KR20220088983A - HUMAN PPARγ ANTAGONIST PREDICTION METHOD BASED ON LEARNING MODEL AND ANALYSIS APPARATUS - Google Patents

Abstract

The learning model-based method for predicting an antagonist of human PPAR gamma comprises the steps of: an analysis device receiving an identifier of a candidate compound; the second classification of the candidate compound by inputting, by the analysis device, an in silico molecular descriptor of the candidate compound classified as negative in the primary classification into a supervised learning classification model; The analysis device inputs the compound fingerprint of the candidate compound classified as negative in the secondary classification, the molecular descriptor, and the protein-ligand interaction fingerprint for human PPAR gamma ligand into the artificial neural network model 3 It includes the step of classifying the car.

Description

Learning model-based human PPAR gamma antagonist prediction method and analysis device {HUMAN PPARγ ANTAGONIST PREDICTION METHOD BASED ON LEARNING MODEL AND ANALYSIS APPARATUS}

The technique described below relates to a technique for predicting antagonists of human PPAR gamma using a learning model.

PPAR (peroxisome proliferator-activated receptor) is a type of hormone receptor present in the nucleus and is known as a transcription factor that regulates genes related to lipid and glucose metabolism and energy homeostasis. PPAR gamma is known to be involved in various diseases. Typically, PPAR gamma is known to play an important role in the initiation of pulmonary fibrosis expression.

QSAR (quantitative structure-activity relationships) is a methodology to evaluate the potential hazard of a chemical using only a computer. In order to develop a QSAR that can comprehensively respond to all novel organic compounds, the diversity of learning data on which predictions are based should be broadened. Therefore, comprehensive predictive QSAR must utilize large-scale, high-throughput screening (HTS) data such as the Tox21 (toxicology in the 21st century) project.

US Patent Publication No. US 2011-0071142

The large amount of chemicals tested with HTS has enormous structural diversity. However, the conventional QSAR methodology has limitations in predicting the entire diversity. First, it is difficult to predict the properties of organic compounds with an infinite variety of possible structures, and second, the algorithm is inevitably biased in the criteria for evaluating which of the properties of a chemical are more important for prediction.

The technology to be described below is intended to provide a technique for predicting human PPAR gamma antagonists only with chemical molecular structure information on in silico.

The learning model-based method for predicting an antagonist of human PPAR gamma comprises the steps of: an analysis device receiving an identifier of a candidate compound; the second classification of the candidate compound by inputting, by the analysis device, an in silico molecular descriptor of the candidate compound classified as negative in the primary classification into a supervised learning classification model; The analysis device inputs the compound fingerprint of the candidate compound classified as negative in the secondary classification, the molecular descriptor, and the protein-ligand interaction fingerprint for human PPAR gamma ligand into the artificial neural network model 3 It includes the step of classifying the car.

The analysis device for predicting an antagonist of human PPAR gamma is an input device that receives a compound fingerprint of a candidate compound, an in silico molecular descriptor and a protein-ligand interaction fingerprint for human PPAR gamma ligand, human PPAR gamma ligand A storage device for storing an unsupervised learning clustering model, a supervised learning classification model, and an artificial neural network model for predicting whether or not an antagonist to When the primary classification result is negative, the molecular descriptor is input to the supervised learning classification model to secondary classify the candidate compound, and when the secondary classification result is negative, the compound fingerprint; and a computing device for tertiary classification by inputting the molecular descriptor and the protein-ligand interaction fingerprint into the artificial neural network model.

The technology to be described below accurately predicts the antagonist of human PPAR gamma by resolving overfitting of analytical properties while considering only properties suitable for the purpose of analysis among the properties of chemical substances.

1 is an example of an antagonist prediction system of human PPAR gamma.
Figure 2 is an example of the human PPAR gamma antagonist prediction model training process.
3 is an example of an antagonist prediction process of human PPAR gamma.
4 is an experimental result for the prediction of antagonists of human PPAR gamma.
5 is another experimental result for the prediction of antagonists of human PPAR gamma.
6 is an example of the configuration of the analysis device.

The technology to be described below can apply various changes and can have various embodiments, and specific embodiments are illustrated in the drawings and described in detail. However, this is not intended to limit the technology described below to specific embodiments, and it should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the technology described below.

Terms such as first, second, A, and B may be used to describe various components, but the components are not limited by the above terms, and only for the purpose of distinguishing one component from other components. is used only as For example, a first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component without departing from the scope of the present invention. and/or includes a combination of a plurality of related listed items or any of a plurality of related listed items.

In terms of terms used herein, the singular expression should be understood to include the plural expression unless the context clearly dictates otherwise, and terms such as "comprises" include the described feature, number, step, operation, element. , parts or combinations thereof are to be understood, but not to exclude the possibility of the presence or addition of one or more other features or numbers, step operation components, parts or combinations thereof.

Prior to a detailed description of the drawings, it is intended to clarify that the classification of the constituent parts in the present specification is merely a division according to the main function each constituent unit is responsible for. That is, two or more components to be described below may be combined into one component, or one component may be divided into two or more for each more subdivided function. In addition, each of the constituent units to be described below may additionally perform some or all of the functions of other constituent units in addition to the main function it is responsible for. Of course, it can also be performed by being dedicated to it.

In addition, in performing the method or method of operation, each process constituting the method may occur differently from the specified order unless a specific order is clearly described in context. That is, each process may occur in the same order as specified, may be performed substantially simultaneously, or may be performed in the reverse order.

1 is an example of an antagonist prediction system 100 of human PPAR gamma.

The compound analyzer 110 analyzes the sample to generate data on the structure and properties of the compound for the sample. The compound analyzer 110 may be a device in which researchers analyze specific substances. The compound database (DB) 120 stores information about the generated compound. The compound DB 120 may be an open DB for research purposes. The compound DB 120 may store raw data indicating the structure and characteristics of the compound. Alternatively, the compound DB 120 may be data for analysis in which the structure and characteristics of the compound are calculated as information by analyzing the raw data. The data for analysis may include a protein-ligand interaction fingerprint for a compound fingerprint, a molecular descriptor, and a human PPAR gamma ligand, which will be described later.

The analyzers 130 and 140 predict the antagonist of human PPAR gamma using the input compound data. The analysis device may be implemented in various forms such as a computer device, a PC, a smart device, a server on a network, and the like. In FIG. 1 , the analysis device is shown in the form of a server 130 and a computer terminal 140 .

The analysis devices 130 and 140 may analyze the raw data stored in the compound DB 120 to generate data for analysis (input data) to be input to the learning model.

The analyzers 130 and 140 calculate a prediction result for whether a specific candidate compound is a human PPAR gamma antagonist candidate. The analyzers 130 and 140 analyze whether a specific candidate compound is a human PPAR gamma antagonist candidate using an in silico learning model.

The users 10 and 20 may check an analysis result for a specific candidate compound. The user 10 may access the server 130 through a user terminal (PC, smart phone, etc.) and check the analysis result performed by the server 130 . The user 20 may check the analysis result through the computer terminal he/she uses.

Hereinafter, a model for predicting an antagonist of PPAR gamma ligand will be described. Hereinafter, “researcher” refers to the person who developed the model. The researcher performed an in silico docking simulation at the ligand binding site of the target receptor based on the chemical structure using an analysis device.

The researcher extracted the chemical properties from the simulation results. This process limits the infinite structural possibilities of organic compounds to the frame of binding sites through virtual experiments, and reduces the number of discrimination rules to be used by the predictive model to a finite number.

The training data used to build the model will be described. The researcher used the published data utilized in the previous study. The researcher used the CompTox Chemistry Dashboard (The CompTox Chemistry Dashboard: a community data resource for environmental chemistry, J Cheminform. 2017 Nov 28;9(1):61). From the initial data, only structures applicable to the Molecular Operating Environment (MOE) were selected and used. Only compounds capable of expressing compound secondary structure (SMILES) were targeted.

The training data for building the model prepared in this way is 5,218 compounds. Of the training data, 543 are positive and 4,675 are negative. 2,454 model descriptors were used.

Model descriptors use compound fingerprints, protein-ligand-interaction fingerprints and molecular descriptors.

In silico, compounds can be expressed using MOE. The researcher identified the compound structure and properties based on the MOE.

The compound fingerprint uses a Morgan fingerprint. The Morgan fingerprint is calculated as a result of analyzing the input compound using RDKit. The researcher applied Morgan's algorithm to the input compound, yielding 2,048 fingerprints. The molecular structural connectivity presented by the Morgan fingerprint indicates the correspondence of the correlation to the PPAR gamma ligand binding site.

The protein-ligand-interaction fingerprint (PLIF) is a docking simulation result for the PPAR gamma ligand binding site. Docking simulation is performed for the binding site of the PPAR gamma ligand. The PPAR gamma ligand binding site uses the 5LSG structure constructed from the RCSB Protein Data Bank. Docking simulation was performed using ASEDock software. Of course, other docking simulation software may also be used. The protein-ligand-interaction fingerprint indicates the target amino acid binding to the chemical structure at the site where the PPAR gamma ligand binds. The protein-ligand-interaction fingerprint contains information about contacts in the bonding structure, ions, aromatics, hydrogen bonds (acceptors), side chain donor atoms and the backbone of 27 amino acids. The investigators used all 70 protein-ligand-interaction fingerprints.

Molecular descriptor is composed of factors representing molecular structure and properties in the MOE. Molecular descriptors include physical property descriptors, Huckel theory related factors, partitioned surface area factors, number of atoms, number of bonds, Kier-Hall molecular coupling factors, kappa shape indice, proximity, distance matrix. , partial charge, potential energy, pharmacophore model factors and MOPAC related factors. The researchers used a total of 336 molecular descriptors. Molecular descriptors may indicate the orientation of the molecular conformation that binds to the PPAR gamma ligand binding site.

Various learning models are available for the evaluation of antagonists of human PPAR gamma. The researchers used multiple QASR models based on different algorithms. When predicting the potential of an antagonist, the researchers applied it step-by-step from a model with a strong overfitting tendency, and sequentially selected chemicals with a high potential for an antagonist. This process complements the weaknesses of each algorithm and brings synergy to the strengths.

The researcher used three learning models sequentially. The first model is an unsupervised learning-based clustering model, the second model is a supervised learning-based classification model, and the third model is an artificial neural network model. Each model has models with various methods or structures. Hereinafter, the model finally utilized by the researcher will be described as an example.

An unsupervised learning-based clustering model can use DBSCAN (Density-based spatial clustering of applications with noise). DBSCAN is a density-based algorithm. DBSCAN is an algorithm for classifying noise data by forming clusters with an undetermined number and shape based on density for given data. DBSCAN classifies which cluster the input data is included in. DBSCAN takes two parameters. One is the minimum number of data constituting the cluster, and the other is the distance for identifying surrounding data. The researcher performed clustering based on the cosine similarity.

DBSCAN uses Morgan fingerprint as input data. DBSCAN is affected by the order of input data. The researcher performed the process of processing by assigning weights to the input data. Of course, classification is possible even if this process is omitted.

The researcher built a separate artificial neural network model for processing the Morgan print. The artificial neural network model for processing the input data of DBSCAN is called the preprocessing model. The preprocessing model can be built using the same configuration and method as the third model described above. However, the preprocessing model must be trained only with a single type of data to be input for the candidate compound (corresponding to the Morgan fingerprint in this example). The artificial neural network, which is a preprocessing model, is composed of 6-layers in Keras. When the Morgan fingerprint of the candidate compound is input to the trained artificial neural network, each node of the layer has an activation value. DBSCAN can perform clustering by using the active value of each node as a descriptor.

A classification model based on supervised learning may use a support vector machine (SVM). SVM classifies input data by supervised learning. SVM classifies input data by maximizing the margin for each cluster based on the hyper-rplane.

The artificial neural network model can use various models capable of processing vector data. The researcher used an artificial neural network with 6 layers. The output value of an artificial neural network usually has a value between 0 and 1. The researcher classified the output values into 5 results based on 4 thresholds. The four thresholds were 0.5, 0.75, 0.875 and 0.984375. Of course, in the learning process, criteria or thresholds for benign classification may be corrected.

The computer device may perform a process of learning the learning model. A computer device is a device for processing compound data and training a learning model. A computer device is a device capable of processing data, such as a PC or a server. The computer device may be the aforementioned analysis device or a separate device.

2 is an example of a human PPAR gamma antagonist prediction model training process 200 . The computer device receives data for model training from the compound DB. The compound DB stores structural and property data for a number of compounds. The computer device can define the structure and properties of the compound in the MOE.

The computer device may select only structures applicable to the MOE from data received from the compound DB ( 210 ). That is, the computer device may select only data that can be structured in silico as training data.

The computer device may generate a Morgan fingerprint, an MOE molecular descriptor, and a protein-ligand-interaction fingerprint (PLIF) for the candidate compound.

The computer device may use the three learning models for the evaluation of antagonists of human PPAR gamma for candidate compounds. 2 is an example of providing three learning models individually.

The computer device performs classification by inputting the Morgan fingerprint of the compound into the unsupervised learning-based clustering model, which is the first model ( 220 ). DBSCAN is unsupervised learning, but there is a need to optimize parameters for clustering. Accordingly, the computer device may optimize DBSCAN while repeatedly using a plurality of training data.

On the other hand, the computer device can be used after constant processing by giving weight to the Morgan fingerprint for the compound. As described above, the computer device may input the Morgan fingerprint to an artificial neural network, which is a pre-trained preprocessing model, extract an activation value from each node of the radar, and use the activation value as a presenter to input to DBSCAN.

The computer device inputs the MOE molecular descriptor for the compound into the second model, SVM, to perform classification (230). The computer device receives the prediction result for the current compound as feedback and trains the SVM by comparing it with known label values.

The computer device inputs the Morgan fingerprint, the MOE molecular descriptor, and the PLIF into a third model, an artificial neural network model (ANN), to perform prediction ( 240 ). The computer device receives the prediction result for the current compound as feedback and trains the artificial neural network model by comparing it with known label values.

The computer device adjusts parameters of the model by repeating the training process with respect to the first model, the second model, and the third model so that each model sufficiently predicts the correct answer.

The analyzer predicts whether the candidate compound is an antagonist of PPAR gamma or a candidate antagonist by using the learning model learned through the above-described process. 3 is an example of an antagonist prediction process 300 of human PPAR gamma.

The analysis device receives data for model training from the compound DB. The compound DB stores information about the target compound to be evaluated for antagonist candidate recognition. The compound DB stores structural and property data for a number of compounds. The analyzer can define the structure and properties of the compound in the MOE.

The analyzer may select only structures applicable to the MOE from data received from the compound DB (310). That is, the analysis device may select only data that can be structured in silico as the analysis target data.

The assay device may generate a Morgan fingerprint, an MOE molecular descriptor, and a protein-ligand-interaction fingerprint (PLIF) for a candidate compound.

The analyzer uses three learning models for the evaluation of antagonists of human PPAR gamma for candidate compounds.

The analysis apparatus performs classification by inputting the Morgan fingerprint of the compound to the first model, an unsupervised learning-based clustering model (DBSCAN) ( 320 ).

On the other hand, the analysis device can be used after constant processing by giving weight to the Morgan fingerprint for the compound. As described above, the analysis apparatus may input the Morgan fingerprint to the artificial neural network learned in advance, extract an activation value from each node of the radar, and use the activation value as a presenter to input to DBSCAN. The analysis device may determine the active value while inputting the Morgan fingerprint to the artificial neural network several times. The analysis device may determine a value indicating an activity independently or commonly during an iterative process as final input data while repeatedly determining a value indicating an activity value.

When DBSCAN classifies the current candidate compound as positive, the analyzer selects the candidate compound as a human PPAR gamma antagonist candidate.

When the DBSCAN predicts the current candidate compound as negative, the analysis device ignores the prediction result of the DBSCAN and inputs the data of the current candidate compound to the training of the second model to perform a new prediction. The analyzer performs classification by inputting the MOE molecular descriptor for the current candidate compound into the SVM, which is the second model ( 330 ).

When the SVM classifies the current candidate compound as positive, the analyzer selects the candidate compound as a human PPAR gamma antagonist candidate.

When the SVM predicts the current candidate compound as negative, the analysis device ignores the prediction result of the SVM, and inputs the data of the current candidate compound to the training of the third model to perform a new prediction. The analysis device inputs the Morgan fingerprint, MOE molecular descriptor, and PLIF of the current candidate compound into an artificial neural network model (ANN) as a third model to perform prediction ( 340 ).

When the artificial neural network model classifies the current candidate compound as positive, the analysis device selects the candidate compound as a human PPAR gamma antagonist candidate. When the artificial neural network model classifies the current candidate compound as negative, the analysis device finally classifies the candidate compound as not a human PPAR gamma antagonist candidate.

4 is an experimental result for the prediction of antagonists of human PPAR gamma. 4 is a result of evaluating a compound in the public DB using a model constructed as a researcher. Whether a compound in the public DB is an antagonist of human PPAR gamma is known in advance.

The researcher used the CompTox Chemistry Dashboard. From the initial data, only structures applicable to MOE were selected and used. Only compounds capable of expressing compound secondary structure (SMILES) were targeted. The training data for building the model prepared in this way is 5,218 compounds. Of the training data, 543 were positive and 4,675 were negative. 2,454 presenters were used for the model. 2,048 Morgan fingerprints, 336 molecular descriptors, and 70 PLIFs were used.

DBSCAN uses Morgan fingerprint as input data. The researcher performed the process of selecting the input data. The researcher used an artificial neural network model (pre-processing model described above) learned in advance. The artificial neural network is composed of 6-layers in Keras. When the Morgan fingerprint of the candidate compound is input to the trained artificial neural network, each node of the layer has an active value. The researcher independently determined the activity value by repeating it 4 times.

DBSCAN can perform clustering by using the active value of each node as a presenter.

SVM classifies input data by supervised learning. SVM classifies the input data by maximizing the margin for each cluster based on the hyperplane.

The artificial neural network model can use various models capable of processing vector data. The researcher used an artificial neural network with 6 layers. The output value of an artificial neural network usually has a value between 0 and 1. The researcher classified the output values into 5 results based on 4 thresholds. The four thresholds were 0.5, 0.75, 0.875 and 0.984375. Among the five groups, the more forward the more likely the positive group, the higher the classification.

4 shows PPV (positive predict value) and NPV (negative positive value) values. Classification was sequentially using DBSCAN, SVM, and ANN. The researcher validated the model using validation data with known label values. 4 shows verification data consistent with the overall prediction result. In DBSCAN, 78.13% of the results classified as positive hit as positive data for verification. The remaining 21.87% were misclassified negative data for verification.

SVM classified the data classified as negative by DBSCAN. In SVM, 41.75% of the results classified as positive were correct as positive data for verification, and the remaining 58.25% were misclassified negative data for verification.

ANN classified the data classified as negative in the SVM as a target. The classification results are divided into five sets. 5 is a classification using the above-described threshold value. The ^3rd set is the lowest group, and the ^7th set is the highest group. That is, the ^7th set is the group with the highest negative probability. Based on the 7th set, ^97.85 % of the results classified as negative were hit as negative data for verification. Therefore, it can be seen that it has considerable accuracy.

5 is another experimental result for the prediction of antagonists of human PPAR gamma. 5 shows sensitivity and specificity. 5 shows prediction results for the entire verification data. ^1st set to ^7th set in FIG. 5 correspond to the results of classification by each learning model in FIG. 4 . DBSCAN classified 13.81% of positive data for verification as positive. DBSCAN classified 99.55.% of the negative data for verification as negative. SVM classified 59.67% of positive data as positive, and the cumulative value from DBSCAN was 73.48%. SVM classified 89.88% of the negative data as negative. ANN classified 14.18% of the positive data as positive based on the threshold between the 6th and 7th sets, and the cumulative value from DBSCAN was 87.66% of the total positive data for training. In addition, 65.20% of the negative data were classified as negative under the same criteria. Therefore, it can be seen that it has a significant classification efficiency.

6 is an example of the configuration of the analysis device 400 . The analysis device 400 corresponds to the aforementioned analysis devices (130 and 140 in FIG. 1). The analysis device 400 may be physically implemented in various forms. For example, the analysis device 400 may have the form of a computer device such as a PC, a server of a network, a chipset dedicated to data processing, and the like.

The analysis device 400 may include a storage device 410 , a memory 420 , an arithmetic device 430 , an interface device 440 , a communication device 450 , and an output device 460 .

The storage device 410 may store raw MOE data for a compound to be analyzed. The storage device 410 may store a Morgan fingerprint, an MOE molecular descriptor, and a protein-ligand-interaction fingerprint (PLIF) for a compound to be analyzed.

The storage device 410 may store a program code for analyzing a compound to be analyzed through the same process as described above.

The storage device 410 may store the aforementioned compound DB.

The storage device 410 may store a learning model used for analysis. The storage device 410 may store previously learned DBSCAN, SVM, and ANN.

The storage device 410 may store the analysis result (the human PPAR gamma antagonist evaluation result).

The memory 420 may store data and information generated by the analysis device 400 during the human PPAR gamma antagonist prediction process.

The interface device 440 is a device that receives predetermined commands and data from the outside. The interface device 440 may receive compound data of samples from a physically connected input device or an external storage device. The interface device 440 may receive raw data representing the structure or characteristics of a compound or data for analysis from the compound DB. Data for analysis include Morgan fingerprints, MOE molecular descriptors and protein-ligand-interaction fingerprints (PLIF). The interface device 440 may receive commands or information necessary for the human PPAR gamma antagonist prediction process. For example, the interface device 440 may receive an identifier of a compound to be analyzed.

The communication device 450 means a configuration for receiving and transmitting certain information through a wired or wireless network. The communication device 450 may receive compound data of samples from an external object. The communication device 450 may receive raw data representing the structure or characteristics of a compound or data for analysis from the compound DB. Data for analysis include Morgan fingerprints, MOE molecular descriptors and protein-ligand-interaction fingerprints (PLIF). The communication device 450 may receive commands or information necessary for the human PPAR gamma antagonist prediction process. For example, the communication device 450 may receive an identifier of a compound to be analyzed. The communication device 450 may transmit an analysis result of the compound to be analyzed to an external object.

The communication device 450 and the interface device 440 are devices that receive predetermined data or commands from the outside. The communication device 450 or the interface device 440 may be referred to as an input device to receive predetermined data.

The output device 460 is a device that outputs certain information. The output device 460 may output an interface necessary for a data processing process, an analysis result, and the like. The output device 460 may output the human PPAR gamma antagonist prediction result for the candidate compound.

The computing unit 430 performs an antagonist prediction of human PPAR gamma for the candidate compound using the program and the learning model stored in the storage unit 410 .

The computing device 430 may generate a Morgan fingerprint, an MOE molecular descriptor, and a protein-ligand-interaction fingerprint (PLIF) by using raw data for a compound to be analyzed.

The calculation unit 430 performs primary classification of the candidate compound by inputting the compound fingerprint of the candidate compound into DBSCAN. When the primary classification result is negative, the calculating unit 430 performs secondary classification on the candidate compound by inputting the molecular descriptor of the candidate compound into the SVM. When the secondary classification result is negative, the computing device 430 inputs the compound fingerprint, molecular descriptor, and protein-ligand interaction fingerprint of the candidate compound into the ANN to perform tertiary classification.

The computational device 430 may perform primary classification by inputting to the artificial neural network model, which is a pre-trained pre-processing model, of the compound fingerprint of the candidate compound, and inputting the values activated in the nodes of the trained artificial neural network model into the DBSACN to perform primary classification. have.

When the primary classification result, the secondary classification result, or the tertiary classification result is positive, the computing device 430 may determine the candidate compound as a candidate antagonist for human PPAR gamma ligand.

The computing device 430 may be a device such as a processor, an AP, or a program embedded chip that processes data and processes a predetermined operation.

In addition, as described above, the method for predicting an antagonist of human PPAR gamma may be implemented as a program (or application) including an executable algorithm that can be executed in a computer. The program may be provided by being stored in a temporary or non-transitory computer readable medium.

The non-transitory readable medium refers to a medium that stores data semi-permanently, rather than a medium that stores data for a short moment, such as a register, cache, memory, and the like, and can be read by a device. Specifically, the various applications or programs described above are CD, DVD, hard disk, Blu-ray disk, USB, memory card, ROM (read-only memory), PROM (programmable read only memory), EPROM (Erasable PROM, EPROM) Alternatively, it may be provided by being stored in a non-transitory readable medium such as an EEPROM (Electrically EPROM) or flash memory.

Temporarily readable media include Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced SDRAM (Enhanced) SDRAM, ESDRAM), Synchronous DRAM (Synclink DRAM, SLDRAM) and Direct Rambus RAM (Direct Rambus RAM, DRRAM) refers to a variety of RAM.

This embodiment and the drawings attached to this specification only clearly show a part of the technical idea included in the above-described technology, and within the scope of the technical idea included in the specification and drawings of the aforementioned technology, those skilled in the art can easily It will be said that it is obvious that all inferred modified examples and specific embodiments are included in the scope of the above-described technology.

Claims (12)

receiving, by the analysis device, an identifier of a candidate compound;
first classifying the candidate compound by the analysis device inputting the compound fingerprint of the candidate compound into an unsupervised learning clustering model;
secondarily classifying the candidate compound by inputting, by the analysis device, an in silico molecular descriptor of the candidate compound classified as negative in the primary classification into a supervised learning classification model; and
The analysis device inputs the compound fingerprint of the candidate compound classified as negative in the secondary classification, the molecular descriptor, and the protein-ligand interaction fingerprint for human PPAR gamma ligand into the artificial neural network model. Including the step of tertiary classification,
The unsupervised learning clustering model, the supervised learning classification model, and the artificial neural network model are models for predicting whether or not antagonists of human PPAR gamma ligand for each input data,
When any one of the unsupervised learning clustering model, the supervised learning classification model, and the artificial neural network model is classified as positive with respect to the candidate compound, the analysis device selects the candidate compound as an antagonist candidate for human PPAR gamma ligand. A method for predicting antagonists of human PPAR gamma based on a learning model that predicts substances. According to claim 1,
The unsupervised learning clustering model is a density-based spatial clustering of applications with noise (DBSCAN), and the supervised learning classification model is a support vector machine (SVM). According to claim 1,
The compound fingerprint is a Morgan fingerprint to which the Morgan algorithm is applied,
The molecular descriptor is a learning model-based method for predicting an antagonist of human PPAR gamma, which is a factor indicating molecular structure and properties in a Molecular Operating Environment (MOE). According to claim 1,
The molecular descriptor is a physical property descriptor, Huckel theory related factor, partitioned surface area factor, number of atoms, number of bonds, Kier-Hall molecular coupling factor, kappa shape indice, proximity, distance An antagonist prediction method of human PPAR gamma based on a learning model including matrix, partial charge, potential energy, pharmacophore model factors and MOPAC-related factors. According to claim 1,
The protein-ligand interaction fingerprint is a learning model-based method for predicting human PPAR gamma antagonists including information on target amino acids for the PPAR gamma ligand binding site as a result of docking simulation for the PPAR gamma ligand binding site. According to claim 1,
The analysis device inputs to an artificial neural network model trained in advance among the compound fingerprints of the candidate compound, and inputs the values activated in nodes of the learned artificial neural network model to the unsupervised learning clustering model to perform the primary classification A learning model-based method for predicting antagonists of human PPAR gamma. an input device for receiving a compound fingerprint of a candidate compound, an in silico molecular descriptor, and a protein-ligand interaction fingerprint for human PPAR gamma ligand;
a storage device for storing an unsupervised learning clustering model, a supervised learning classification model, and an artificial neural network model for predicting whether or not an antagonist to human PPAR gamma ligand; and
The compound fingerprint is input to the unsupervised learning clustering model to first classify the candidate compound, and when the primary classification result is negative, the molecular descriptor is input to the supervised learning classification model to input the candidate compound Calculating device for secondary classification, and tertiary classification by inputting the compound fingerprint, the molecular descriptor, and the protein-ligand interaction fingerprint to the artificial neural network model when the secondary classification result is negative including,
When any one of the unsupervised learning clustering model, the supervised learning classification model, and the artificial neural network model is classified as positive for the candidate compound, the computing device selects the candidate compound as an antagonist candidate for human PPAR gamma ligand. An analysis device that predicts the antagonist of human PPAR gamma predicted by the substance. 8. The method of claim 7,
The unsupervised learning clustering model is DBSCAN (Density-based spatial clustering of applications with noise), and the supervised learning classification model is an SVM (support vector machine) analysis apparatus for predicting an antagonist of human PPAR gamma. 8. The method of claim 7,
The compound fingerprint is a Morgan fingerprint to which the Morgan algorithm is applied,
The molecular descriptor is an analysis device for predicting an antagonist of human PPAR gamma, which is a factor indicating molecular structure and properties in Molecular Operating Environment (MOE). 8. The method of claim 7,
The molecular descriptor is a physical property descriptor, Huckel theory related factor, partitioned surface area factor, number of atoms, number of bonds, Kier-Hall molecular coupling factor, kappa shape indice, proximity, distance An analysis device for predicting antagonists of human PPAR gamma, including matrix, partial charge, potential energy, pharmacophore model factors and MOPAC-related factors. 8. The method of claim 7,
The protein-ligand interaction fingerprint is a docking simulation result for the PPAR gamma ligand binding site, and an analysis device for predicting human PPAR gamma antagonists including information on target amino acids for the PPAR gamma ligand binding site. 8. The method of claim 7,
The computing device inputs to an artificial neural network model trained in advance among the compound fingerprints of the candidate compound, and inputs the values activated in nodes of the learned artificial neural network model to the unsupervised learning clustering model to perform the primary classification An analysis device that predicts antagonists of human PPAR gamma.

Images