug by the imparting of the directionality
  • the third data set includes an item including side effect type information for a drug pair combined in the order of a first drug and a second drug
  • the preprocessed third data set may be processed to further include side effect type information on the drug pair combined in the order of the second drug and the first drug by the directionality.
  • the at least one artificial intelligence model includes a multi-input, single-output first artificial intelligence model predicting the side effect level and a multi-input, multi-output second artificial intelligence model predicting the side effect type. models can be included.
  • the method may further include transmitting data indicating the level and type of side effects between the pair of drugs to another device.
  • An apparatus for analyzing drug-drug interaction (DDI) may include a memory storing at least one artificial intelligence model, and a processor connected to the memory.
  • the processor as data sets for learning, a first data set for the chemical structure of drugs, a second data set for the grade of side effects between the drugs, and a type of side effect between the drugs Obtaining a third data set for , generating detailed attribute information of each of the drugs by preprocessing the first data set, and preprocessing the second data set to classify the classes included in the second data set
  • By normalizing, assigning a direction to the second data set, and preprocessing the third data set expressions representing side effect types included in the third data set are extracted, the expressions are normalized, and the third data set A direction is given to a set, the at least one artificial intelligence model is learned using the preprocessed first data set, the preprocessed second data set, and the preprocessed third data set, and the at least one artificial intelligence Control can be performed to determine the level and type
  • the time, human, and material resources required for drug-drug interaction (DDI) analysis can be reduced.
  • FIG. 1 shows the structure of a drug-drug interaction (DDI) analysis system according to an embodiment of the present invention.
  • FIG. 2 shows the structure of an artificial neural network applicable to a system according to an embodiment of the present invention.
  • FIG 3 shows the structure of a system according to an embodiment of the present invention.
  • FIG. 4 illustrates a concept of acquiring learning data and independent variables in a system according to an embodiment of the present invention.
  • FIG. 5 shows a functional structure of a system according to an embodiment of the present invention.
  • BDSI Binary data of Drug Structural Information
  • FIG. 7 illustrates an example of an operation of deriving an Index of Similarity between Drugs (ISD) from a chemical structure according to an embodiment of the present invention.
  • FIG 8 illustrates an example of an operation of matching attribute information and level/type information according to an embodiment of the present invention.
  • FIG. 9 illustrates an example of an operation of learning and evaluating an artificial intelligence model according to an embodiment of the present invention.
  • 10A and 10B show an example of an artificial intelligence model for a side effect type system according to an embodiment of the present invention.
  • 11A and 11B show an example of an artificial intelligence model for a side effect grade system according to an embodiment of the present invention.
  • FIG. 12 illustrates an example of an artificial intelligence model for a side effect level system according to an embodiment of the present invention.
  • FIG. 13 illustrates a circular operation of prediction-verification-learning of an artificial intelligence model according to an embodiment of the present invention.
  • FIG. 14 illustrates an example of an artificial intelligence model for a side effect type system according to an embodiment of the present invention.
  • 15 illustrates a procedure for analyzing interactions between drugs in a system according to an embodiment of the present invention.
  • FIG. 16 illustrates an embodiment of a procedure for learning and prediction in a system according to an embodiment of the present invention.
  • FIG. 17 illustrates an embodiment of a procedure for performing learning in a system according to an embodiment of the present invention.
  • FIG. 18 illustrates an embodiment of a procedure for performing prediction in a system according to an embodiment of the present invention.
  • the present invention proposes a technique for analyzing drug-drug interaction (DDI) using an intelligent artificial intelligence algorithm.
  • the present invention is to provide a system for analyzing interactions between drugs under various environments such as a cloud environment and a local environment.
  • a system may be named 'CombiRisk'.
  • CombiRisk is a system that can quickly predict and recommend an appropriate drug combination for complex drugs to users by analyzing interactions between drugs based on big data and artificial intelligence technology, which are the core technologies of the 4th industry.
  • the CombiRisk system based on domestic and international drug big data and deep learning technology, is a decision-making support system that predicts compatibility between the main ingredients of drugs, and is designed to help more efficiently deploy time, human, and material resources for combination drug research. will be.
  • FIG 1 shows the structure of an interaction analysis system between drugs according to an embodiment of the present invention.
  • the system includes a user device 110a, a user device 110b, and a server 120 connected to a communication network.
  • FIG. 1 illustrates two user devices 110a and 110b, three or more user devices may exist.
  • the user device 110a and the user device 110b are end devices used by a user who intends to perform interaction analysis between drugs using the system according to an embodiment of the present invention.
  • the user device 110a and the user device 110b may acquire input data (eg, information on drugs that are objects of interaction analysis) and transmit the input data to the server 120 through a communication network.
  • Each of the user devices 110a and 110b may include a communication unit for communication, a storage unit for storing data and programs, a display unit for displaying information, an input unit for user input, and a processor for control.
  • each of the user devices 110a and 110b may be a general-purpose device (eg, a smart phone, tablet, laptop computer, or desktop computer) installed with an application or program for system access or a sys-only access terminal.
  • the server 120 performs calculations for analyzing interactions between drugs according to embodiments of the present invention.
  • the server 120 may provide various functions for an interaction analysis system between drugs and operate an artificial intelligence model.
  • An example of an artificial neural network applicable to the present invention will be described with reference to FIG. 2 below.
  • the server 120 may perform learning for the artificial intelligence model using the learning data.
  • the server 120 may be a local server existing in a local network or a remote access server (eg, a cloud server) connected through an external network.
  • the server 120 may include a communication unit for communication, a storage unit for storing data and programs, and a processor for control.
  • an artificial neural network as shown in FIG. 2 may be understood as a structure of artificial intelligence models stored in the server 120.
  • an artificial neural network includes an input layer 210, at least one hidden layer 220, and an output layer 230.
  • Each of the layers 210, 220, and 230 is composed of a plurality of nodes, and each node is connected to an output of at least one node belonging to the previous layer.
  • Each node adds a bias to the inner product of each output value of the nodes in the previous layer and the corresponding connection weight, and then generates a non-linear activation function
  • the output value multiplied by is delivered to at least one neuron in the next layer.
  • Each layer can be further divided into input nodes, perceptrons, and output nodes.
  • the artificial neural network shown in FIG. 2 may be formed by learning (eg, machine learning, deep learning, etc.).
  • artificial neural network models used in various embodiments of the present invention include fully convolutional neural networks, convolutional neural networks, recurrent neural networks, and restricted Boltzmann machines. , RBM) and at least one of a deep belief neural network (DBN), but is not limited thereto.
  • RBM convolutional neural networks
  • DNN restricted Boltzmann machines
  • machine learning methods other than deep learning may also be included.
  • a hybrid model combining deep learning and machine learning may also be included.
  • a deep learning-based model may be applied to extract features of an image, and a machine learning-based model may be applied when the image is classified or recognized based on the extracted features.
  • the machine learning-based model may include a Support Vector Machine (SVM), AdaBoost, and the like, but is not limited thereto.
  • SVM Support Vector Machine
  • AdaBoost AdaBoost
  • a deep neural network can be applied, and interactions between drugs that may appear on pharmacokinetic (PK) can be predicted. That is, the system according to an embodiment of the present invention predicts the risk level and type of side effects due to interactions between new drugs by applying deep learning technology to interaction information between drugs, and provides predicted results.
  • a self-database may be established, and characteristics of each drug may be extracted based on this.
  • drug characteristics include drug structure information, structural similarity between drugs, absorption/distribution/metabolism/excretion/toxicity information (ADMET), and interaction information with proteins.
  • ADMET is information used as a criterion for describing the distribution of a drug in vivo.
  • a system according to an embodiment of the present invention may be composed of a DDI grade system that predicts risk in 5 steps and a DDI type system that predicts what type of DDI will occur. there is.
  • the server 120 includes a DDI type system 310 and a DDI level system 320 .
  • the DDI type system 310 may be implemented based on a dragbank database.
  • the DDI level system 320 may be implemented based on a drug bank database, a drugscom database, a combination prescription database of the Health Insurance Review and Assessment Service, a combination drug database, a combination contraindication database, and the like.
  • the DDI type system 310 provides the types of DDIs expected to occur. For example, the DDI type system 310 determines the risk of a specific symptom (e.g., rhabdomyolysis) when drug a as a subject drug and drug b as an affecting drug are combined. may predict an increase in risk or severity.
  • the DDI type system 310 may be referred to as 'RiskDescription system', 'RiskDescription system model', 'CombiType system', 'CombiType system model', 'side effect type system', and the like.
  • the DDI level system 320 may include a DNN for predicting risk.
  • the DDI level system 320 may be referred to as 'RiskGrade system', 'RiskGrade system model', 'CombiGrade system', 'CombiGrade system model', 'side effect level system', and the like.
  • the degree of risk may be classified into 5 levels. For example, the risk levels classified into 5 levels are shown in [Table 1] below.
  • learning data and independent variables may be obtained from the chemical structure of a drug.
  • Detailed attribute information (420-1, 420-2) can be obtained.
  • the detailed attribute information 420-1 and 420-2 may include BDSI, ISD, IIPD, IISD, and ADMET.
  • a concatenation 430 of feature vectors is generated based on the feature information 420-1 and 420-2.
  • the concatenation 430 of feature vectors may be used for training or predictive operation of an artificial intelligence model.
  • the system for predicting interactions between drugs uses a plurality of DDI databases (eg, Drugbank, Drugscom, public data portal, Health Insurance Review and Assessment Service, Korea Institute of Drug Safety and Management, etc.) in an integrated manner. Therefore, the risk of overfitting is significantly reduced as the artificial intelligence model is learned using various data without being biased toward one type of database.
  • the drug-drug interaction prediction system can predict the outcome of the drug-drug interaction and at the same time predict the severity of the drug-drug interaction. Accordingly, resources required for complex drug development and drug prescription can be utilized more efficiently.
  • the drug-to-drug interaction prediction system considers not only the response of the drug-to-drug interaction, but also the directionality of the interaction between two drugs, and determines the subject drug and the affected drug. drug) can provide predictions.
  • 5 shows a functional structure of a system according to an embodiment of the present invention. 5 may be understood as a functional configuration of the server 120 of FIG. 1 .
  • the server includes a data collection unit 510, a pre-processing unit 520, a data classification unit 530, a learning unit 540, an artificial intelligence model 550, an input data acquisition unit 560, and an analysis unit.
  • a section 570 is included.
  • Original data may include data in various forms and contents.
  • original data may include paid purchase data and public data.
  • original data may include three data sets.
  • the original data may include a drug chemical structure data set, a drug side effect level data set, and a drug side effect type data set.
  • the pre-processing unit 520 processes original data for learning. In other words, the pre-processing unit 520 processes the original data into a learnable form.
  • the preprocessor 520 may generate detailed attribute information indicating attributes of a drug from the drug structure data set.
  • the detailed attribute information includes BDSI (Binary data of Drug Structural Information), ISD (Index of Similarity between Drugs), IIPD (Index of Interaction between Protein and Drug), IISD (Index of Interaction Similarity between Drugs), ADMET ( Absortion Distribution Metabolism Excretion Toxicity).
  • the detailed attribute information may further include items other than BDSI, ISD, IIPD, IISD, and ADMET listed above, or at least one of the listed items may be replaced with another item.
  • the pre-processing unit 520 normalizes the data set of the side effect level between drugs and the type of side effect between drugs according to predefined criteria, assigns a direction, matches the detailed attribute information of the drug, and then determines the independent variable and the dependent variable. variables can be created.
  • the data classification unit 530 classifies the preprocessed data according to the use in the learning procedure of the artificial intelligence model 550 .
  • the data classification unit 530 may classify data into training data, verification data, and test data. Specifically, 60% of the preprocessed data may be classified as training data, 20% as verification data, and 20% as test data.
  • the learning unit 540 performs learning and evaluation of the artificial intelligence model 550 using training data, verification data, and test data provided from the data classification unit 530 .
  • an artificial intelligence model 550 may be trained and evaluated.
  • performance evaluation is performed using test data 930 .
  • the performance does not meet the required criterion (eg, accuracy greater than or equal to a critical rate)
  • re-learning may be performed.
  • the learning unit 540 performs learning using only some of the training data and verification data, performs evaluation using a part of the test data, and then learns using another part according to the evaluation result. can be performed additionally.
  • the artificial intelligence model 550 includes a deep neural network.
  • the deep neural network includes an input layer and an output layer, and includes at least one hidden layer. Each layer consists of at least one input node, at least one perceptron, and at least one output node.
  • a deep neural network can be quickly built by building a neural layer using a Python-based Keras library, a phytoch library, or the like, using an artificial intelligence development library.
  • a Python-based Keras library, a phytoch library, a tensorflow library, or the like may be utilized, or other programming languages (eg, JAVA, C, etc.) may be utilized.
  • Each of the DDI type system 310 and the DDI level system 320 included in the CombiRisk system may be designed to include about 6 to 8 layers.
  • each layer may include a batch normalization layer, a dense layer, and a dropout layer.
  • the batch normalization layer Through the batch normalization layer, the data is converted into a state that is better for learning, learning is performed in the dense layer, and the overfitting probability can be reduced in the dropout layer.
  • the number of perceptrons (neurons) in each layer can be designed between a minimum of 15 and a maximum of 2048. It is desirable to apply and test several algorithms to resolve the imbalance between classes in the training data. For example, a focal loss algorithm may be applied that lowers a loss value of a class with good prediction and slightly lowers a loss value of a class with poor prediction. In this case, learning is performed more intensively for classes with poor prediction.
  • the input data acquisition unit 560 acquires input data input to the artificial intelligence model 550 for prediction operation.
  • the input data includes drug information.
  • the drug information may include detailed information on properties of the drug (eg, BDSI, ISD, IIPD, IISD, ADMET, etc.).
  • the drug information may include a drug ID.
  • a pre-processing unit generating detailed attribute information by pre-processing the input data may be further included.
  • the pre-processing unit 520 that pre-processes the learning data may pre-process the input data.
  • the analysis unit 570 inputs detailed attribute information included in the input data or detailed attribute information generated from the input data to the artificial intelligence model 550, obtains output data of the artificial intelligence model 550, and obtains output. Generate analysis results based on the data.
  • the generated analysis result may be internally stored or transmitted to the outside (eg, the user device 110a or the user device 110b).
  • the drug chemical structure data set may include a drug identifier (ID) and a simplified molecular input line entry system (SMILES).
  • ID drug identifier
  • SMILES of each compound can be obtained from chemical substance databases such as Pubchem and Drugbank. However, since there are differences in the form of SMILES for each database, preprocessing may be required. However, the aforementioned SMILES is an example of a chemical structure data set, and other chemical structure data may be used for various embodiments. For example, replacing or paralleling SMILES, compound data data (mol file, mol2 file, sdf (structural-data file)), InChI (International Chemical Identifier), chemical formula, 3D structure information, etc. this can be used
  • the data set of side effect level between drugs includes a first drug ID, a second drug ID, and a level value representing the degree of side effect.
  • the level value is one of predefined candidate values, and each candidate value indicates one of the levels listed in [Table 1].
  • the cross-drug side effect type data set includes an ID of a first drug having an effect, an ID of a second drug having an effect, and type information indicating a type of side effect.
  • the type information may be expressed as a sentence describing what kind of side effect the first drug causes the second drug.
  • type information “sub-dug may decrease the anticoagulant activities of aff_drug”, “sub-dug may decrease the antihypertensive activities of aff_drug”, “sub-dug can cause a decrease in the absorption of aff_drug”, “sub -dug can cause an increase in the absorption of aff_drug”.
  • the data set of side effect level between drugs and the type of side effect between drugs is related to the combination of drugs included in the drug structure data set
  • more items than those included in the drug structure data set can include
  • the drug structure data set includes structural information of about 13,000 drug items
  • each of the drug-to-drug side effect level data set and the drug-to-drug side effect type data set may include about 1,500,000 interaction-related items. .
  • the pre-processing unit 520 generates detailed attribute information including BDSI, ISD, IIPD, IISD, and ADMET from the chemical structure of the drug.
  • BDSI basic attribute information
  • ISD ISD
  • IIPD IIPD
  • ADMET ADMET from the chemical structure of the drug.
  • BDSI represents unique information of compounds, and is information designed to confirm the characteristics or similarity of molecules in drugs.
  • the pre-processing unit 520 calculates each value according to the distance of each element and according to the structure and to which element it is combined, and expresses the calculated values as binary values. That is, BDSI represents the average molecular structure and characteristics of various molecules using surrounding elements. An example of a process of deriving BDSI from a chemical structure is shown in FIG. 6 below.
  • FIG. 6 shows an example of an operation of deriving a BDSI from a chemical structure according to an embodiment of the present invention.
  • 6 illustrates BDSI generation for compound 610.
  • structures 620 at distances 0, 2, and 4 are identified.
  • the identified structures 620 are converted into numerical IDs 630 .
  • the IDs 630 are converted into a list representation 640, and binary values corresponding to the IDs included in the list representation 640 are rearranged by a hash function, thereby generating the BDSI 650.
  • the number of binaries for expressing DBSI is adjustable, and since it is binary data, it has the advantage that fast operation is possible.
  • An example of the generated BSDI data set is shown in [Table 2] below.
  • drug ID BDSI DB00006 ⁇ 1, 0, 0, 0, 0... , 0, 1, 0, 1, 0 ⁇ DB00007 ⁇ 0, 0, 0, 0, 0... , 0, 0, 0, 0, 1, 0 ⁇ DB00014 ⁇ 1, 0, 0, 1, 0... , 0, 0, 0, 0, 0 ⁇ DB00027 ⁇ 0, 0, 0, 0, 0... , 0, 1, 0, 0, 0 ⁇ DB00035 ⁇ 0, 0, 1, 0, 0... , 0, 0, 0, 0, 1, 0 ⁇ DB00050 ⁇ 1, 0, 0, 0, 1... , 0, 1, 0, 0 ⁇ ... ...
  • ISD is generated based on BDSI and expresses the degree of structural similarity of compounds. That is, the pre-processing unit 520 calculates the similarity of compound structures between drugs based on the molecular structures represented by BDSI.
  • One ISD value per drug is generated. For example, if there are 10,000 drugs, each drug has 10,000 similarity values, as shown in FIG. 7 below, and 10,000 values arranged in order constitute one ISD. 7 illustrates an example of an operation of deriving an ISD from a chemical structure according to an embodiment of the present invention. Referring to FIG. 7 , for a drug 710 having an ID of DB00007, similarity values 720 with all drugs including itself are calculated. By listing the similarity values 730 in a predefined order (eg, ascending drug ID order), the ISD value 730 is generated. An example of the ISD data set generated through this process is shown in [Table 3] below.
  • the pre-processing unit 520 quantifies a series of reactions occurring between drugs and proteins based on the molecular structure expressed by BDSI. To this end, proteins known to play a major role in interactions between drugs and proteins are selected.
  • the pre-processing unit 520 may extract information on a total of 8 types of reactions according to drug and protein structures. For example, the eight types of reactions are: hydrophobic contacts, aromatic face to face, aromatic edge to face, and hydrogen bonding as a hydrogen bond donor. bond (protein as hydrogen bond donor), hydrogen bond (protein as hydrogen bond acceptor), salt bridges (protein positively charged), protein as negatively charged It may include salt bridges (protein negatively charged), salt bridges (ionic bond with metal ion), etc.
  • the IISD is information indicating the degree of similarity of IIPD between drugs.
  • the pre-processing unit 520 calculates the similarity of the IIPD between drugs based on the IIPD For example, if there are 10,000 drugs, each drug has 10,000 similarity values, and the 10,000 similarity values are ordered. The results arranged as above form the IISD of one drug The pre-processing unit 520 determines the IISD for each drug and creates an IISD data set including a plurality of IISDs.
  • ADMET is information that quantifies the level of absorption, distribution, metabolism, excretion, toxicity, etc. of a drug through changes in the concentration of the drug in the body over time from the viewpoint of pharmacokinetics.
  • the pre-processing unit 520 extracts molecular features of the drug from drug structural information, that is, SMILE, and then calculates ADMET.
  • a total of 28 ADMET values representing each of the 6 categories per drug are generated.
  • 28 ADMET values are 1 LogS, LodD, LogP related to basic physical and chemical properties, 2 Caco-2, Pgp-Inhibitor, HIA, F (20%), F (30%), and 3 distribution related to absorption.
  • the pre-processing unit 520 determines an ADMET value set for each drug and generates an IISD data set including a plurality of ADMET value sets.
  • the pre-processing unit 520 generates learning data for the DDI level system 320 from the drug-to-drug side effect level data set.
  • the pre-processing unit 520 may analyze database characteristics for class reclassification and perform pre-processing based on the analysis result. Also, the pre-processing unit 520 may give directionality to data.
  • the pre-processing unit 520 adds an item of “DB00001+DB06605 ⁇ Grade 1”. This is because, if only the data of "DB06605+DB00001 ⁇ Grade 1" is learned, a result other than Grade 1 can be predicted if the combination of "DB00001+DB06605" is entered.
  • the pre-processing unit 520 generates learning data for the DDI type system 310 from the drug-to-drug side effect type data set.
  • the pre-processing unit 520 extracts the type of side effect from type information included in the type data set of side effect between drugs. For example, when the type information is "sub_drug may decrease effectiveness of aff_drug", the preprocessor 520 may extract 'decrease' and 'effectiveness'. As another example, when the type information is "sub_drug may increase the QTc-prolonging activities of aff_drug", the preprocessor 520 may extract 'increase' and 'QTc-prolonging'. Then, the pre-processing unit 520 analyzes the type of side effects and performs pre-processing.
  • the preprocessing unit 520 may organize synonyms, similar side effects, and the like into unified terms. Also, the pre-processing unit 520 may give directionality to data. For example, if there is an item of "DB06605+DB00001 ⁇ increase, QTc-prolonging", the preprocessor 520 means that DB06605 is an affected drug (sub_drug) and DB0001 is an affecting drug (aff_drug), ' A directionality value of 0' may be assigned.
  • the pre-processing unit 520 adds an item of "DB00001+DB06605 ⁇ increase, QTc-prolonging", DB0001 is an affecting drug (aff_drug), and DB06605 is an affected drug (sub_drug). value can be assigned. That is, the direction value '0' means a combination of the former and the latter influencing drug, and the directionality value of '1' means a combination of the former and the latter.
  • the pre-processing unit 520 pre-processes the original data. Then, the data is separated into independent and dependent variables. In order to determine the independent variable and the dependent variable, the preprocessing unit 520 may match detailed attribute information and level/type data such as BDSI, ISD, IIPD, IISD, and ADMET from the chemical structure of the drug. The matching operation is as shown in FIG. 8 .
  • FIG. 8 illustrates an example of an operation of matching attribute information and level/type information according to an embodiment of the present invention.
  • detailed attribute information 810 for each drug ID including BDSI, ISD, IIPD, IISD, and ADMET for each drug ID generated from drug chemical structure information (eg, SMILES) is generated.
  • drug chemical structure information eg, SMILES
  • the DDI level data 820a for each drug ID pair and the DDI type data 820b for each drug ID pair are matched with the detailed attribute information 810 for each drug ID.
  • DDI level/type data for each BDSI pair 830a
  • DDI level/type data for each ISD pair 830b
  • DDI level/type data for each IIPD pair 830c
  • DDI level/type data for each IISD pair 830d
  • DDI level/type data 830e for each ADMET pair is generated.
  • DDI level/type data 830a for each BDSI pair are shown in [Table 4] and [Table 5] below.
  • subject drug ID affected drug ID label ⁇ 1, 0, 0, 0, 0... , 0, 1, 0, 1, 0 ⁇ ⁇ 1, 0, 0, 1, 0... , 0, 0, 0, 0, 0 ⁇ 0 ⁇ 1, 0, 0, 0, 0... , 0, 1, 0, 1, 0 ⁇ ⁇ 0, 0, 1, 0, 0... , 0, 0, 0, 0, 1, 0 ⁇ 0 ⁇ 0, 0, 1, 0, 0... , 0, 0, 0, 0, 1, 0 ⁇ ⁇ 1, 0, 0, 1, 0... , 0, 0, 0, 0, 0 ⁇ One ⁇ 0, 0, 1, 0, 0... , 0, 0, 0, 1, 0 ⁇ ⁇ 1, 0, 0, 0, 0, 1,... , 0, 1, 0, 0 ⁇ 2 ... ... ... ... ...
  • subject drug ID affected drug ID modified summary ⁇ 1,0,0,0,0... ,0,1,0,1,0 ⁇ ⁇ 1,0,0,1,0... ,0,0,0 ⁇
  • the therapeutic efficacy ... ⁇ 1,0,0,0,0... ,0,1,0,1,0 ⁇ ⁇ 0,0,1,0,0... ,0,0,0,1,0 ⁇
  • the therapeutic efficacy ... ⁇ 0,0,1,0,0... ,0,0,0,1,0 ⁇ ⁇ 1,0,0,1,0... ,0,0,0,0 ⁇ subject drug can cause... ⁇ 0,0,1,0,0...
  • the preprocessor 520 determines independent variables and dependent variables from data sets generated through matching. . For example, independent variables such as BDSI, ISD, IIPD, IISD, and ADMET for drug pairs can be generated as shown in [Table 6] to [Table 10].
  • subject drug ID affected drug ID DB06605_BDSI DB00001_BDSI DB06695_BDSI DB00001_BDSI DB01254_BDSI DB00001_BDSI DB01609_BDSI DB00001_BDSI DB01586_BDSI DB00001_BDSI DB02659_BDSI ... ...
  • 'DB*****_DBSI' means the BDSI value of a drug whose drug ID is DB*****.
  • subject drug ID affected drug ID DB06605_ISD DB00001_ISD DB06695_ISD DB00001_ISD DB01254_ISD DB00001_ISD DB00001_ISD DB01609_ISD DB00001_ISD DB01586_ISD DB00001_ISD DB02659_ISD ... ...
  • 'DB*****_ISD' means the ISD value of a drug whose drug ID is DB*****.
  • subject drug ID affected drug ID DB06605_IIPD DB00001_IIPD DB06695_IIPD DB00001_IIPD DB01254_IIPD DB00001_IIPD DB01609_IIPD DB00001_IIPD DB01586_IIPD DB00001_IIPD DB02659_IIPD ... ...
  • 'DB*****_IIPD' means the IIPD value of a drug whose drug ID is DB*****.
  • subject drug ID affected drug ID DB06605_IISD DB00001_IISD DB06695_IISD DB00001_IISD DB01254_IISD DB00001_IISD DB01609_IISD DB00001_IISD DB01586_IISD DB00001_IISD DB02659_IISD ... ...
  • 'DB*****_IISD' means the IISD value of a drug whose drug ID is DB******.
  • 'DB*****_ADMET' means the ADMET value of a drug whose drug ID is DB*****.
  • items for level and type are extracted as dependent variables. For example, dependent variables including a level class of a single output as shown in [Table 11] below and a type class of multiple outputs as shown in [Table 12] below can be created.
  • 10A and 10B show an example of an artificial intelligence model for a side effect type system according to an embodiment of the present invention.
  • 10A and 10B illustrate an artificial intelligence model for a DDI type system 310.
  • detailed attribute information 1004 including BDSI, ISD, IIPD, IISD, and ADMET is determined from drug SMILES data 1002 by a preprocessing process 1010.
  • the detailed attribute information 1004 is provided to the artificial intelligence model 1020 as training data.
  • the artificial intelligence model 1020 includes a plurality of layers, and each layer includes a batch normalization (BN) layer, a dense layer, and a dropout layer.
  • the side effect type 1008 is determined by prediction using an artificial intelligence model 1020.
  • Side effect type 1008 takes the form of multiple outputs.
  • the drug side effect type data 1006 is provided as an output layer of the artificial intelligence model 1020 after going through a preprocessing process 1030, and the artificial intelligence model 1020 is learned by back-propagation operation.
  • the structure of the output layer may depend on the shape of the dependent variable.
  • 11A and 11B show an example of an artificial intelligence model for a side effect level system according to an embodiment of the present invention.
  • 11A and 11B illustrate an artificial intelligence model for the DDI level system 320.
  • detailed attribute information 1104 including BDSI, ISD, IIPD, IISD, and ADMET is determined from drug SMILES data 1102 by a preprocessing process 1110.
  • the detailed attribute information 1104 is provided to the artificial intelligence model 1120 as training data.
  • the artificial intelligence model 1120 includes a plurality of layers, and each layer includes a batch normalization (BN) layer, a dense layer, and a dropout layer.
  • the level of side effects 1108 is determined by prediction using an artificial intelligence model 1120 .
  • the side effect level 1108 takes the form of a single output.
  • the drug side effect level data 1106 is provided as an output layer of the artificial intelligence model 1120 after going through a preprocessing process 1130, and the artificial intelligence model 1120 is learned by back-propagation operation.
  • the structure of the output layer may depend on the shape of the dependent variable.
  • the artificial intelligence model 1250 for the side effect level system has a multi-input, single-output form. Accordingly, when input data 1202 including BDSI, ISD, IIPD, IISD, and ADMET is input, output data 1204 indicating one side effect level is output.
  • the artificial intelligence model 1250 predicts a grade class by analyzing patterns of independent variables such as BDSI, ISD, IIPD, IISD, and ADMET of drug pairs. Additionally, as shown in FIG. 13 , the artificial intelligence model 1250 may verify the result of the prediction 1301 by itself (1303), and may further proceed with learning (1305) while feeding back the verification contents. For example, if the predicted result is ⁇ Drug_1, Drug_2 ⁇ Class '1' ⁇ , the interpretation is ⁇ Drug 1 and Drug 2 have a 'normal level' side effect probability when used together ⁇ .
  • the artificial intelligence model 1450 for the side effect type system has the form of multiple inputs and multiple outputs. Accordingly, when input data 1402 including BDSI, ISD, IIPD, IISD, and ADMET is input, output data 1404 representing side effect types represented by a plurality of items is output.
  • the artificial intelligence model 1450 analyzes patterns of independent variables such as BDSI, ISD, IIPD, IISD, and ADMET of drug pairs and predicts type classes.
  • a pair of drugs e.g., drug 1 (1502-1), drug 2 (1502-2)
  • drug 1 1502-1
  • drug 2 1502-2
  • the input data includes attribute data used for learning by models such as BDSI, ISD, IIPD, IIPDSP, and ADMET of the drug.
  • the RiskDescription system model 1550b also predicts directionality, it provides how the effect of one drug changes the side effects of another drug.
  • the RiskGrade system model (1550a) predicts the predicted result of ⁇ Drug_1, Drug_2 ⁇ Class '1' ⁇
  • the RiskDescription system model (1550b) predicts ⁇ Drug_1(subject_drug), Drug_2(affected_drug) ⁇ Type: 'increase', Type: 'CNS depression', Type: 'hypotesion' predictive results can be output.
  • the conclusion (1506) is ⁇ Drug No. 1 and Drug No. 2 have a 'normal level' side effect probability when used together, and the side effects of 'central nervous system depression' and 'low blood pressure' of Drug No. 1 due to the effect of Drug No. 2 are ' May include 'may increase'.
  • 16 illustrates an embodiment of a procedure for learning and prediction in a system according to an embodiment of the present invention.
  • the subject of operation is described as a 'device', but operations described below may be performed by a server or a user device.
  • Data is data for learning, and may include, for example, a drug chemical structure data set, a drug side effect level data set, and a drug side effect type data set.
  • step S1603 the device performs learning.
  • the device may pre-process data and perform learning using the pre-processed data.
  • step S1605 the device performs prediction. That is, the device obtains output data including a predicted result from input data including information on a drug pair to be analyzed using the learned artificial intelligence model. At this time, the device may pre-process information on the drug pair in a format that can be input to the artificial intelligence model.
  • FIG. 17 illustrates an embodiment of a procedure for performing learning in a system according to an embodiment of the present invention.
  • the subject of operation is described as a 'device', but operations described below may be performed by a server or a user device.
  • step S1701 the device determines detailed attribute information based on the chemical structure. That is, as a pre-processing of drug chemical structure data, the device generates detailed attribute information based on the chemical structure of the drug, for example, attribute information of at least one of BDSI, ISD, IIPD, IISD, and ADMET.
  • step S1703 the device reclassifies the class data and assigns a direction. That is, as a pre-processing of inter-drug side effect type data, the device reclassifies level class values in a predefined format. That is, the device normalizes level information according to drug combinations collected from different sources according to a predefined format into a unified format. In addition, the device provides directionality by adding an item obtained by changing the order of drug combination to the data set.
  • step S1705 the device extracts side effect types from type data, organizes expressions, and gives directions. That is, as a pre-processing of data on the level of side effects between drugs, the device extracts keywords representing types from data in the form of sentences. And, the device replaces expressions in a synonym or synonym relationship with a representative expression. In addition, the device provides directionality by adding an item in which the order of drug combinations is changed to the data set.
  • step S1707 the device performs learning based on the preprocessed data.
  • the device generates a data set mapping the combination of drug attributes and the type/level of side effects by matching the preprocessed data sets.
  • the device performs learning using the created data set. That is, the device performs learning using learning data having drug attribute combinations as independent variables and side effect types/levels as dependent variables.
  • the device performs training on the artificial intelligence model using, as training data, attribute combination information of drugs labeled with side effect type/level. That is, after performing prediction using the learning data, the device updates the weights of the artificial intelligence model through backpropagation.
  • FIG. 18 illustrates an embodiment of a procedure for performing prediction in a system according to an embodiment of the present invention.
  • the subject of operation is described as a 'device', but operations described below may be performed by a server or a user device.
  • step S1801 the device acquires input data.
  • Input data may be input through an input means provided in the device (eg, an input device such as a keyboard, a port connectable to an external storage medium, an interface for receiving signals through a communication network, etc.).
  • an input means eg, an input device such as a keyboard, a port connectable to an external storage medium, an interface for receiving signals through a communication network, etc.
  • step S1803 the device checks the chemical structure of the drug included in the input data. At this time, if the input data does not include chemical structure information (eg SMILES) but includes drug identification information (eg name, ID, etc.), the device converts the chemical structure corresponding to the identification information from an internal or external database. You can search. To this end, the device may connect to the database, transmit a request including identification information of the drug, and receive chemical structure information as a response.
  • chemical structure information eg SMILES
  • drug identification information eg name, ID, etc.
  • step S1805 the device determines detailed property information of the drug based on the chemical structure.
  • the device generates detailed attribute information, for example, attribute information of at least one of BDSI, ISD, IIPD, IISD, and ADMET, based on the chemical structure of the drug.
  • the device acquires output data through prediction.
  • the device generates output data including prediction results from detailed attribute information of a pair of drugs using an artificial intelligence model.
  • the output data may include at least one of side effect level data and side effect type data as a prediction result.
  • the output data may be converted into textual results that users can more easily understand.
  • step S1809 the device provides output data.
  • the output data is provided to the user requesting or ordering the analysis.
  • the output data may be visually output through an output means (eg, a screen, etc.) provided in the device, or may be transmitted to an external device.
  • Exemplary methods of the present invention are presented as a series of operations for clarity of explanation, but this is not intended to limit the order in which steps are performed, and each step may be performed concurrently or in a different order, if desired.
  • other steps may be included in addition to the exemplified steps, other steps may be included except for some steps, or additional other steps may be included except for some steps.
  • various embodiments of the present invention may be implemented by hardware, firmware, software, or a combination thereof.
  • ASICs application specific integrated circuits
  • DSPs digital signal processors
  • DSPDs digital signal processing devices
  • PLDs programmable logic devices
  • FPGAs field programmable gate arrays
  • ASICs application specific integrated circuits
  • DSPs digital signal processors
  • DSPDs digital signal processing devices
  • PLDs programmable logic devices
  • FPGAs field programmable gate arrays
  • It may be implemented by a processor (general processor), controller, microcontroller, microprocessor, or the like.
  • the scope of the present invention is software or machine-executable instructions (eg, operating systems, applications, firmware, programs, etc.) that cause operations according to methods of various embodiments to be executed on a device or computer, and such software or includes a non-transitory computer-readable medium in which instructions and the like are stored and executable on a device or computer;

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Epidemiology (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Medicinal Chemistry (AREA)
  • Theoretical Computer Science (AREA)
  • Chemical & Material Sciences (AREA)
  • Software Systems (AREA)
  • Toxicology (AREA)
  • Physics & Mathematics (AREA)
  • Pathology (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Databases & Information Systems (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Pharmacology & Pharmacy (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

본 발명은 약물들 간 상호작용을 분석하기 위한 방법 및 장치에 대한 것으로, 장치의 동작 방법은, 학습을 위한 데이터 셋(set)들로서, 약물들의 화학구조에 대한 제1 데이터 셋, 상기 약물들 간 부작용 수준(grade)에 대한 제2 데이터 셋, 상기 약물들 간 부작용 유형(type)에 대한 제3 데이터 셋을 획득하는 단계, 상기 제1 데이터 셋을 전처리함으로써, 상기 약물들 각각의 세부 속성 정보를 생성하는 단계, 상기 제2 데이터 셋을 전처리함으로써, 상기 제2 데이터 셋에 포함된 클래스를 정규화하고, 상기 제2 데이터 셋에 방향성을 부여하는 단계, 상기 제3 데이터 셋을 전처리함으로써, 상기 제3 데이터 셋에 포함된 부작용 유형을 나타내는 표현들을 추출하고, 상기 표현들을 정규화하고, 상기 제3 데이터 셋에 방향성을 부여하는 단계, 상기 전처리된 제1 데이터 셋, 상기 전처리된 제2 데이터 셋, 상기 전처리된 제3 데이터 셋을 이용하여 적어도 하나의 인공지능 모델을 학습하는 단계, 및 상기 적어도 하나의 인공지능 모델을 이용하여 한 쌍의 약물들의 정보로부터 상기 한 쌍의 약물들 간 부작용의 수준 및 유형을 결정하는 단계를 포함할 수 있다.

Description

약물들 간 상호작용을 분석하기 위한 방법 및 장치
본 발명은 약물들 간 상호작용(Drug-Drug Interaction, DDI) 분석에 관한 것으로, 특히, 인공지능 알고리즘을 이용하여 약물들 간 상호작용을 분석하기 위한 방법 및 장치에 대한 것이다.
문명이 발달함에 따라 인간의 삶은 풍요로워졌지만, 새로운 질병들이 계속적으로 발생하고 있다. 질병으로부터 자유롭고 건강하게 오래 살고자 하는 사람들의 소망에 따라, 새로운 약을 만들기 위한 수많은 시도와 노력이 이루어지고 있다. 이러한 가운데, 제약산업 분야에서 4차 산업 혁명은 대규모 데이터를 바탕으로 인공지능, 바이오 융합기술 등을 통해 약물 개발에 새로운 방향을 제시하고 있다.
약물들 간 상호작용(Drug-Drug Interaction, DDI)은 약물 개발에 있어 크게 고려해야 할 대상 중 하나이다. 약물들 간 상호작용이란 넓은 의미로 한 약물의 효능 혹은 독성이 다른 약물이나, 음식, 환경화학물질 등에 의해 변형될 때 일어나는 현상을 말한다. 즉, 약물들 간 상호작용은 임상에서는 2가지 이상의 약물을 동시에 혹은 짧은 간격을 두고 함께 사용할 때 서로 영향을 주는 것을 의미한다. 국내 복합제 임상시험 기준에 따르면, 안전성 측면에서 '병리기전 및 치료기전 등을 고려할 때, 개개 주성분의 병용투여 시 약동학적 및 약력학적 상호작용이 나타날 가능성이 있는 약물인 경우, 병용투여의 안전성 및 유효성에 대한 면밀한 평가가 요구된다'라는 기준이 명시되어 있다. 즉, 약물상호작용을 평가함에 있어서, 약물이 체내에 미치는 직접적인 영향과 약물 간에 나타나는 반응에 의한 영향 등이 모두 고려되어야 한다. 하지만, 현존하는 약물의 종류만 해도 수천 가지가 넘고, 약물들 간 상호작용에 관한 허가사항에 대해 국가별, 기관별로 제각각인 것인 것으로 드러나 약물 간의 조합이 적합한지 판단할 수 있는 자료들 조차도 불완전한 실정이다.
본 발명은 인공지능 알고리즘을 이용하여 약물들 간 상호작용(Drug-Drug Interaction, DDI)을 효과적으로 분석하기 위한 방법 및 장치를 제공하기 위한 것이다.
본 발명은 인공지능 알고리즘을 이용하여 약물들 간 상호작용 분석에 필요한 시간적, 인적, 물적 자원의 소요를 줄이기 위한 방법 및 장치를 제공하기 위한 것이다.
본 발명은 인공지능 알고리즘에 기반하여 복합제를 위한 적절한 약물 조합을 빠르게 예측하기 위한 방법 및 장치를 제공하기 위한 것이다.
본 발명은 인공지능 알고리즘에 기반하여 복합제를 위한 적절한 약물 조합을 빠르게 추천하기 위한 방법 및 장치를 제공하기 위한 것이다.
본 발명에서 이루고자 하는 기술적 과제들은 이상에서 언급한 기술적 과제들로 제한되지 않으며, 언급하지 않은 또 다른 기술적 과제들은 아래의 기재로부터 본 발명이 속하는 기술분야에서 통상의 지식을 가진 자에게 명확하게 이해될 수 있을 것이다.
본 발명의 일 실시 예에 따른, 약물들 간 상호작용(Drug-Drug Interaction, DDI)을 분석하는 방법은, 학습을 위한 데이터 셋(set)들로서, 약물들의 화학구조에 대한 제1 데이터 셋, 상기 약물들 간 부작용 수준(grade)에 대한 제2 데이터 셋, 상기 약물들 간 부작용 유형(type)에 대한 제3 데이터 셋을 획득하는 단계, 상기 제1 데이터 셋을 전처리함으로써, 상기 약물들 각각의 세부 속성 정보를 생성하는 단계, 상기 제2 데이터 셋을 전처리함으로써, 상기 제2 데이터 셋에 포함된 클래스를 정규화하고, 상기 제2 데이터 셋에 방향성을 부여하는 단계, 상기 제3 데이터 셋을 전처리함으로써, 상기 제3 데이터 셋에 포함된 부작용 유형을 나타내는 표현들을 추출하고, 상기 표현들을 정규화하고, 상기 제3 데이터 셋에 방향성을 부여하는 단계, 상기 전처리된 제1 데이터 셋, 상기 전처리된 제2 데이터 셋, 상기 전처리된 제3 데이터 셋을 이용하여 적어도 하나의 인공지능 모델을 학습하는 단계, 및 상기 적어도 하나의 인공지능 모델을 이용하여 한 쌍의 약물들의 정보로부터 상기 한 쌍의 약물들 간 부작용의 수준 및 유형을 결정하는 단계를 포함할 수 있다.
본 발명의 일 실시 예에 따라, 상기 적어도 하나의 인공지능 모델을 학습하는 단계는, 상기 전처리된 제1 데이터 셋을 상기 전처리된 제2 데이터 셋 및 상기 전처리된 제3 데이터 셋과 매칭함으로써, 약물의 속성 조합에 대하여 부작용 수준 및 유형을 맵핑한 학습 데이터 셋을 생성하는 단계를 포함할 수 있다.
본 발명의 일 실시 예에 따라, 상기 한 쌍의 약물들의 정보로부터 상기 한 쌍의 약물들 간 부작용의 수준 및 유형을 결정하는 단계는, 상기 한 쌍의 약물들의 정보를 전처리함으로써, 상기 한 쌍의 약물들 각각의 세부 속성 정보를 생성하는 단계, 및 상기 세부 속성 정보를 상기 적어도 하나의 인공지능 모델의 입력 데이터로서 입력하는 단계를 포함할 수 있다.
본 발명의 일 실시 예에 따라, 상기 세부 속성 정보는, 각 약물의 BDSI(Binary data of Drug Structural Information), ISD(Index of Similarity between Drugs), IIPD(Index of Interaction between Protein and Drug), IISD(Index of Interaction Similarity between Drugs), ADMET(Absortion Distribution Metabolism Excretion Toxicity)을 포함할 수 있다.
본 발명의 일 실시 예에 따라, 상기 제2 데이터 셋은, 제1 출처로부터 수집된 제1 약물들 간 부작용 수준 데이터 및 제2 출처로부터 수집된 제2 약물들 간 부작용 수준 데이터를 포함하며, 상기 제1 약물들 간 부작용 수준 데이터 및 상기 제2 약물들 간 부작용 수준 데이터는 동일한 클래스를 서로 다른 표현들로 지시하며, 상기 동일한 클래스를 지시하는 서로 다른 표현들은, 상기 전처리를 통해 정규화될 수 있다.
본 발명의 일 실시 예에 따라, 상기 제3 데이터 셋은, 제1 약물 쌍의 부작용의 유형을 표현하는 제1 문장 및 제2 약물 쌍의 부작용의 유형을 표현하는 제2 문장을 포함하며, 상기 제1 문장 및 상기 제2 문장 각각은, 적어도 하나의 유형을 나타내는 표현을 포함하며, 상기 제1 문장 및 상기 제2 문장은, 동일한 의미의 유형을 지시하는 서로 다른 표현들을 포함하고, 동일한 의미의 유형을 지시하는 서로 다른 표현들은, 상기 전처리를 통해 하나의 용어로 대체될 수 있다.
본 발명의 일 실시 예에 따라, 상기 제2 데이터 셋은, 제1 약물 및 제2 약물의 순서로 조합된 약물 쌍에 대한 부작용 수준 정보를 포함하는 항목을 포함하며, 상기 전처리된 제2 데이터 셋은, 상기 방향성의 부여에 의해, 상기 제2 약물 및 상기 제1 약물의 순서로 조합된 약물 쌍에 대한 부작용 수준 정보를 더 포함하도록 가공될 수 있다.
본 발명의 일 실시 예에 따라, 상기 제3 데이터 셋은, 제1 약물 및 제2 약물의 순서로 조합된 약물 쌍에 대한 부작용 유형 정보를 포함하는 항목을 포함하며, 상기 전처리된 제3 데이터 셋은, 상기 방향성의 부여에 의해, 상기 제2 약물 및 상기 제1 약물의 순서로 조합된 약물 쌍에 대한 부작용 유형 정보를 더 포함하도록 가공될 수 있다.
본 발명의 일 실시 예에 따라, 상기 적어도 하나의 인공지능 모델은, 상기 부작용 수준을 예측하는 다중 입력 단일 출력의 제1 인공지능 모델 및 상기 부작용 유형을 예측하는 다중 입력 다중 출력의 제2 인공지능 모델을 포함할 수 있다.
본 발명의 일 실시 예에 따라, 상기 방법은, 상기 한 쌍의 약물들 간 부작용의 수준 및 유형을 나타내는 데이터를 다른 장치로 송신하는 단계를 더 포함할 수 있다.
본 발명의 일 실시 예에 따른 약물들 간 상호작용(Drug-Drug Interaction, DDI)을 분석하는 장치는, 적어도 하나의 인공지능 모델을 저장한 메모리, 및 상기 메모리와 연결된 프로세서를 포함할 수 있다. 상기 프로세서는, 학습을 위한 데이터 셋(set)들로서, 약물들의 화학구조에 대한 제1 데이터 셋, 상기 약물들 간 부작용 수준(grade)에 대한 제2 데이터 셋, 상기 약물들 간 부작용 유형(type)에 대한 제3 데이터 셋을 획득하고, 상기 제1 데이터 셋을 전처리함으로써, 상기 약물들 각각의 세부 속성 정보를 생성하고, 상기 제2 데이터 셋을 전처리함으로써, 상기 제2 데이터 셋에 포함된 클래스를 정규화하고, 상기 제2 데이터 셋에 방향성을 부여하고, 상기 제3 데이터 셋을 전처리함으로써, 상기 제3 데이터 셋에 포함된 부작용 유형을 나타내는 표현들을 추출하고, 상기 표현들을 정규화하고, 상기 제3 데이터 셋에 방향성을 부여하고, 상기 전처리된 제1 데이터 셋, 상기 전처리된 제2 데이터 셋, 상기 전처리된 제3 데이터 셋을 이용하여 상기 적어도 하나의 인공지능 모델을 학습하고, 상기 적어도 하나의 인공지능 모델을 이용하여 한 쌍의 약물들의 정보로부터 상기 한 쌍의 약물들 간 부작용의 수준 및 유형을 결정하도록 제어할 수 있다.
본 발명에 대하여 위에서 간략하게 요약된 특징들은 후술하는 본 발명의 상세한 설명의 예시적인 양상일 뿐이며, 본 발명의 범위를 제한하는 것은 아니다.
본 발명에 따르면, 약물들 간 상호작용(Drug-Drug Interaction, DDI) 분석에 필요한 시간적, 인적, 물적 자원의 소요가 감소될 수 있다.
본 발명에서 얻을 수 있는 효과는 이상에서 언급한 효과들로 제한되지 않으며, 언급하지 않은 또 다른 효과들은 아래의 기재로부터 본 발명이 속하는 기술분야에서 통상의 지식을 가진 자에게 명확하게 이해될 수 있을 것이다.
도 1은 본 발명의 일 실시 예에 따른 약물들 간 상호작용(Drug-Drug Interaction, DDI) 분석 시스템의 구조를 도시한다.
도 2는 본 발명의 일 실시 예에 따른 시스템에 적용 가능한 인공 신경망의 구조를 도시한다.
도 3은 본 발명의 일 실시 예에 따른 시스템의 구조를 도시한다.
도 4는 본 발명의 일 실시 예에 따른 시스템에서 학습 데이터 및 독립 변인 획득의 개념을 도시한다.
도 5는 본 발명의 일 실시 예에 따른 시스템의 기능적 구조를 도시한다.
도 6은 본 발명의 일 실시 예에 따라 화학적 구조로부터 BDSI(Binary data of Drug Structural Information)를 도출하는 동작의 예를 도시한다.
도 7은 본 발명의 일 실시 예에 따라 화학적 구조로부터 ISD(Index of Similarity between Drug)를 도출하는 동작의 예를 도시한다.
도 8은 본 발명의 일 실시 예에 따라 속성 정보 및 수준/유형 정보를 매칭하는 동작의 예를 도시한다.
도 9는 본 발명의 일 실시 예에 따라 인공지능 모델을 학습 및 평가하는 동작의 예를 도시한다.
도 10a 및 도 10b는 본 발명의 일 실시 예에 따른 부작용 유형(type) 시스템을 위한 인공지능 모델의 예를 도시한다.
도 11a 및 도 11b는 본 발명의 일 실시 예에 따른 부작용 수준(grade) 시스템을 위한 인공지능 모델의 예를 도시한다.
도 12는 본 발명의 일 실시 예에 따른 부작용 수준 시스템을 위한 인공지능 모델의 예를 도시한다.
도 13는 본 발명의 일 실시 예에 따른 인공지능 모델의 예측-검증-학습의 순환적 동작을 도시한다.
도 14는 본 발명의 일 실시 예에 따른 부작용 유형 시스템을 위한 인공지능 모델의 예를 도시한다.
도 15는 본 발명의 일 실시 예에 따른 시스템에서 약물들 간 상호작용을 분석하기 위한 절차를 도시한다.
도 16은 본 발명의 일 실시 예에 따른 시스템에서 학습 및 예측을 위한 절차의 일 실시 예를 도시한다.
도 17은 본 발명의 일 실시 예에 따른 시스템에서 학습을 수행하는 절차의 일 실시 예를 도시한다.
도 18은 본 발명의 일 실시 예에 따른 시스템에서 예측을 수행하는 절차의 일 실시 예를 도시한다.
이하에서는 첨부한 도면을 참고로 하여 본 발명의 실시 예에 대하여 본 발명이 속하는 기술 분야에서 통상의 지식을 가진 자가 용이하게 실시할 수 있도록 상세히 설명한다. 그러나, 본 발명은 여러 가지 상이한 형태로 구현될 수 있으며 여기에서 설명하는 실시 예에 한정되지 않는다.
본 발명의 실시 예를 설명함에 있어서 공지 구성 또는 기능에 대한 구체적인 설명이 본 발명의 요지를 흐릴 수 있다고 판단되는 경우에는 그에 대한 상세한 설명은 생략한다. 그리고, 도면에서 본 발명에 대한 설명과 관계없는 부분은 생략하였으며, 유사한 부분에 대해서는 유사한 도면 부호를 붙였다.
본 발명은 지능적 인공지능 알고리즘을 이용하여 약물들 간 상호작용(Drug-Drug Interaction, DDI)을 분석하기 위한 기술에 대해 제안한다. 구체적으로, 본 발명은 클라우드 환경, 로컬 환경 등 다양한 환경 하에서 약물들 간 상호작용을 분석하는 시스템을 제공하기 위한 것이다.
본 발명의 다양한 실시 예들에 따른 시스템은 'CombiRisk'라 명명될 수 있다. CombiRisk는 4차 산업의 핵심 기술인 빅데이터 및 인공지능 기술을 바탕으로 약물 간의 상호작용을 분석하여 사용자에게 복합제를 위한 적절한 약물 조합을 빠르게 예측 및 추천해줄 수 있는 시스템이다. 국내 및 국외의 약물 빅데이터와 딥러닝 기술 기반인 CombiRisk 시스템은 약물의 주성분 간의 적합성을 예측하는 의사 결정 지원 시스템으로서, 복합제 연구를 위한 시간, 인적, 물적 자원들을 보다 효율적으로 배치할 수 있도록 돕기 위한 것이다.
도 1은 본 발명의 일 실시 예에 따른 약물들 간 상호작용 분석 시스템의 구조를 도시한다.
도 1을 참고하면, 시스템은 통신망에 연결된 사용자 장치(110a), 사용자 장치(110b), 서버(120)를 포함한다. 도 1은 2개의 사용자 장치들(110a, 110b)를 예시하였으나, 3개 이상의 사용자 장치들이 존재할 수 있다.
사용자 장치(110a) 및 사용자 장치(110b)는 본 발명의 실시 예에 따른 시스템을 이용하여 약물들 간 상호작용 분석을 수행하고자 하는 사용자에 의해 사용되는 종단 장치이다. 사용자 장치(110a) 및 사용자 장치(110b)는 입력 데이터(예: 상호작용 분석의 대상인 약물들의 정보)를 획득하고, 입력 데이터를 통신 망을 통해 서버(120)로 송신할 수 있다. 사용자 장치들(110a, 110b) 각각은 통신을 위한 통신부, 데이터 및 프로그램을 저장하는 저장부, 정보를 표시하기 위한 표시부, 사용자의 입력을 위한 입력부, 제어를 위한 프로세서를 포함할 수 있다. 예를 들어, 사용자 장치들(110a, 110b) 각각은 시스템 접속을 위한 어플리케이션 또는 프로그램을 설치한 범용 장치(예: 스마트폰, 타블렛, 랩탑 컴퓨터, 데스크탑 컴퓨터) 또는 시스 전용 접속 단말일 수 있다.
서버(120)는 본 발명의 실시 예들에 따른 약물들 간 상호작용을 분석하기 위한 연산을 수행한다. 서버(120)는 약물들 간 상호작용 분석 시스템을 위한 다양한 기능들을 제공하며, 인공지능 모델을 운용할 수 있다. 본 발명에 적용 가능한 인공 신경망의 일 예는 이하 도 2를 참고하여 설명된다. 또한, 서버(120)는 학습 데이터를 이용하여 인공지능 모델을 위한 학습을 수행할 수 있다. 여기서, 서버(120)는 로컬 네트워크에 존재하는 로컬 서버이거나, 외부 망을 통해 연결되는 원격 접속 서버(예: 클라우드 서버)일 수 있다. 서버(120)는 통신을 위한 통신부, 데이터 및 프로그램을 저장하는 저장부, 제어를 위한 프로세서를 포함할 수 있다.
도 2는 본 발명의 일 실시 예에 따른 시스템에 적용 가능한 인공 신경망의 구조를 도시한다. 도 2와 같은 인공 신경망은 서버(120)에 저장된 인공지능 모델들의 구조로 이해될 수 있다. 도 2를 참고하면, 인공 신경망은 입력 계층(input layer)(210), 적어도 하나의 은닉 계층(hidden layer)(220), 출력 계층(output layer)(230)으로 이루어진다. 계층들(210, 220, 230) 각각은 복수의 노드(node)들로 구성되어 있으며, 노드들 각각은 이전 계층에 속한 적어도 하나의 노드의 출력과 연결되어 있다. 각 노드는 이전 계층의 노드들의 각 출력 값과 그에 상응하는 연결 가중치(weight)를 내적(inner product)한 값에 바이어스(bias)를 더한 후, 비선형(non-linear)인 활성화 함수(activation function)와 곱한 출력 값을 다음 계층의 적어도 하나의 뉴런에게 전달한다. 각 계층은 입력 노드, 퍼셉트론, 출력 노드로 다시 구분될 수 있다.
도 2와 같은 인공 신경망은 학습(예: 기계 학습(machine learning), 딥 러닝(deep learning) 등)에 의해 형성될 수 있다. 또한, 본 발명의 다양한 실시 예에서 사용되는 인공 신경망 모델은 완전 합성곱 신경망(fully convolutional neural network), 합성곱 신경망(convolutional neural network), 순환 신경망(recurrent neural network), 제한 볼츠만 머신(restricted Boltzmann machine, RBM) 및 심층 신뢰 신경망(deep belief neural network, DBN) 중 적어도 하나를 포함할 수 있으나, 이에 한정되지 않는다. 또는, 딥러닝 이외의 머신 러닝 방법도 포함할 수 있다. 또는 딥러닝과 머신 러닝을 결합한 하이브리드 형태의 모델도 포함할 수 있다. 예컨대, 딥러닝 기반의 모델을 적용하여 영상의 특징을 추출하고, 상기 추출된 특징에 기초하여 영상을 분류하거나 인식할 때는 머신 러닝 기반의 모델을 적용할 수도 있다. 머신 러닝 기반의 모델은 서포트 벡터 머신(Support Vector Machine, SVM), 에이다부스트(AdaBoost) 등을 포함할 수 있으나, 이에 한정되지 않는다.
*본 발명의 실시 예에 따른 시스템의 딥러닝을 위해, 심층 신경망(Deep Neural Network, DNN)이 적용될 수 있고, 약물동태(Pharmacokinetic, PK) 상에서 나타날 수 있는 약물 간의 상호작용이 예측될 수 있다. 즉, 본 발명의 실시 예에 따른 시스템은 약물들 간 상호작용 정보에 딥러닝 기술을 적용함으로써, 새로운 약물들 간의 상호작용에 의한 부작용의 위험 수준 및 유형을 예측하고, 예측된 결과를 제공한다. 이를 위해, 국내외에서 제공되는 DDI 정보들을 수집한 후, 자체 데이터베이스가 구축될 수 있고, 이를 기반으로 각 약물들의 특징들이 추출될 수 있다. 예를 들어, 약물의 특징들은 약물의 구조 정보, 약물 간 구조적 유사도, 흡수/분포/대사/배출/독성 정보(Absortion/ Distribution/ Metabolism/ Excretion/ Toxicity, ADMET), 단백질과의 상호작용 정보 등을 포함할 수 있다. 여기서, ADMET는 생체 내에서 약물의 분포를 묘사하는 기준으로서 사용되는 정보이다.
본 발명의 실시 예에 따른 시스템은, 이하 도 3과 같이, 위험도를 5-단계로 예측하는 DDI 수준(grade) 시스템과 어떤 유형의 DDI가 발생할 것인지 예측하는 DDI 유형(Type) 시스템으로 구성될 수 있다.
도 3은 본 발명의 일 실시 예에 따른 시스템의 구조를 도시한다. 도 3을 참고하면, 서버(120)는 DDI 유형 시스템(310), DDI 수준 시스템(320)을 포함한다. DDI 유형 시스템(310)은 드러그뱅크(dragbank) 데이터베이스에 기반하여 구현될 수 있다. DDI 수준 시스템(320)은 드러그뱅크 데이터베이스, 드럭스컴(drugscom) 데이터베이스, 건강보험심사평가원 복합처방 데이터베이스, 약물 복합제 데이터베이스, 병용금기 데이터베이스 등에 기반하여 구현될 수 있다.
DDI 유형 시스템(310)은 발생이 예상되는 DDI의 유형을 제공한다. 예를 들어, DDI 유형 시스템(310)은 영향받는 약물(subject drug)인 약물a 및 영향주는 약물(affecting drug)인 약물b가 조합될 때 특정 증상(예: 횡문근융해(rhabdomyolysis))의 위험(risk) 또는 심각성(severity)이 증가할 있음을 예측(predict)할 수 있다. DDI 유형 시스템(310)은 'RiskDescription 시스템', 'RiskDescription 시스템 모델', 'CombiType 시스템', 'CombiType 시스템 모델', '부작용 유형 시스템' 등으로 지칭될 수 있다.
DDI 수준 시스템(320)은 위험도를 예측하기 위한 DNN을 포함할 수 있다. DDI 수준 시스템(320)은 'RiskGrade 시스템', 'RiskGrade 시스템 모델', 'CombiGrade 시스템', 'CombiGrade 시스템 모델', '부작용 수준 시스템' 등으로 지칭될 수 있다. 본 발명의 일 실시 예에 따라, 위험도는 5개 레벨들로 분류될 수 있다. 예를 들어, 5개 레벨들로 분류된 위험도는 이하 [표 1]과 같다.
5-단계 위험도
단계 1 사용금지(contraindicated)
단계 2 심각(major)
단계 3 보통(moderate)
단계 4 사소함(minor)
단계 5 사용가능(available)
도 4는 본 발명의 일 실시 예에 따른 시스템에서 학습 데이터 및 독립 변인 획득의 개념을 도시한다. 도 4를 참고하면, 학습 데이터 및 독립 변인들은 약물의 화학적 구조로부터 획득될 수 있다. 약물a(410-1) 및 약물b(410-2)의 화학적 구조 정보를 분석함으로써, 약물a(410-1) 및 약물b(410-2)의 속성에 대한 세부 속성 정보(420-1, 420-2)가 획득될 수 있다. 예를 들어, 세부 속성 정보(420-1, 420-2)는 BDSI, ISD, IIPD, IISD, ADMET 등을 포함할 수 있다. 이후, 특성에 대한 정보(420-1, 420-2)에 기반하여 특징 벡터(feature vector)의 연접(concatenation)(430)이 생성된다. 특징 벡터의 연접(430)은 인공지능 모델의 학습 또는 예측 동작을 위해 사용될 수 있다.
본 발명의 다양한 실시 예들에 약물들 간 상호작용 예측 시스템은 복수의 DDI 데이터베이스들(예: Drugbank, Drugscom, 공공데이터포탈, 건강보험심사평가원, 한국의약품안전관리원 등)을 통합적으로 사용한다. 따라서, 한 종류의 데이터베이스에 치우침 없이, 인공지능 모델이 다양한 데이터를 이용하여 학습됨에 따라, 과적합의 위험이 현저히 낮아진다. 또한, 본 발명의 다양한 실시 예들에 약물들 간 상호작용 예측 시스템은 약물 간 상호작용의 결과를 예측할 수 있을 뿐만 아니라, 동시에 약물들 간 상호작용의 심각도를 예측할 수 있다. 이에 따라, 복합제 개발 및 약물 처방 등에 필요한 자원이 보다 효율적으로 활용될 수 있다. 또한, 본 발명의 다양한 실시 예들에 약물들 간 상호작용 예측 시스템은 약물 간 상호작용의 반응뿐만 아니라, 2가지 약물의 상호작용에 방향성까지 고려하여 영향주는 약물(subject drug) 및 영향받는 약물(affected drug)을 구분한 예측을 제공할 수 있다.
도 5는 본 발명의 일 실시 예에 따른 시스템의 기능적 구조를 도시한다. 도 5는 도 1의 서버(120)의 기능적 구성으로 이해될 수 있다.
도 5를 참고하면, 서버는 데이터 수집부(510), 전처리부(520), 데이터 분류부(530), 학습부(540), 인공지능 모델(550), 입력데이터 획득부(560), 분석부(570)를 포함한다.
데이터 수집부(510)는 학습을 위한 원본 데이터를 수집한다. 원본 데이터는 다양한 형태 및 내용의 데이터를 포함할 수 있다. 예를 들어, 원본 데이터는 유료 구매 데이터 및 공공 데이터를 포함할 수 있다. 일 실시 예에 따라, 원본 데이터는 3가지의 데이터 셋(set)들을 포함할 수 있다. 구체적으로, 원본 데이터는 약물 화학구조 데이터 셋, 약물 간 부작용 수준 데이터 셋, 약물 간 부작용 유형 데이터 셋을 포함할 수 있다.
전처리부(520)는 학습을 위해 원본 데이터를 가공한다. 다시 말해, 전처리부(520)는 원본 데이터를 학습 가능한 형태로 가공한다. 다양한 실시 예들에 따라, 전처리부(520)는 약물 구조 데이터 셋으로부터 약물의 속성을 나타내는 세부 속성 정보를 생성할 수 있다. 예를 들어, 세부 속성 정보는 BDSI(Binary data of Drug Structural Information), ISD(Index of Similarity between Drugs), IIPD(Index of Interaction between Protein and Drug), IISD(Index of Interaction Similarity between Drugs), ADMET(Absortion Distribution Metabolism Excretion Toxicity)을 포함할 수 있다. 본 발명의 다양한 실시 예들에 따라, 세부 속성 정보는 위 나열된 BDSI, ISD, IIPD, IISD, ADMET 외 다른 항목을 더 포함하거나, 또는 나열된 항목들 중 적어도 하나가 다른 항목으로 대체될 수 있다. 또한, 전처리부(520)는 약물 간 부작용 수준 데이터 셋, 약물 간 부작용 유형 데이터 셋을 미리 정의된 기준에 따라 정규화하고, 방향성을 부여하고, 약물의 세부 속성 정보와 매칭한 후, 독립 변인 및 종속 변인을 생성할 수 있다.
데이터 분류부(530)는 전처리된 데이터를 인공지능 모델(550)의 학습 절차에서의 용도에 따라 분류한다. 예를 들어, 데이터 분류부(530)는 데이터를 학습 데이터, 검증 데이터, 테스트 데이터로 분류할 수 있다. 구체적으로, 전처리된 데이터 중 60%는 학습 데이터로, 20%는 검증 데이터로, 20%는 테스트 데이터로 분류될 수 있다.
학습부(540)는 데이터 분류부(530)로부터 제공되는 학습 데이터, 검증 데이터, 테스트 데이터를 이용하여 인공지능 모델(550)의 학습 및 평가를 수행한다. 예를 들어, 도 9와 같이, 인공지능 모델(550)이 학습 및 평가될 수 있다. 도 9를 참고하면, 학습 데이터(910) 및 검증 데이터(920)를 이용하여 인공지능 모델(550)이 학습된 후, 테스트 데이터(930)를 이용하여 성능 평가가 이루어진다. 이때, 성능이 요구되는 기준(예: 임계 비율 이상의 정확도)을 충족하지 못하면, 재학습이 이루어질 수 있다. 재학습을 고려하여, 학습부(540)는 학습 데이터 및 검증 데이터를 일부만을 사용하여 학습을 수행하고, 테스트 데이터의 일부를 이용하여 평가를 수행한 후, 평가 결과에 따라 다른 일부를 이용하여 학습을 추가적으로 수행할 수 있다.
인공지능 모델(550)은 심층 신경망을 포함한다. 심층 신경망은 도 2를 참고하여 설명한 바와 같이, 입력 계층, 출력 계층을 포함하며, 적어도 하나의 은닉 계층을 포함한다. 각 계층은 적어도 하나의 입력 노드, 적어도 하나의 퍼셉트론, 적어도 하나의 출력 노드로 구성된다. 본 발명의 일 실시 예에 따라, 인공지능 개발 라이브러리를 활용하여 파이썬(Python) 기반의 Keras 라이브러리, phytoch 라이브러리 등을 활용하여 신경층을 쌓음으로써, 심층 신경망이 빠르게 구축될 수 있다. 예를 들어, 파이썬(Python) 기반의 Keras 라이브러리, phytoch 라이브러리, tensorflow 라이브러리 등이 활용되거나, 다른 프로그래밍 언어(예: JAVA, C 등)가 활용될 수 있다.
CombiRisk 시스템에 포함되는 DDI 유형 시스템(310), DDI 수준 시스템(320) 각각은 약 6 내지 8개 계층을 포함하도록 설계될 수 있다. 이때, 각 층은 배치 정규화(Batch Normalization) 층, 밀집(Dense) 층, 드랍아웃(Dropout) 층을 포함할 수 있다. 배치 정규화 층을 통해 데이터가 보다 학습하기에 좋은 상태로 변환되고, 밀집 층에서 학습이 이루어지고, 드랍아웃 층에서 과대적합 확률이 감소될 수 있다. 각 층의 퍼셉트론(뉴런) 개수는 최소 15개에서 최대 2048개 사이로 설계될 수 있다. 학습 데이터 내 클래스 간 불균형을 해소하기 위한 여러 알고리즘들이 적용 및 테스트되는 것이 바람직하다. 예를 들어, 예측이 잘되는 클래스의 손실(loss) 값은 낮추고, 예측이 잘되지 않는 클래스의 손실 값은 소폭 낮추게하는 포컬 로스(focal loss) 알고리즘이 적용될 수 있다. 이 경우, 예측이 잘되지 않는 클래스에 대해 더 집중적으로 학습이 수행된다.
입력데이터 획득부(560)는 예측 동작을 위해 인공지능 모델(550)에 입력되는 입력 데이터를 획득한다. 예를 들어, 입력 데이터는 약물 정보를 포함한다. 여기서, 약물 정보는 약물의 속성에 대한 세부 정보(예: BDSI, ISD, IIPD, IISD, ADMET 등)를 포함할 수 있다. 또는, 약물 정보는 약물 ID를 포함할 수 있다. 입력 데이터가 약물 ID인 경우, 도 5에 도시되지 아니하였으나, 입력 데이터를 전처리함으로써 세부 속성 정보를 생성하는 전처리부가 더 포함될 수 있다. 또는, 학습 데이터를 전처리하는 전처리부(520)이 입력 데이터를 전처리할 수 있다.
분석부(570)는 입력 데이터에 포함된 세부 속성 정보 또는 입력 데이터로부터 생성된 세부 속성 정보를 인공지능 모델(550)에 입력하고, 인공지능 모델(550)의 출력 데이터를 획득하고, 획득된 출력 데이터에 기반하여 분석 결과를 생성한다. 생성된 분석 결과는 내부에 저장되거나, 외부(예: 사용자 장치(110a), 사용자 장치(110b))로 송신될 수 있다.
본 발명의 일 실시 예에 따라, 약물 화학구조 데이터 셋은 약물 ID(identifier) 및 SMILES(Simplified Molecular Input Line Entry System)를 포함할 수 있다. 여기서, SMILES는 화학물질의 구조를 나타내는 문자열이다. SMILES는 매우 간결한 구조의 문자열 표기 방식에 따르며, 화합물의 구조적 특징을 압축적으로 추상화한 표현이다. SMILES에 따르면, 원자는 표준 원소기호로 나타내고, 수소원자는 가능한 모든 곳에 연결되어 있다 가정하여 표기 생략되고, 이웃한 원자는 바로 인접해서 기재하고, 2중결합은 '=' 으로, 3중결합은 '#'으로, 결합 가지은 괄호 '()'로 표현되고, 고리 구조는 서로 연결된 원자에 숫자를 표시함으로써 표현된다. 예를 들어, 에탄올의 SMILES 표현은 CCO, 벤젠의 SMILES 표현은 C1=CC=CC=C1, 안트라센의 SMILES 표현은 C1=CC=C2C=C3C=CC=CC3=CC2=C1이다. Pubchem, Drugbank 등의 화학 물질 데이터베이스로부터, 각 화합물의 SMILES이 획득될 수 있다. 다만, 데이터베이스마다 SMILES 형태에 차이가 존재하므로, 전처리가 필요할 수 있다. 다만, 전술한 SMILES는 화학구조 데이터 셋의 일 예이며, 다른 화학구조 데이터가 다양한 실시 예들을 위해 사용될 수 있따. 예를 들어, SMILES를 대체하거나 병행하여, 화합물 데이터 자료(mol file, mol2 file, sdf(structural-data file)), InChI(International Chemical Identifier), 화학식(Chemical Formula), 3D 구조(structure) 정보 등이 사용될 수 있다.
본 발명의 일 실시 예에 따라, 약물 간 부작용 수준 데이터 셋은 제1 약물 ID, 제2 약물 ID, 부작용 정도를 나타내는 수준 값을 포함한다. 수준 값은 미리 정의된 후보 값들 중 하나로서, 각 후보 값은 [표 1]에 나열된 레벨들 중 하나를 지시한다. 약물 간 부작용 유형 데이터 셋은 영향을 주는 제1 약물 ID, 영향을 받는 제2 약물 ID, 부작용 유형을 나타내는 유형 정보를 포함한다. 유형 정보는 제1 약물이 제2 약물에게 어떠한 부작용을 일으키는지를 기술한 문장으로 표현될 수 있다. 예를 들어, 유형 정보는 "sub-dug may decrease the anticoagulant activities of aff_drug", "sub-dug may decrease the antihypertensive activities of aff_drug", "sub-dug can cause a decrease in the absorption of aff_drug", "sub-dug can cause an increase in the absorption of aff_drug" 등으로 표현될 수 있다.
본 발명의 일 실시 예에 따라, 약물 간 부작용 수준 데이터 셋 및 약물 간 부작용 유형 데이터 셋은 약물 구조 데이터 셋에 포함된 약물들의 조합에 관련되므로, 약물 구조 데이터 셋에 포함되는 항목들보다 많은 항목들을 포함할 수 있다. 예를 들어, 약물 구조 데이터 셋이 13,000 여개의 약물 항목들의 구조 정보를 포함하는 경우, 약물 간 부작용 수준 데이터 셋 및 약물 간 부작용 유형 데이터 셋 각각은 약 1,500,000 여개의 상호작용 관련 항목들을 포함할 수 있다.
도 5를 참고하여 설명한 실시 예에서, 전처리부(520)는 약물의 화학구조로부터 BDSI, ISD, IIPD, IISD, ADMET를 포함하는 세부 속성 정보를 생성한다. 각 속성에 대해 설명하면 다음과 같다.
BDSI는 화합물들의 고유 정보를 나타내며, 약물에서의 분자의 특징이나 유사도를 확인하기 위해 디자인되는 정보이다. 전처리부(520)는 각 원소를 기준으로 거리에 따라 어떤 구조인지, 어떤 원소와 결합되어 있는지에 따라 각 값들을 계산하며, 계산된 값들을 이진(binary) 값으로 표현한다. 즉, BDSI는 주변 원소들을 이용하여 평균적인 분자 구조와 다양한 분자들의 특징을 나타낸다. 화학적 구조로부터 BDSI가 도출되는 과정의 예는 이하 도 6과 같다.
도 6은 본 발명의 일 실시 예에 따라 화학적 구조로부터 BDSI를 도출하는 동작의 예를 도시한다. 도 6은 화합물(610)에 대한 BDSI 생성을 예시한다. 도 6을 참고하면, 화합물(610)의 각 원소에 대하여, 거리 0, 2, 4 각각의 범위에서의 구조들(620)이 확인된다. 확인된 구조들(620)은 수치로 표현되는 ID들(630)로 변환된다. ID들(630)은 리스트 표현(640)으로 변환되고, 리스트 표현(640)에 포함되는 ID들에 대응하는 이진 값들이 해시 함수(hash function)에 의해 재배치됨으로써, BDSI(650)이 생성된다.
DBSI를 표현하기 위한 바이너리(binary)의 개수는 조절 가능하며, 이진 데이터이므로 빠른 연산이 가능하다는 장점을 가진다. 생성된 BSDI 데이터 셋의 일 예는 이하 [표 2]와 같다.
drug ID BDSI
DB00006 {1, 0, 0, 0, 0 …, 0, 1, 0, 1, 0}
DB00007 {0, 0, 0, 0, 0 …, 0, 0, 0, 1, 0}
DB00014 {1, 0, 0, 1, 0 …, 0, 0, 0, 0, 0}
DB00027 {0, 0, 0, 0, 0 …, 0, 1, 0, 0, 0}
DB00035 {0, 0, 1, 0, 0 …, 0, 0, 0, 1, 0}
DB00050 {1, 0, 0, 0, 1 …, 0, 1, 0, 0, 0}
... ...
ISD는 BDSI에 기반하여 생성되며, 화합물의 구조적 유사도를 표현한다. 즉, 전처리부(520)는 BDSI에 의해 표현되는 분자 구조에 기반하여 약물들 간의 화합물 구조의 유사도를 계산한다. 약물 당 하나의 ISD 값이 생성된다. 예를 들어, 10,000개의 약물들이 존재하는 경우, 이하 도 7과 같이, 1개의 약물 당 10,000개의 유사도 값들을 가지게 되며, 유사도 값들을 순서대로 나열한 10,000개의 값들이 하나의 ISD를 구성한다. 도 7은 본 발명의 일 실시 예에 따라 화학적 구조로부터 ISD를 도출하는 동작의 예를 도시한다. 도 7을 참고하면, DB00007의 ID를 가지는 약물(710)에 대하여, 자신을 포함한 모든 약물들과의 유사도 값들(720)이 계산된다. 유사도 값들(730)을 미리 정의된 순서(예: 약물 ID 오름차순)로 나열함으로써, ISD 값(730)이 생성된다. 이러한 과정을 통해 생성된 ISD 데이터 셋의 일 예는 이하 [표 3]과 같다.
drug ID ISD
DB00006 {1, 0.34234, 0.64534 … 0.756454}
DB00007 {0.34234, 1, 0.342425 … 0.123546}
DB00014 {0.64534, 0.342425, 1 … 0.856523}
DB00027 {0.133345, 0.623244, 0.136542 … 0.643534}
DB00035 {0.845634, 0.234562, 0.734211 … 0.892344}
DB00050 {0.522423, 0.642324, 0.718964 … 0.716342}
... ...
IIPD는 약물 및 단백질 간 반응을 표현하는 정보이다. 전처리부(520)는 BDSI에 의해 표현된 분자 구조에 기반하여 약물과 단백질 사이에서 일어나는 일련의 반응들을 수치화한다. 이를 위해, 약물과 단백질 간의 상호작용에 주요 역할을 하는 것으로 알려진 단백질들이 선정된다. 전처리부(520)는 약물과 단백질 구조에 따라 총 8가지 유형의 반응들에 대한 정보를 추출할 수 있다. 예를 들어, 8가지 유형의 반응들은, 소수성 접촉(hydrophobic contacts), 전면 대 전면 방향족(aromatic face to face), 경계 대 전면 방향족(aromatic edge to face), 단백질이 수소 결합 공여체로서의 수소 결합(hydrogen bond (protein as hydrogen bond donor), 단백질이 수소 결합 수용체로서의 수소 결합(hydrogen bond (protein as hydrogen bond acceptor)), 단백질이 양전하로 하전된 염다리(salt bridges (protein positively charged)), 단백질이 음전하로 하전된 염다리(salt bridges (protein negatively charged)), 금속 이온으로 이온 결합된 염다리(salt bridges (ionic bond with metal ion)) 등을 포함할 수 있다.IISD는 약물들 간 IIPD의 유사도를 나타내는 정보이다. 전처리부(520)는 IIPD에 기반하여 약물들 간 IIPD의 유사도를 계산한다. 예를 들어, 10,000개의 약물들이 존재하는 경우, 1개의 약물 당 10,000개의 유사도 값들을 가지며, 10,000개의 유사도 값들을 순서대로 나열한 결과가 하나의 약물의 IISD를 구성한다. 전처리부(520)는 약물 별로 IISD를 결정하고, 복수의 IISD들을 포함하는 IISD 데이터 셋을 생성한다.
ADMET은 약물 동태학 관점에서 시간에 따라 체내 약물 농도의 변화를 통해 약물의 흡수, 분배, 대사, 배출, 독성 등의 수준을 수치화한 정보이다. 전처리부(520)는 약물 구조 정보, 즉, SMILE로부터 약물의 분자적 특징(molecular feature)을 추출한 후, ADMET를 계산한다. 1개의 약물 당 6가지 범주들 각각을 의미하는 총 28개의 ADMET 값이 생성된다. 예를 들어, 28개의 ADMET 값들은 ①기초 물리 화학적 특성에 관련된 LogS, LodD, LogP, ②흡수에 관련된 Caco-2, Pgp-Inhibitor, HIA, F(20%), F(30%), ③분배에 관련된 PPB, VD, BBB, ④대사에 관련된 CYP1A2-Inhibitor, CYP 1A2-Substrate, CYP 3A4-Inhibitor, CYP 3A4-Substrate, CYP 2C9-Inhibitor, CYP 2C9-Substrate, CYP 2C19-Inhibitor, CYP 2C19-Substrate, CYP 2D6-Inhibitor, CYP 2D6-Substrate, ⑤배출에 관련된 Clearance, T1/2, ⑥독성에 관련된 hERG, H-HT, Ames, Skin sensitivity, LD50를 포함한다. 전처리부(520)는 약물 별로 ADMET 값 집합을 결정하고, 복수의 ADMET 값 집합들을 포함하는 IISD 데이터 셋을 생성한다.
전처리부(520)는 약물 간 부작용 수준 데이터 셋으로부터 DDI 수준 시스템(320)을 위한 학습 데이터를 생성한다. '0'(Major), '1'(Moderate), '2'(Minor)로 클래스가 구분된 원본 데이터베이스도 있지만, 다른 표현으로 클래스가 구분되는 원본 데이터 베이스 또는 클래스 구분이 따로 없는 원본 데이터베이스도 존재한다. 따라서, 전처리부(520)는 클래스 재분류를 위해 데이터베이스 특성을 분석하고, 분석 결과에 기반하여 전처리를 수행할 수 있다. 또한, 전처리부(520)는 데이터에 방향성을 부여할 수 있다. 예를 들어, "DB06605+DB00001→Grade 1"의 항목이 저장되어 있으면, 전처리부(520)는 "DB00001+DB06605→Grade 1"의 항목을 추가한다. 왜냐하면, "DB06605+DB00001→Grade 1"의 데이터만 학습할 경우, "DB00001+DB06605"의 조합을 입력하는 경우 Grade 1이 아닌 다른 결과가 예측될 수 있기 때문이다.
전처리부(520)는 약물 간 부작용 유형 데이터 셋으로부터 DDI 유형 시스템(310)을 위한 학습 데이터를 생성한다. 전처리부(520)는 약물 간 부작용 유형 데이터 셋에 포함되는 유형 정보로부터 부작용 유형을 추출한다. 예를 들어, 유형 정보가 "sub_drug may decrease effectiveness of aff_drug"인 경우, 전처리부(520)는 'decrease', 'effectiveness'를 추출할 수 있다. 다른 예로, 유형 정보가 "sub_drug may increase the QTc-prolonging activities of aff_drug"인 경우, 전처리부(520)는 'increase', 'QTc-prolonging'를 추출할 수 있다. 그리고, 전처리부(520)는 부작용 유형을 분석하고, 전처리를 수행한다. 예를 들어, 전처리부(520)는 동의어, 유사 부작용 등을 통일된 용어로 정리할 수 있다. 또한, 전처리부(520)는 데이터에 방향성을 부여할 수 있다. 예를 들어, "DB06605+DB00001→increase, QTc-prolonging"의 항목이 존재하면, 전처리부(520)는, DB06605는 영향받는 약물(sub_drug), DB0001는 영향주는 약물(aff_drug)이라는 의미로, '0'이라는 방향성 값을 부여할 수 있다. 더불어, 전처리부(520)는 "DB00001+DB06605→increase, QTc-prolonging"의 항목을 추가하고, DB0001은 영향주는 약물(aff_drug), DB06605는 영향받는 약물(sub_drug)이라는 의미로 '1'이라는 방향성 값을 부여할 수 있다. 즉, 방향성 값 '0'은 영향받는 약물이 전자, 영향주는 약물이 후자인 조합을 의미하고, 방향성 값 '1'은 영향주는 약물이 전자, 영향받는 약물이 후자인 조합을 의미한다.
전술한 바와 같이, 전처리부(520)는 원본 데이터에 대한 전처리를 수행한다. 이후, 데이터는 독립 변인 및 종속 변인으로 분리된다. 독립 변인 및 종속 변인을 결정하기 위해, 전처리부(520)는 약물의 화학구조로부터 BDSI, ISD, IIPD, IISD, ADMET 등의 세부 속성 정보 및 수준/유형 데이터를 매칭할 수 있다. 매칭 동작은 도 8과 같다.
도 8은 본 발명의 일 실시 예에 따라 속성 정보 및 수준/유형 정보를 매칭하는 동작의 예를 도시한다. 도 8을 참고하면, 먼저 약물들이 화학구조 정보(예: SMILES)로부터 생성된 약물 ID 별 BDSI, ISD, IIPD, IISD, ADMET를 포함하는 약물 ID 별 세부 속성 정보(810)가 생성된다. 이어, 약물 ID 쌍(pair) 별 DDI 수준 데이터(820a) 및 약물 ID 쌍 별 DDI 유형 데이터(820b)가 약물 ID 별 세부 속성 정보(810)와 매칭된다. 매칭에 의해, BDSI 쌍 별 DDI 수준/유형 데이터(830a), ISD 쌍 별 DDI 수준/유형 데이터(830b), IIPD 쌍 별 DDI 수준/유형 데이터(830c), IISD 쌍 별 DDI 수준/유형 데이터(830d), ADMET 쌍 별 DDI 수준/유형 데이터(830e)가 생성된다. 예를 들어, BDSI 쌍 별 DDI 수준/유형 데이터(830a)는 이하 [표 4] 및 [표 5]와 같다.
subject drug ID affected drug ID label
{1, 0, 0, 0, 0 …, 0, 1, 0, 1, 0} {1, 0, 0, 1, 0 …, 0, 0, 0, 0, 0} 0
{1, 0, 0, 0, 0 …, 0, 1, 0, 1, 0} {0, 0, 1, 0, 0 …, 0, 0, 0, 1, 0} 0
{0, 0, 1, 0, 0 …, 0, 0, 0, 1, 0} {1, 0, 0, 1, 0 …, 0, 0, 0, 0, 0} 1
{0, 0, 1, 0, 0 …, 0, 0, 0, 1, 0} {1, 0, 0, 0, 1,…, 0, 1, 0, 0, 0} 1
{1, 0, 0, 0, 0 …, 0, 1, 0, 1, 0} {1, 0, 0, 0, 1,…, 0, 1, 0, 0, 0} 2
... ... ...
subject drug ID affected drug ID modified summary
{1,0,0,0,0 …,0,1,0,1,0} {1,0,0,1,0 …,0,0,0,0,0} The therapeutic efficacy ...
{1,0,0,0,0 …,0,1,0,1,0} {0,0,1,0,0 …,0,0,0,1,0} The therapeutic efficacy ...
{0,0,1,0,0 …,0,0,0,1,0} {1,0,0,1,0 …,0,0,0,0,0} subject drug can cause ...
{0,0,1,0,0 …,0,0,0,1,0} {1,0,0,0,1,…,0,1,0,0,0} subject drug can cause ...
{1,0,0,0,0 …,0,1,0,1,0} {1,0,0,0,1,…,0,1,0,0,0} subject drug may decrease ...
... ... ...
ISD 쌍 별 DDI 수준/유형 데이터(830b), IIPD 쌍 별 DDI 수준/유형 데이터(830c), IISD 쌍 별 DDI 수준/유형 데이터(830d), ADMET 쌍 별 DDI 수준/유형 데이터(830e)도 [표 4] 및 [표 5]와 유사한 형태로 구성될 수 있다. 즉, 수준/유형 데이터에서 약물 ID 열을 해당 약물의 ISD, IIPD, IISD, ADMET으로 대체함으로써, ISD 쌍 별 DDI 수준/유형 데이터(830b), IIPD 쌍 별 DDI 수준/유형 데이터(830c), IISD 쌍 별 DDI 수준/유형 데이터(830d), ADMET 쌍 별 DDI 수준/유형 데이터(830e)가 생성될 수 있다.전처리부(520)는 매칭을 통해 생성된 데이터 셋들로부터 독립 변인 및 종속 변인을 결정한다. 예를 들어, 약물 쌍(pair)에 대한 BDSI, ISD, IIPD, IISD, ADMET 등의 독립 변인들이 이하 [표 6] 내지 [표 10]과 같이 생성될 수 있다.
subject drug ID affected drug ID
DB06605_BDSI DB00001_BDSI
DB06695_BDSI DB00001_BDSI
DB01254_BDSI DB00001_BDSI
DB00001_BDSI DB01609_BDSI
DB00001_BDSI DB01586_BDSI
DB00001_BDSI DB02659_BDSI
... ...
[표 6]에서, 'DB*****_DBSI'는 약물 ID가 DB*****인 약물의 BDSI 값을 의미한다.
subject drug ID affected drug ID
DB06605_ISD DB00001_ISD
DB06695_ISD DB00001_ISD
DB01254_ISD DB00001_ISD
DB00001_ISD DB01609_ISD
DB00001_ISD DB01586_ISD
DB00001_ISD DB02659_ISD
... ...
[표 7]에서, 'DB*****_ISD'는 약물 ID가 DB*****인 약물의 ISD 값을 의미한다.
subject drug ID affected drug ID
DB06605_IIPD DB00001_IIPD
DB06695_IIPD DB00001_IIPD
DB01254_IIPD DB00001_IIPD
DB00001_IIPD DB01609_IIPD
DB00001_IIPD DB01586_IIPD
DB00001_IIPD DB02659_IIPD
... ...
[표 8]에서, 'DB*****_IIPD'는 약물 ID가 DB*****인 약물의 IIPD 값을 의미한다.
subject drug ID affected drug ID
DB06605_IISD DB00001_IISD
DB06695_IISD DB00001_IISD
DB01254_IISD DB00001_IISD
DB00001_IISD DB01609_IISD
DB00001_IISD DB01586_IISD
DB00001_IISD DB02659_IISD
... ...
[표 9]에서, 'DB*****_IISD'는 약물 ID가 DB*****인 약물의 IISD 값을 의미한다.
subject drug ID affected drug ID
DB06605_ADMET DB00001_ADMET
DB06695_ADMET DB00001_ADMET
DB01254_ADMET DB00001_ADMET
DB00001_ADMET DB01609_ADMET
DB00001_ADMET DB01586_ADMET
DB00001_ADMET DB02659_ADMET
... ...
[표 10]에서, 'DB*****_ADMET'는 약물 ID가 DB*****인 약물의 ADMET 값을 의미한다.그리고, 수준 및 유형에 대한 항목들이 종속 변인으로서 추출된다. 예를 들어, 이하 [표 11]과 같은 단일 출력의 수준 클래스, 이하 [표 12]와 같은 다중 출력의 유형 클래스를 포함하는 종속 변인들이 생성될 수 있다.
label
0
0
1
1
2
2
...
QTc-Prolonging hepatotoxic liver damage infection hypothyroid hypomania hyperthemia hyperkalemic hypertension ...
0 0 1 0 0 0 1 0 0 ...
0 0 1 0 0 1 0 0 0 ...
0 0 0 0 0 0 0 0 1 ...
0 0 0 0 1 0 0 0 0 ...
1 0 0 0 0 0 1 0 0 ...
1 0 0 0 0 0 0 0 1 ...
... ... ... ... ... ... ... ... ... ...
도 10a 및 도 10b는 본 발명의 일 실시 예에 따른 부작용 유형 시스템을 위한 인공지능 모델의 예를 도시한다. 도 10a 및 도 10b는 DDI 유형 시스템(310)을 위한 인공지능 모델을 예시한다. 도 10a 및 도 10b는 참고하면, 전처리 과정(1010)에 의해, 약물 SMILES 데이터(1002)로부터 BDSI, ISD, IIPD, IISD, ADMET를 포함하는 세부 속성 정보(1004)가 결정된다. 세부 속성 정보(1004)가 인공지능 모델(1020)에 학습 데이터로서 제공된다. 인공지능 모델(1020)은 복수의 계층들을 포함하며, 각 계층은 배치 정규화(Batch Normalization, BN) 층, 밀집(Dense) 층, 드랍아웃(Dropout) 층을 포함한다. 인공지능 모델(1020)을 이용한 예측에 의해 부작용 유형(1008)이 결정된다. 부작용 유형(1008)은 다중 출력의 형태를 가진다. 약물 간 부작용 유형 데이터(1006)는 전처리 과정(1030)을 거친 후, 인공지능 모델(1020)의 출력 계층으로 제공되고, 역전파(back-propagation) 연산에 의해 인공지능 모델(1020)이 학습된다. 여기서, 출력 계층의 구조는 종속 변인의 형태에 의존할 수 있다.
도 11a 및 도 11b는 본 발명의 일 실시 예에 따른 부작용 수준 시스템을 위한 인공지능 모델의 예를 도시한다. 도 11a 및 도 11b는 DDI 수준 시스템(320)을 위한 인공지능 모델을 예시한다. 도 11을 참고하면, 전처리 과정(1110)에 의해, 약물 SMILES 데이터(1102)로부터 BDSI, ISD, IIPD, IISD, ADMET를 포함하는 세부 속성 정보(1104)가 결정된다. 세부 속성 정보(1104)가 인공지능 모델(1120)에 학습 데이터로서 제공된다. 인공지능 모델(1120)은 복수의 계층들을 포함하며, 각 계층은 배치 정규화(Batch Normalization, BN) 층, 밀집(Dense) 층, 드랍아웃(Dropout) 층을 포함한다. 인공지능 모델(1120)을 이용한 예측에 의해 부작용 수준(1108)이 결정된다. 부작용 수준(1108)은 단일 출력의 형태를 가진다. 약물 간 부작용 수준 데이터(1106)는 전처리 과정(1130)을 거친 후, 인공지능 모델(1120)의 출력 계층으로 제공되고, 역전파(back-propagation) 연산에 의해 인공지능 모델(1120)이 학습된다. 여기서, 출력 계층의 구조는 종속 변인의 형태에 의존할 수 있다.
도 12는 본 발명의 일 실시 예에 따른 부작용 수준 시스템을 위한 인공지능 모델의 예를 도시한다. 도 12를 참고하면, 부작용 수준 시스템을 위한 인공지능 모델(1250)은 다중 입력, 단일 출력의 형태를 가진다. 이에 따라, BDSI, ISD, IIPD, IISD, ADMET를 포함하는 입력 데이터(1202)가 입력되면, 하나의 부작용 수준을 나타내는 출력 데이터(1204)가 출력된다. 인공지능 모델(1250)은 약물 쌍의 BDSI, ISD, IIPD, IISD, ADMET 등 독립 변인들의 패턴을 분석함으로서, 수준 클래스(grade class)를 예측한다. 추가적으로, 도 13과 같이, 인공지능 모델(1250)은 예측(1301) 결과를 스스로 검증(1303)하고, 검증 내용을 피드백하면서 학습(1305)을 더 진행할 수 있다. 예를 들어, 예측 결과가 『Drug_1, Drug_2 → Class '1'』이면, 해석은 『약물 1번과 약물 2번은 함께 사용 시 '보통 수준'의 부작용 발생 확률 있음』이다.
도 14는 본 발명의 일 실시 예에 따른 부작용 유형 시스템을 위한 인공지능 모델의 예를 도시한다. 도 14를 참고하면, 부작용 유형 시스템을 위한 인공지능 모델(1450)은 다중 입력, 다중 출력의 형태를 가진다. 이에 따라, BDSI, ISD, IIPD, IISD, ADMET를 포함하는 입력 데이터(1402)가 입력되면, 복수의 항목들로 표현되는 부작용 유형을 나타내는 출력 데이터(1404)가 출력된다. 인공지능 모델(1450)은 약물 쌍의 BDSI, ISD, IIPD, IISD, ADMET 등 독립변인들의 패턴을 분석하고, 유형 클래스들을 예측한다. 예를 들어, 예측 결과가 『Drug_1(subject_drug), Drug_2(affected_drug) → Type : 'increase', Type : 'CNS depression', Type : 'hypotesion'』이면, 해석은 『약물 1번은 약물 2번의 영향으로 함께 사용 시 '중추신경계 저하'와 '저혈압' 부작용이 '증가'할 확률이 있음』이다.
도 15는 본 발명의 일 실시 예에 따른 시스템에서 약물들 간 상호작용을 분석하기 위한 절차를 도시한다. 도 15를 참고하면, 부작용의 수준(Grade)과 유형(Type)을 예측하고자 하는 1쌍의 약물들(예: 약물1(1502-1), 약물2(1502-2))이 수준을 판단하는 RiskGrade 시스템 모델(1550a) 및 유형을 판단하는 RiskDescription 시스템 모델(1550b)로 입력된다. 입력되는 데이터는 해당 약물의 BDSI, ISD, IIPD, IIPDSP, ADMET 등 모델이 학습에 사용되었던 속성 데이터를 포함한다. RiskGrade 시스템 모델(1550a) 및 유형을 판단하는 RiskDescription 시스템 모델(1550b) 각각은 예측 결과들(1504-1, 1504-2)을 출력한다. 예측 결과들(1504-1, 1504-2)에 기반하여, 약물 간 부작용의 수준 및 유형을 예측한 결론(1506)이 얻어진다.
RiskDescription 시스템 모델(1550b)는 방향성도 예측하므로, 어느 약물의 영향으로 다른 약물의 부작용이 어떻게 변화하는지를 제공한다. 예를 들어, RiskGrade 시스템 모델(1550a)은 『Drug_1, Drug_2 → Class '1'』의 예측 결과를, RiskDescription 시스템 모델(1550b)은 『Drug_1(subject_drug), Drug_2(affected_drug) → Type : 'increase', Type : 'CNS depression', Type : 'hypotesion』의 예측 결과를 출력할 수 있다. 이 경우, 결론(1506)은 『약물 1번과 약물 2번은 함께 사용 시 '보통 수준'의 부작용 발생확률이 있으며, 약물 2번의 영향으로 약물 1번의 '중추신경계 저하'와 '저혈압' 부작용이 '증가'할 수 있다』를 포함할 수 있다.
도 16은 본 발명의 일 실시 예에 따른 시스템에서 학습 및 예측을 위한 절차의 일 실시 예를 도시한다. 이하 설명에서, 동작 주체는 '장치'로 설명되나, 후술되는 동작들은 서버 또는 사용자 장치에 의해 수행될 수 있다.
도 16을 참고하면, S1601 단계에서, 장치는 데이터를 획득한다. 데이터는 학습을 위한 데이터로서, 예를 들어, 약물 화학구조 데이터 셋, 약물 간 부작용 수준 데이터 셋, 약물 간 부작용 유형 데이터 셋을 포함할 수 있다.
S1603 단계에서, 장치는 학습을 수행한다. 학습을 수행하기 위해, 장치는 데이터에 대한 전처리를 수행하고, 전처리된 데이터를 이용하여 학습을 수행할 수 있다.
S1605 단계에서, 장치는 예측을 수행한다. 즉, 장치는 학습된 인공지능 모델을 이용하여 분석하고자 하는 약물 쌍에 대한 정보를 포함하는 입력 데이터로부터 예측 결과를 포함하는 출력 데이터를 획득한다. 이때, 장치는 인공지능 모델에 입력 가능한 형식으로 약물 쌍에 대한 정보를 전처리할 수 있다.
도 17은 본 발명의 일 실시 예에 따른 시스템에서 학습을 수행하는 절차의 일 실시 예를 도시한다. 이하 설명에서, 동작 주체는 '장치'로 설명되나, 후술되는 동작들은 서버 또는 사용자 장치에 의해 수행될 수 있다.
도 17을 참고하면, S1701 단계에서, 장치는 화학구조에 기반하여 세부 속성 정보를 결정한다. 즉, 약물 화학구조 데이터에 대한 전처리로서, 장치는 약물의 화학구조에 기반하여 세부 속성 정보, 예를 들어, BDSI, ISD, IIPD, IISD, ADMET 중 적어도 하나의 속성 정보를 생성한다.
S1703 단계에서, 장치는 등급 데이터에 대해 클래스를 재분류하고, 방향성을 부여한다. 즉, 약물 간 부작용 유형 데이터에 대한 전처리로서, 장치는 미리 정의된 형식으로 수준 클래스 값들을 재분류한다. 즉, 장치는 미리 정의된 형식에 따라 서로 다른 출처(source)들로부터 수집된 약물 조합에 따른 수준 정보를 통일된 형식으로 정규화한다. 그리고, 장치는 데이터 셋에 약물 조합의 순서를 변경한 항목을 추가함으로써, 방향성을 부여한다.
S1705 단계에서, 장치는 유형 데이터로부터 부작용 유형을 추출하고, 표현을 정리하고, 방향성을 부여한다. 즉, 약물 간 부작용 수준 데이터에 대한 전처리로서, 장치는 문장 형식의 데이터로부터 유형을 표현하는 키워드들을 추출한다. 그리고, 장치는 동의어, 유사어 관계에 있는 표현들을 대표 표현으로 대체한다. 또한, 장치는 데이터 셋에 약물 조합의 순서를 변경한 항목을 추가함으로써, 방향성을 부여한다.
S1707 단계에서, 장치는 전처리된 데이터에 기반하여 학습을 수행한다. 이를 위해, 장치는 전처리된 데이터 셋들을 매칭함으로써 약물의 속성 조합 및 부작용 유형/수준을 맵핑한 데이터 셋을 생성한다. 그리고, 장치는 생성된 데이터 셋을 이용하여 학습을 수행한다. 즉, 장치는 약물의 속성 조합들을 독립 변인으로, 부작용 유형/수준을 종속 변인으로 가지는 학습 데이터를 이용하여 학습을 수행한다. 다시 말해, 장치는 부작용 유형/수준으로 라벨링(labeling)된 약물의 속성 조합 정보를 학습 데이터로서 사용하여 인공지능 모델에 대한 학습을 수행한다. 즉, 장치는 학습 데이터를 이용하여 예측을 수행한 후, 역전파 동작을 통해 인공지능 모델의 가중치를 갱신한다.
도 18은 본 발명의 일 실시 예에 따른 시스템에서 예측을 수행하는 절차의 일 실시 예를 도시한다. 이하 설명에서, 동작 주체는 '장치'로 설명되나, 후술되는 동작들은 서버 또는 사용자 장치에 의해 수행될 수 있다.
도 18을 참고하면, S1801 단계에서, 장치는 입력 데이터를 획득한다. 입력 데이터는 장치에 구비된 입력 수단(예: 키보드 등의 입력 장치, 외부 저장 매체와 연결 가능한 포트, 통신 망을 통한 신호를 수신하는 인터페이스 등)을 통해 입력될 수 있다.
S1803 단계에서, 장치는 입력 데이터에 포함된 약물의 화학구조를 확인한다. 이때, 입력 데이터가 화학구조 정보(예: SMILES)를 포함하지 아니하고, 약물의 식별 정보(예: 명칭, ID 등)을 포함하면, 장치는 식별 정보에 대응하는 화학구조를 내부 또는 외부의 데이터베이스로부터 검색할 수 있다. 이를 위해, 장치는 데이터베이스에 접속하고, 약물의 식별 정보를 포함하는 요청을 송신한 후, 응답으로서 화학구조 정보를 수신할 수 있다.
S1805 단계에서, 장치는 화학구조에 기반하여 약물의 세부 속성 정보를 결정한다. 장치는 약물의 화학구조에 기반하여 세부 속성 정보, 예를 들어, BDSI, ISD, IIPD, IISD, ADMET 중 적어도 하나의 속성 정보를 생성한다.
S1807 단계에서, 장치는 예측을 통해 출력 데이터를 획득한다. 장치는 인공지능 모델을 이용하여 한 쌍의 약물들의 세부 속성 정보로부터 예측 결과를 포함하는 출력 데이터를 생성한다. 여기서, 출력 데이터는 예측 결과로서 부작용 수준 데이터 및 부작용 유형 데이터 중 적어도 하나를 포함할 수 있다. 다른 실시 예에 따라, 출력 데이터는 사용자가 보다 용이하게 이해할 수 있는 문자화된 결과로 변환될 수 있다.
S1809 단계에서, 장치는 출력 데이터를 제공한다. 여기서, 출력 데이터는 분석을 요청 또는 명령한 사용자에게 제공된다. 예를 들어, 출력 데이터는 장치에 구비된 출력 수단(예: 스크린 등)을 통해 시각적으로 출력되거나, 또는 외부 장치로 송신될 수 있다.
본 발명의 예시적인 방법들은 설명의 명확성을 위해서 동작의 시리즈로 표현되어 있지만, 이는 단계가 수행되는 순서를 제한하기 위한 것은 아니며, 필요한 경우에는 각각의 단계가 동시에 또는 상이한 순서로 수행될 수도 있다. 본 발명에 따른 방법을 구현하기 위해서, 예시하는 단계에 추가적으로 다른 단계를 포함하거나, 일부의 단계를 제외하고 나머지 단계를 포함하거나, 또는 일부의 단계를 제외하고 추가적인 다른 단계를 포함할 수도 있다.
본 발명의 다양한 실시 예는 모든 가능한 조합을 나열한 것이 아니고 본 발명의 대표적인 양상을 설명하기 위한 것이며, 다양한 실시 예에서 설명하는 사항들은 독립적으로 적용되거나 또는 둘 이상의 조합으로 적용될 수도 있다.
또한, 본 발명의 다양한 실시 예는 하드웨어, 펌웨어(firmware), 소프트웨어, 또는 그들의 결합 등에 의해 구현될 수 있다. 하드웨어에 의한 구현의 경우, 하나 또는 그 이상의 ASICs(Application Specific Integrated Circuits), DSPs(Digital Signal Processors), DSPDs(Digital Signal Processing Devices), PLDs(Programmable Logic Devices), FPGAs(Field Programmable Gate Arrays), 범용 프로세서(general processor), 컨트롤러, 마이크로 컨트롤러, 마이크로 프로세서 등에 의해 구현될 수 있다.
본 발명의 범위는 다양한 실시 예의 방법에 따른 동작이 장치 또는 컴퓨터 상에서 실행되도록 하는 소프트웨어 또는 머신-실행가능한 명령들(예를 들어, 운영체제, 애플리케이션, 펌웨어(firmware), 프로그램 등), 및 이러한 소프트웨어 또는 명령 등이 저장되어 장치 또는 컴퓨터 상에서 실행 가능한 비-일시적 컴퓨터-판독가능 매체(non-transitory computer-readable medium)를 포함한다

Claims (11)

  1. 약물들 간 상호작용(Drug-Drug Interaction, DDI)을 분석하는 방법에 있어서,
    상기 약물들 간 상호작용을 분석하기 위해 구성된 프로세서에서, 학습을 위한 데이터 셋(set)들로서, 약물들의 화학구조에 대한 제1 데이터 셋, 상기 약물들 간 부작용 수준(grade)에 대한 제2 데이터셋, 상기 약물들 간 부작용 유형(type)에 대한 제3 데이터 셋을 획득하는 단계;
    상기 프로세서에서, 상기 제1 데이터 셋을 전처리함으로써, 상기 약물들 각각의 세부 속성 정보를 생성하는 단계;
    상기 프로세서에서, 상기 제2 데이터 셋을 전처리함으로써, 상기 제2 데이터 셋에 포함된 클래스를 정규화하고, 상기 제2 데이터셋에 방향성을 부여하는 단계;
    상기 프로세서에서, 상기 제3 데이터 셋을 전처리함으로써, 상기 제3 데이터 셋에 포함된 부작용 유형을 나타내는 표현들을 추출하고, 상기 표현들을 정규화하고, 상기 제3 데이터셋에 방향성을 부여하는 단계;
    상기 프로세서에서, 상기 전처리된 제1 데이터 셋, 상기 전처리된 제2 데이터 셋, 상기 전처리된 제3 데이터 셋을 이용하여 메모리에 저장된 적어도 하나의 인공지능 모델을 학습하는 단계; 및
    상기 프로세서에서, 상기 적어도 하나의 인공지능 모델을 이용하여 한 쌍의 약물들의 정보로부터 상기 한 쌍의 약물들 간 부작용의 수준 및 유형을 결정하는 단계를 포함하는 방법.
  2. 청구항 1에 있어서,
    상기 적어도 하나의 인공지능 모델을 학습하는 단계는,
    상기 프로세서에서, 상기 전처리된 제1 데이터 셋을 상기 전처리된 제2 데이터 셋 및 상기 전처리된 제3 데이터 셋과 매칭함으로써, 약물의 속성 조합에 대하여 부작용 수준 및 유형을 맵핑한 학습 데이터 셋을 생성하는 단계를 포함하는 방법.
  3. 청구항 1에 있어서,
    상기 한 쌍의 약물들의 정보로부터 상기 한 쌍의 약물들 간 부작용의 수준 및 유형을 결정하는 단계는,
    상기 프로세서에서, 상기 한 쌍의 약물들의 정보를 전처리함으로써, 상기 한 쌍의 약물들 각각의 세부 속성 정보를 생성하는 단계; 및
    상기 프로세서에서, 상기 세부 속성 정보를 상기 적어도 하나의 인공지능 모델의 입력 데이터로서 입력하는 단계를 포함하는 방법.
  4. 청구항 1에 있어서,
    상기 세부 속성 정보는, 각 약물의 BDSI(Binary data of Drug Structural Information), ISD(Index of Similarity between Drugs), IIPD(Index of Interaction between Protein and Drug), IISD(Index of Interaction Similarity between Drugs), ADMET(Absortion Distribution Metabolism Excretion Toxicity)을 포함하는 방법.
  5. 청구항 1에 있어서,
    상기 제2 데이터 셋은, 제1 출처로부터 수집된 제1 약물들 간 부작용 수준 데이터 및 제2 출처로부터 수집된 제2 약물들 간 부작용 수준 데이터를 포함하며,
    상기 제1 약물들 간 부작용 수준 데이터 및 상기 제2 약물들 간 부작용 수준 데이터는 동일한 클래스를 서로 다른 표현들로 지시하며,
    상기 동일한 클래스를 지시하는 서로 다른 표현들은, 상기 전처리를 통해 정규화되는 방법.
  6. 청구항 1에 있어서,
    상기 제3 데이터 셋은, 제1 약물 쌍의 부작용의 유형을 표현하는 제1 문장 및 제2 약물 쌍의 부작용의 유형을 표현하는 제2 문장을 포함하며,
    상기 제1 문장 및 상기 제2 문장 각각은, 적어도 하나의 유형을 나타내는 표현을 포함하며,
    상기 제1 문장 및 상기 제2 문장은, 동일한 의미의 유형을 지시하는 서로 다른 표현들을 포함하고,
    동일한 의미의 유형을 지시하는 서로 다른 표현들은, 상기 전처리를 통해 하나의 용어로 대체되는 방법.
  7. 청구항 1에 있어서,
    상기 제2 데이터 셋은, 제1 약물 및 제2 약물의 순서로 조합된 약물 쌍에 대한 부작용 수준 정보를 포함하는 항목을 포함하며,
    상기 전처리된 제2 데이터 셋은, 상기 방향성의 부여에 의해, 상기 제2 약물 및 상기 제1 약물의 순서로 조합된 약물 쌍에 대한 부작용 수준 정보를 더 포함하도록 가공되는 방법.
  8. 청구항 1에 있어서,
    상기 제3 데이터 셋은, 제1 약물 및 제2 약물의 순서로 조합된 약물 쌍에 대한 부작용 유형 정보를 포함하는 항목을 포함하며,
    상기 전처리된 제3 데이터 셋은, 상기 방향성의 부여에 의해, 상기 제2 약물 및 상기 제1 약물의 순서로 조합된 약물 쌍에 대한 부작용 유형 정보를 더 포함하도록 가공되는 방법.
  9. 청구항 1에 있어서,
    상기 적어도 하나의 인공지능 모델은, 상기 부작용 수준을 예측하는 다중 입력 단일 출력의 제1 인공지능 모델 및 상기 부작용 유형을 예측하는 다중 입력 다중 출력의 제2 인공지능 모델을 포함하는 방법.
  10. 청구항 1에 있어서,
    상기 한 쌍의 약물들 간 부작용의 수준 및 유형을 나타내는 데이터를 다른 장치로 송신하는 단계를 더 포함하는 방법.
  11. 약물들 간 상호작용(Drug-Drug Interaction, DDI)을 분석하는 장치에 있어서,
    적어도 하나의 인공지능 모델을 저장한 메모리; 및
    상기 메모리와 연결된 프로세서를 포함하고,
    학습을 위한 데이터 셋(set)들로서, 약물들의 화학구조에 대한 제1 데이터 셋, 상기 약물들 간 부작용 수준(grade)에 대한 제2 데이터셋, 상기 약물들 간 부작용 유형(type)에 대한 제3 데이터 셋을 획득하고,
    상기 제1 데이터 셋을 전처리함으로써, 상기 약물들 각각의 세부 속성 정보를 생성하고,
    상기 제2 데이터 셋을 전처리함으로써, 상기 제2 데이터 셋에 포함된 클래스를 정규화하고, 상기 제2 데이터셋에 방향성을 부여하고,
    상기 제3 데이터 셋을 전처리함으로써, 상기 제3 데이터 셋에 포함된 부작용 유형을 나타내는 표현들을 추출하고, 상기 표현들을 정규화하고, 상기 제3 데이터셋에 방향성을 부여하고,
    상기 전처리된 제1 데이터 셋, 상기 전처리된 제2 데이터 셋, 상기 전처리된 제3 데이터 셋을 이용하여 상기 적어도 하나의 인공지능 모델을 학습하고, 및
    상기 적어도 하나의 인공지능 모델을 이용하여 한 쌍의 약물들의 정보로부터 상기 한 쌍의 약물들 간 부작용의 수준 및 유형을 결정하도록 제어하는 장치.
PCT/KR2022/095058 2021-06-16 2022-03-17 약물들 간 상호작용을 분석하기 위한 방법 및 장치 WO2022265480A1 (ko)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
KR1020210078199A KR102339561B1 (ko) 2021-06-16 2021-06-16 약물들 간 상호작용을 분석하기 위한 방법 및 장치
KR10-2021-0078199 2021-06-16

Publications (1)

Publication Number Publication Date
WO2022265480A1 true WO2022265480A1 (ko) 2022-12-22

Family

ID=79033231

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/KR2022/095058 WO2022265480A1 (ko) 2021-06-16 2022-03-17 약물들 간 상호작용을 분석하기 위한 방법 및 장치

Country Status (2)

Country Link
KR (1) KR102339561B1 (ko)
WO (1) WO2022265480A1 (ko)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117976244B (zh) * 2024-04-01 2024-06-07 天津理工大学 基于多维特征的药物相互作用预测方法及装置

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102339561B1 (ko) * 2021-06-16 2021-12-16 닥터노아바이오텍 주식회사 약물들 간 상호작용을 분석하기 위한 방법 및 장치
KR102464996B1 (ko) * 2022-05-31 2022-11-09 주식회사 원스글로벌 의약품 표준화 데이터 구축 방법 및 이러한 방법을 수행하는 장치

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020095261A1 (en) * 2000-11-03 2002-07-18 J. Gut Method and system for registration, identifying and processing of drug specific data
US20150324693A1 (en) * 2014-05-06 2015-11-12 International Business Machines Corporation Predicting drug-drug interactions based on clinical side effects
US20170177803A1 (en) * 2015-12-21 2017-06-22 International Business Machines Corporation Predicting drug-drug interactions and specific adverse events
KR20200023689A (ko) * 2018-08-20 2020-03-06 아주대학교산학협력단 인공지능 기반의 약물유해반응 탐지 방법 및 그 시스템
KR20200072585A (ko) * 2018-11-30 2020-06-23 이율희 인공지능에 기반한 대상 물질의 유해성과 위해성 예측 방법
KR102339561B1 (ko) * 2021-06-16 2021-12-16 닥터노아바이오텍 주식회사 약물들 간 상호작용을 분석하기 위한 방법 및 장치

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102299674B1 (ko) 2019-10-10 2021-09-07 인천대학교 산학협력단 두 약물 간의 혼용 효과를 예측하기 위한 인공지능 기반의 예측모델을 생성할 수 있는 전자 장치 및 그 동작 방법

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020095261A1 (en) * 2000-11-03 2002-07-18 J. Gut Method and system for registration, identifying and processing of drug specific data
US20150324693A1 (en) * 2014-05-06 2015-11-12 International Business Machines Corporation Predicting drug-drug interactions based on clinical side effects
US20170177803A1 (en) * 2015-12-21 2017-06-22 International Business Machines Corporation Predicting drug-drug interactions and specific adverse events
KR20200023689A (ko) * 2018-08-20 2020-03-06 아주대학교산학협력단 인공지능 기반의 약물유해반응 탐지 방법 및 그 시스템
KR20200072585A (ko) * 2018-11-30 2020-06-23 이율희 인공지능에 기반한 대상 물질의 유해성과 위해성 예측 방법
KR102339561B1 (ko) * 2021-06-16 2021-12-16 닥터노아바이오텍 주식회사 약물들 간 상호작용을 분석하기 위한 방법 및 장치

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117976244B (zh) * 2024-04-01 2024-06-07 天津理工大学 基于多维特征的药物相互作用预测方法及装置

Also Published As

Publication number Publication date
KR102339561B1 (ko) 2021-12-16

Similar Documents

Publication Publication Date Title
WO2022265480A1 (ko) 약물들 간 상호작용을 분석하기 위한 방법 및 장치
WO2020213843A1 (ko) 사용자 맞춤형 의료정보 제공 시스템 및 이의 구동방법
WO2021177730A1 (ko) 음성 및 연하 장애를 유발하는 질환 진단 장치 및 그 진단 방법
WO2019168253A1 (ko) 계층적으로 사용자 표현을 이해하고 답변을 생성하는 대화형 상담 챗봇 장치 및 방법
CN110797124A (zh) 一种模型多端协同训练方法、医疗风险预测方法和装置
WO2021010744A1 (ko) 음성 인식 기반의 세일즈 대화 분석 방법 및 장치
WO2019093675A1 (ko) 빅데이터 분석을 위한 데이터 병합 장치 및 방법
WO2020032467A1 (ko) 인공지능 기반 질의응답 도출방법, 장치 및 프로그램
Zhao et al. Do transformers parse while predicting the masked word?
WO2021045332A1 (ko) 암호화폐 거래를 분석하기 위한 데이터 획득 방법 및 장치
WO2021251558A1 (ko) 임상시험 검색을 위한 데이터 분류 장치, 시스템 및 방법
WO2022035018A1 (ko) 클리닉 매니저 서비스 플랫폼 및 이를 제공하는 시스템
WO2022145877A1 (ko) 주기적으로 업데이트 되는 유전자 변이 검사 결과 리포트 자동 발행 시스템
WO2021071000A1 (ko) 신약 후보 물질 도출 방법 및 장치
WO2016190495A1 (ko) 비정형 데이터 기반 룰 관리 방법 및 그 장치
WO2018212396A1 (ko) 데이터를 분석하는 방법, 장치 및 컴퓨터 프로그램
Celebi et al. Evaluation of knowledge graph embedding approaches for drug-drug interaction prediction using linked open data
CN116402066A (zh) 多网络特征融合的属性级文本情感联合抽取方法及系统
WO2024090712A1 (ko) 공감대 형성을 통한 심리 치료용 인공지능 대화 시스템
WO2023113452A1 (ko) 인공 지능 기반 의료 텍스트의 노이즈 데이터 필터링 방법, 장치 및 프로그램
EP4258107A1 (en) Method and system for automated discovery of artificial intelligence and machine learning assets in an enterprise
CN113536784A (zh) 文本处理方法、装置、计算机设备和存储介质
WO2022181907A1 (ko) 대변 이미지 분석 기반 영양 정보 제공 방법, 장치 및 시스템
WO2019117400A1 (ko) 유전자 네트워크 구축 장치 및 방법
CN114547335A (zh) 业务数据处理方法、装置、设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22825403

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 18567683

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE