KR102363456B1 - Method and system for analyzing genome and medical information and developing drug component based on artificial intelligence - Google Patents

Abstract

The present invention relates to analysis of genome and medical information and development of drug components based on artificial intelligence. A system for analyzing an analysis genome and providing a drug component development platform comprises: an input unit for acquiring first data indicating a nucleotide sequence of the genome and second data including medical information; a first analysis unit for analyzing the first data and confirming the structure of protein generated by the genome based on the analysis result for the data; a second analysis unit for confirming information about symptoms occurring in a subject by analyzing the second data; a design unit which generates a structure of a compound for use as a drug component based on the structure of the protein, the analysis result for the first data, and the analysis result for the second data, and analyzes the properties of the compound; and an output unit for outputting the analysis result for the data and information related to the structure and properties of the compound. Therefore, the genome and medical information can be analyzed more effectively.

Description

AI-based genome and medical information analysis and pharmaceutical substance development method and system

The present invention relates to genome and medical information analysis and drug substance development, and to a method and system for analyzing genome and medical information based on artificial intelligence and developing a related drug substance.

Modern medicine is making remarkable progress. However, new diseases are also constantly occurring. In particular, infectious diseases pose a great threat to human society. Therefore, it is necessary to effectively study the disease and develop a new drug to respond to the disease based on the research results.

In general, it takes a lot of time and money to develop a new drug. For example, an analysis procedure to obtain basic data for new drug development is performed, based on this, a new drug substance with medicinal effect is designed, and verification is performed again. All of these processes involve chemical experiments, which require a lot of effort and time.

M. Skalic et al., "From Target to Drug: Generative Modeling for the Multimodal Structure-Based Ligand Design", Molecular Pharmaceutics, 16:4282-4291. (2019.08.22.)

An object of the present invention is to provide a method and system for effective gene and medical information analysis for diseases and development of pharmaceutical substances.

The present invention is to provide a method and system for analyzing a gene based on artificial intelligence.

The present invention is to provide a method and system for analyzing medical information based on artificial intelligence.

The present invention is to provide a method and system for developing pharmaceutical substances based on artificial intelligence.

The technical problems to be achieved in the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those of ordinary skill in the art to which the present invention belongs from the description below. will be able

A system for providing a genome and medical information analysis and pharmaceutical substance development platform according to an embodiment of the present invention includes an input unit for acquiring first data representing a nucleotide sequence of a genome and second data including medical information, the first A first analyzer that analyzes data and checks the structure of a protein generated by the genome based on the analysis result for the data, a second that identifies information about symptoms occurring in the subject by analyzing the second data An analysis unit, a design unit that generates a structure of a compound for use as a pharmaceutical substance based on the analysis unit, the structure of the protein, the analysis result for the first data, and the analysis result for the second data, and analyzes the properties of the compound; and an output unit for outputting an analysis result of the data and information related to the structure and properties of the compound. Here, the analysis unit and the design unit may operate using at least one artificial neural network.

A method for analyzing a genome and medical information and developing a drug substance according to an embodiment of the present invention includes acquiring first data representing a nucleotide sequence of a genome and second data including medical information, the first data analyzing and confirming the structure of the protein generated by the genome based on the analysis result for the data, verifying information on symptoms occurring in the subject by analyzing the second data, the structure of the protein, generating a structure of a compound for use as a pharmaceutical substance based on an analysis result for the first data and an analysis result for the second data, analyzing properties of the compound, and analysis result for the data and outputting information related to the structure and properties of the compound. Here, the analysis of the data, confirmation of the structure of the protein, generation of the structure of the compound, and analysis of properties of the compound may be performed using at least one artificial neural network.

The features briefly summarized above with respect to the invention are merely exemplary aspects of the detailed description of the invention that follows, and do not limit the scope of the invention.

According to the present invention, it is possible to more effectively analyze genes and medical information, and to more effectively develop pharmaceutical substances for responding to related diseases.

The effects obtainable in the present invention are not limited to the above-mentioned effects, and other effects not mentioned may be clearly understood by those of ordinary skill in the art to which the present invention belongs from the following description. will be.

1 is a diagram illustrating a system for analyzing a genome and developing a drug substance according to an embodiment of the present invention.
2 is a diagram showing the structure of a genome analysis and drug substance development system according to an embodiment of the present invention.
3 is a diagram illustrating an operation procedure of a system for analyzing a genome and developing a drug substance according to an embodiment of the present invention.
4 is a diagram illustrating an example of a genome analysis interface provided by the genome analysis platform according to an embodiment of the present invention.
5 is a diagram illustrating an example of a pharmaceutical material development interface provided by the pharmaceutical material development platform according to an embodiment of the present invention.
6 is a diagram showing the structure of an artificial neural network applicable to a genome analysis and drug substance development system according to an embodiment of the present invention.
7 is a diagram illustrating a connection structure of artificial neural networks applicable to a genome analysis and drug substance development system according to an embodiment of the present invention.
8 is a diagram illustrating a procedure for performing transfer learning in a genome analysis and drug substance development system according to an embodiment of the present invention.

Hereinafter, with reference to the accompanying drawings, embodiments of the present invention will be described in detail so that those of ordinary skill in the art to which the present invention pertains can easily implement them. However, the present invention may be embodied in several different forms and is not limited to the embodiments described herein.

In describing an embodiment of the present invention, if it is determined that a detailed description of a well-known configuration or function may obscure the gist of the present invention, a detailed description thereof will be omitted. And, in the drawings, parts not related to the description of the present invention are omitted, and similar reference numerals are attached to similar parts.

Accordingly, the present invention proposes a technology for diagnosing and analyzing a disease. Furthermore, the present invention proposes a technique for designing a compound that can be used as a medicinal substance, ie, a candidate substance, for responding to a diagnosed and analyzed disease.

1 is a diagram illustrating a system for analyzing a genome and developing a drug substance according to an embodiment of the present invention.

Referring to FIG. 1 , the genome analysis and drug substance development system includes a local device 110a, a local device 110b, and a server 120 connected to a communication network. 1 illustrates two local devices 110a, 110b,

The local device 110a and the local device 110b are used by a user who wants to diagnose and analyze a disease using the system. The local device 110a and the local device 110b may obtain input data, transmit the input data to the server 120 through the communication network, and receive data including the result of analysis from the server 120 . can

The server 120 provides a platform for diagnosing and analyzing a disease according to embodiments of the present invention and further designing a compound to be used as a medicinal substance, and performs diagnosis, analysis, and design algorithms. According to various embodiments, the algorithms of diagnosis, analysis, and design may be performed based on artificial intelligence. The server 120 performs operations such as molecular diagnosis, genetic analysis, medical information analysis, and compound design for pharmaceutical substances based on data received from at least one of the local device 110a and the local device 110b, and results It transmits data to at least one of the local device 110a and the local device 110b. For example, the server 120 may be a cloud server.

As described above, according to an embodiment, the local device 110a and the local device 110b are terminals that perform data input and output functions, and the server 120 performs diagnosis, analysis, and design functions. can According to another embodiment, the local device 110a and the local device 110b may perform at least some calculations for diagnosis, analysis, and design. The degree of sharing of calculations for diagnosis, analysis, and design may be different for each local device. Furthermore, according to another embodiment, a local device including all functions of the server 120 may also exist. In this case, the local device 110a or the local device 110b may perform diagnosis, analysis, and design operations even if it is not connected to a communication network.

As described with reference to FIG. 1 , the server 120 may provide a platform for analyzing/diagnosing a disease and developing/designing a pharmaceutical substance. The platform according to an embodiment forms a green zone for new drug development and may provide disease diagnosis and prediction services. Furthermore, the platform according to an embodiment provides a bioinformatics service based on genome big data information using artificial intelligence.

2 is a diagram showing the structure of a genome analysis and drug substance development system according to an embodiment of the present invention. The components illustrated in FIG. 2 may be included in one of a local device (local device 110a or local device 110b in FIG. 1) and a server (eg, server 120 in FIG. 1), and each component How it is deployed in the local device and the server may vary according to various embodiments. Accordingly, the connection between the components may be based on an internal circuit or an external communication network.

Referring to FIG. 2 , the system includes a data input unit 210 , a genetic information analysis unit 220 , a medical information analysis unit 230 , a design unit 240 , and an output unit 250 .

The data input unit 210 is a means for inputting data to be analyzed. Here, the data includes genomic data and medical information data. Medical information data is the result of a subject's medical examination, and includes diagnostic charts, images (eg, X-rays, computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET)).

etc.) and the like. The genomic data includes at least one of data related to a subject (eg, a patient) or data related to a virus/bacterium causing a disease. Specifically, the genomic data includes information on the DNA or RNA base sequence of a subject or virus/bacterium. For example, the data input unit 210 may receive dielectric data through an external communication network or receive dielectric data through local hardware (eg, a memory port or a user input device). A user who wants to use the system may input genome data through the data input unit 210 . In this case, the genome data may be input in the form of a file configured according to a predefined format.

The genetic information analysis unit 220 analyzes the genome data input through the data input unit 210 . The genetic information analysis unit 220 may analyze the genomic data to determine the genetic characteristics present in the nucleotide sequence of the subject or virus/bacterium, and estimate the protein structure or intracellular activity that may be generated according to the nucleotide sequence. there is. Accordingly, the protein structure generating unit 226 may acquire data related to diseases caused by the infiltration of external viruses/bacteria as well as diseases resulting from the subject's gene. To this end, the genetic information analysis unit 220 includes a genome analysis unit 222 , a molecular diagnosis analysis unit 224 , and a protein structure generation unit 226 .

The genome analyzer 222 may analyze the genome data using various techniques. The genome analyzer 222 analyzes the subject's genome data to determine the genetic characteristics. The genome analyzer 222 may check the variation of a specific nucleotide sequence on the subject's genome. For example, the genome analyzer 222 is a single nucleotide polymorphism (SNP) analysis method, a single strand conformation polymorphism (SSCP) analysis method, an Amplified Fragment Length Polymorphism (AFLP) analysis method, a Random Amplified Polymorphic DNAs (RAPD) analysis method, an AS-PCR ( Genomic data can be analyzed using a variety of analysis techniques, such as Allele-Specific PCR) analysis, Dynamic Allele-Specific Hybridization (DASH), Whole-Genome Sequencing (WGS), and Next Generation Sequencing (NGS).

The molecular diagnosis analysis unit 224 performs molecular diagnosis based on the genome data. In other words, the molecular diagnostic analysis unit 224 identifies various molecular-level activities that may occur in the subject's cells based on the genomic data. Specifically, the molecular diagnostic analysis unit 224 detects changes in various molecular levels occurring in cells through numerical values or images. For example, the molecular diagnostic analysis unit 224 may perform nucleic acid analysis such as DNA or RNA, protein analysis, intracellular metabolome analysis, and the like.

The protein structure generating unit 226 predicts the structure of a protein that can be generated by the subject's gene based on the genomic data. That is, the protein structure generating unit 226 predicts the structure of the disease-causing protein. If the genome data of the virus/bacterium is secured, the protein structure generator 226 may also predict the structure of a protein that may be generated by the virus/bacterium gene.

As described above, the genetic information analysis unit 220 includes a genome analysis unit 222 , a molecular diagnosis analysis unit 224 , and a protein structure generation unit 226 . Analysis operations performed by the genetic information analyzer 220 are performed using genomic data (eg, an electronic file) rather than an actual gene. Accordingly, as compared to an analysis using an actual gene, the analysis can be performed in a shorter time and at a low cost. For effective analysis of a vast amount of genome data, the genetic information analyzer 220 may use an artificial neural network (ANN) based on machine learning or deep learning. In this case, the genetic information analyzer 220 may use a pre-trained artificial neural network, or may use it after directly learning an initial artificial neural network.

The medical information analysis unit 230 analyzes medical information data. By analyzing the medical information data, the medical information analyzing unit 230 may acquire information on symptoms of diseases occurring in the subject's body. The medical information data may include image data. In this case, for effective analysis, the medical information analyzer 230 may filter the image to improve image quality or to enhance main information. For example, the filtering may include at least one of fast Fourier transform (FFT), histogram equalization, motion artifact removal, or noise canceling. In addition, the medical information analysis unit 230 may extract feature points from the medical image, determine a symptom based on the extracted feature points, and convert the determined symptom into data. That is, based on the analysis result of the medical image, the medical information analysis unit 230 may determine the state of the subject, whether the subject is infected, the progress of the disease, the severity of the disease, and the like.

The design unit 240 is based on the genomic data and medical information input through the data input unit 210 , the analysis result of the genome data by the genetic information analysis unit 220 , and the analysis result of the medical information by the medical information analysis unit 230 . based on designing compounds that can be used as pharmaceutical substances. To this end, the design unit 240 includes a characteristic compound generation unit 242 and an ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) prediction unit 244 .

The characteristic compound generating unit 242 generates a structure of the characteristic compound corresponding to the protein structure predicted by the protein structure generating unit 226 . A characteristic compound is generated to have a structure capable of interacting with (eg, binding to) a predicted protein, and is generated to have a property capable of inhibiting disease by interacting with the predicted protein.

The ADMET prediction unit 244 performs absorption, distribution, metabolism, excretion, and toxicity tests for the compound generated by the characteristic compound generation unit 242 . In other words, the ADMET prediction unit 244 analyzes the ADMET profile and drug action mechanism of the compound generated by the characteristic compound generation unit 242 . That is, the ADMET prediction unit 244 analyzes how the compound generated by the characteristic compound generation unit 242 acts in the subject's cells based on the obtained genomic data. At this time, the ADMET prediction unit 244 may build a mathematical model for the characteristics of the subject based on the obtained genomic data, and generate an ADMET profile based on the constructed mathematical model.

As described above, the design unit 240 includes a characteristic compound generation unit 242 and an ADMET prediction unit 244 . Analysis operations performed by the design unit 240 are performed using data (eg, an electronic file) of a compound rather than an actual compound. For effective design, the genetic information analysis unit 220 may use an artificial neural network based on machine learning or deep learning. In this case, as an input of the artificial neural network, the analysis result of the genome generated by the genetic information analysis unit 220 and the analysis result of the medical image generated by the medical information analysis unit 230 may be used. Accordingly, compared to the analysis using an actual compound, the design can be made in a shorter time and at a low cost. In this case, the design unit 240 may use a pre-learned artificial neural network or may use the initial artificial neural network after directly learning it.

The output unit 250 outputs the results generated by the genetic information analysis unit 220 , the medical information analysis unit 230 , and the design unit 240 . For example, the output unit 250 may output the result as a tangible object (eg, a print) or in the form of a digital file. In this case, the digital file may be stored in a storage means within the system or transmitted to a pre-designated destination (eg, a designated e-mail, a designated phone number, a designated application) through an external communication network.

In the structure described with reference to FIG. 2 , at least one function of the genetic information analysis unit 220 and the design unit 240 and at least some functions of the data input unit 210 and the output unit 250 are performed by a processor. can be performed. In addition, although not shown in FIG. 2 , the system may include an interface means for interaction with a user. For example, the system may include input means such as a keyboard, mouse, touch screen, and the like, and display means, such as a monitor, touch screen, projector, and the like. In the following description, it may be understood that the input from the user and the description of the interface shown to the user are inputted through the input means and displayed through the display means.

3 is a diagram illustrating an operation procedure of a system for analyzing a genome and developing a drug substance according to an embodiment of the present invention. Hereinafter, for convenience of description, a subject performing the operations of FIG. 3 is referred to as a 'system'.

Referring to FIG. 3 , in step 301 , the system acquires genomic data and medical information. The obtained genomic data may include DNA or RNA nucleotide sequence data of the subject. Additionally, DNA or RNA nucleotide sequence data of viruses or bacteria that cause disease may be further included.

In step 303, the system analyzes the genomic data. The system may perform various genome analysis techniques based on the genome data. The system can use the genome data as an input value and analyze the genome data using an artificial neural network. To this end, the system can process the genomic data into a form that can be input to the artificial neural network. Through this, the system can estimate the genetic characteristics of the subject or virus/bacterium, molecular activity, structure of the produced protein, and the like. Due to the use of artificial neural networks, the system can acquire analytical data on the genome without actual chemical experiments.

In step 305, the system analyzes the medical information. In particular, according to an embodiment, the system may analyze a medical image of the subject and identify symptoms based on the analysis result. Specifically, the system may extract feature points after filtering the image, and determine a symptom based on the extracted feature points. In this case, the system may use an artificial neural network. For example, the system may use a convolutional neural network (CNN) to identify and classify lesions. In this case, a different artificial neural network may be used depending on the type of image.

At step 307 , the system designs the characteristic compound and analyzes the properties of the designed characteristic compound. That is, based on the analysis result performed in step 303 and the analysis result of the medical information performed in step 305, the system designs a compound that specifically responds to the subject or virus/bacterium. For example, the system may contain a compound that interacts with a protein produced by a subject's gene that causes a disease, a compound that destroys a virus/bacterium, a compound that inhibits the activity of a protein produced by the virus/bacterium, a compound that inhibits the activity of a protein produced by the virus/bacterium At least one of the compounds that inhibit protein production by The designed compound may be referred to as a candidate substance, a candidate drug, and the like. And, the system generates an ADMET profile for the designed at least one compound. That is, the system can perform simulations to predict how the designed compound will behave within the subject's cells. In other words, the system performs a reactivity test of a genome to a candidate substance, a side effect test, and an activity prediction. The operation of step 307 may be performed based on artificial intelligence using an artificial neural network. Due to the use of artificial neural networks, the system can acquire compounds and analytical data on compounds without actual chemical experiments.

In step 309, the system outputs the analysis result. The analysis result may be output in the form of a tangible object (eg, printed matter) or in the form of a digital file. In this case, the digital file may be stored in a storage means within the system or transmitted to a pre-designated destination (eg, a designated e-mail, a designated phone number, a designated application) through an external communication network.

The subject of the operations described with reference to FIG. 3 has been described as a system. However, the above-described operations may be performed by one device (eg, the local devices 110a and 110b or the server 120 ) or may be performed by two or more devices. When performed by two or more devices, between some operations, an operation in which necessary information is exchanged between the two or more devices may be added. For example, when step 301 is performed by a local device and step 303 is performed by a server, an operation in which the local device transmits genome data to the server prior to step 303 may be added.

According to the embodiment described with reference to FIG. 3 , the system designs a characteristic compound and analyzes the properties of the designed compound. Although not shown in FIG. 3 , if the properties of the analyzed compound do not achieve the reference effect as a medicinal substance or have unacceptable side effects, the system may exclude the compound from the candidate substance. In this case, the system may redesign another compound, or reselect another one of a plurality of previously designed compounds, and then analyze the properties. This operation may be repeated until a compound meeting the criteria is determined or a predetermined failure condition is satisfied.

A user who intends to perform disease diagnosis and analysis and drug substance development using the system may input necessary data using a local device and command analysis. Accordingly, the local device or server provides a result to the user by performing the operations described with reference to FIG. 3 . In this case, the local device may provide an interface according to various embodiments for interaction of a user's data input, command input, and the like. That is, the local device may display various interfaces using the display means. Examples of interfaces are described below with reference to FIGS. 4 and 5 .

4 is a diagram illustrating an example of a genome analysis interface provided by the genome analysis platform according to an embodiment of the present invention. Referring to FIG. 4 , the genome analysis interface includes a menu 410 , a search bar 420 , and function items 430 . Each of the function items 431 to 439 may be composed of an image and a name expressing an analysis target or technique. For example, the function items 430 are epigenomics analysis items 431, exome analysis items 432, GWAS (genome-wide association study) analysis items 433, metabolite analysis ( metabolomics item (434), pangenomics item (435), protein analysis item (436), target sequencing item (437), transcriptome analysis item (438), and at least one of Whole-Genome Sequencing (WGS) analysis items 439 . After inputting the genome data, the user can proceed with the corresponding analysis technique by clicking or selecting the item of the function to be utilized for analysis.

Using the interface as shown in FIG. 4, in the genome analysis, based on the extracted genome data, the next-generation nucleotide sequence such as full genome analysis, microRNA target gene and disease association prediction analysis, mass transcript analysis, gene interaction network analysis, etc. Tools for analysis and the like may be provided. Such a platform can be understood as a convergence technology of IT with disease diagnosis and drug substance development, which enables the utilization of data that is significantly increased every year according to the trend of big data and convergence and by the improvement of NGS technology. That is, the present invention is based on analysis software for analyzing and interpreting a vast amount of genomic information in various fields, and it is possible to provide a platform that can be used in various industries by using it to various researchers or companies.

5 is a diagram illustrating an example of a pharmaceutical material development interface provided by the pharmaceutical material development platform according to an embodiment of the present invention. Referring to FIG. 5 , the drug substance development interface includes a menu 510 , a search bar 520 , and function items 530 . Each of the function items 530 may be composed of an image and a name expressing an analysis target or technique. For example, the function items 530 are ADME (Absorption, Distribution, Metabolism, Excretion) analysis item 531, Basic Science Research B item 532, and biomarker development item. (533), lead identification item (534), lead optimization item (535), protein interaction analysis item (536), target validation item (537) , and at least one of a toxicity analysis item 538 . After the analysis result for the genomic data is derived, the user can proceed with the analysis technique by clicking or selecting the function item to be utilized for drug substance development.

In the case of the pharmaceutical industry with growth potential, problems arise due to the lack of step-by-step linkage between existing new drug development projects and the existence of a gap with the industrial field, and there is a limit to deriving tangible results. Various uncertainties in the field of candidate substance derivation, optimization, non-clinical, and clinical fields increase the possibility of market failure. Accordingly, the present invention intends to provide a big data-based platform utilizing artificial intelligence. Through this, developers will be able to increase the efficiency in the process of developing new drugs.

As described above, the system and platform according to an embodiment of the present invention provide functions for analyzing a genome and developing a pharmaceutical substance based on the analysis result. In this case, the calculation algorithm for analysis and development is performed using data, not an actual genome sample. Furthermore, a calculation algorithm for analysis and development may be implemented based on artificial intelligence (AI).

Artificial intelligence means that machines such as computers perform thinking, learning, and analysis that are possible with human intelligence. Recently, the technology for applying artificial intelligence to the medical industry is increasing. To implement artificial intelligence, artificial neural networks are widely used. An example of an artificial neural network applicable to the present invention is shown in FIG. 6 below.

6 is a diagram showing the structure of an artificial neural network applicable to a genome analysis and drug substance development system according to an embodiment of the present invention. Referring to FIG. 6 , the artificial neural network includes an input layer 610 , at least one hidden layer 620 , and an output layer 630 . Each of the layers 610 , 620 , and 630 includes a plurality of nodes, and each of the nodes is connected to the 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 of the previous layer and the corresponding connection weight, and then a non-linear activation function The output value multiplied by is delivered to at least one neuron in the next layer.

Artificial neural network models used in various embodiments of the present invention include a fully convolutional neural network, a convolutional neural network, a recurrent neural network, and a restricted Boltzmann machine (RBM). ) and at least one of a deep belief neural network (DBN), but is not limited thereto. Alternatively, machine learning methods other than deep learning may be included. Alternatively, it may include a hybrid model that combines deep learning and machine learning. For example, when a feature of an image is extracted by applying a deep learning-based model, and an image is classified or recognized based on the extracted feature, a machine learning-based model may be applied. The machine learning-based model may include, but is not limited to, a support vector machine (SVM), an AdaBoost, and the like.

An artificial neural network configured similarly to the example of FIG. 6 may be used. In this case, the system according to an embodiment of the present invention may have a plurality of artificial neural networks. The plurality of artificial neural networks may be classified according to functions. For example, the plurality of artificial neural networks may include at least one artificial neural network for gene analysis and at least one artificial neural network for drug substance development. In addition, the at least one artificial neural network for genetic analysis may be divided into at least one artificial neural network for analyzing the genome of a subject and at least one artificial neural network for analyzing the genome of a virus/bacterium. Furthermore, the plurality of artificial neural networks may further include at least one artificial neural network for selecting an artificial neural network to be used for analysis or development.

7 is a diagram illustrating a connection structure of artificial neural networks applicable to a genome analysis and drug substance development system according to an embodiment of the present invention.

Referring to FIG. 7 , the system includes a first artificial neural network set 710 including a plurality of artificial neural networks for genome analysis, a second artificial neural network set 720 including a plurality of artificial neural networks for medical image data analysis, and and a third artificial neural network set 730 including a plurality of artificial neural networks for drug substance development.

The plurality of artificial neural networks included in the first artificial neural network set 710 may be classified according to functions (eg, an analysis method, whether the analysis target is a target or a virus/bacterium). For example, the first artificial neural network set 710 may include artificial neural networks corresponding to each of the items 431 to 408 illustrated in FIG. 4 . Also, a plurality of artificial neural networks included in the first artificial neural network set 710 may have different structures. Specifically, the plurality of artificial neural networks may be different from each other in the number of input nodes, types of input values, and the like. Accordingly, the system may confirm the command by the user through the user interface, and process the input genome data according to the types of input values required by the artificial neural network corresponding to the confirmed command.

A plurality of artificial neural networks included in the second artificial neural network set 720 may be classified according to functions (eg, an analysis method, whether an analysis target is a target or a virus/bacterium). For example, the second artificial neural network set 720 may include artificial neural networks corresponding to the type of medical image (eg, X-ray, CT, or MRI). Also, a plurality of artificial neural networks included in the second artificial neural network set 720 may have different structures. Specifically, the plurality of artificial neural networks may be different from each other in the number of input nodes, types of input values, and the like. Accordingly, the system may confirm a command by the user through the user interface, and process the input medical image data according to the types of input values required by the artificial neural network corresponding to the confirmed command.

A plurality of artificial neural networks included in the third artificial neural network set 730 may be classified according to functions (eg, an analysis method, whether the analysis target is a target or a virus/bacterium). For example, the third artificial neural network set 730 may include artificial neural networks corresponding to each of the items 531 to 509 illustrated in FIG. 5 . Also, a plurality of artificial neural networks included in the third artificial neural network set 730 may have different structures. Specifically, the plurality of artificial neural networks may be different from each other in the number of input nodes, types of input values, and the like. Therefore, the system confirms the command by the user through the user interface, and the genomic analysis result data and medical Image analysis result data can be processed.

The artificial neural networks included in the second artificial neural network set 720 may be learned using medical image data. For medical image data, it may be difficult to secure sufficient learning data for reasons such as protection of personal information. Accordingly, the artificial neural networks included in the second artificial neural network set 720 may be learned using learning data secured through data argumentation. In particular, data augmentation may be performed on data on lesions in medical image data. That is, the system can secure the learning data by extracting a lesion region from the input medical image data and performing data augmentation on the extracted lesion region data. For example, data augmentation may include rotation, scaling, and the like.

According to an embodiment, the third artificial neural network set 730 includes an artificial neural network (hereinafter referred to as a 'compound design model') for designing a characteristic compound. The compound design model uses a result inferred from at least one of the artificial neural networks included in the first artificial neural network set 710 as an input. In addition, the compound design model can use as input the results of attribute analysis (eg, ADMET profile) of other compounds (eg, compounds that were not selected as final compounds and were rejected). In addition, the compound design model can use as input data on protein-to-compound interactions collected from the outside.

To this end, the system may include a data collection unit (not shown) that performs data crawling. The data collection unit collects information on compounds related to the structure of proteins obtained through genome analysis using an external data network and provides them to be used in the compound design model. For example, the data collection unit may collect information about the compound by periodically or event-based web search or academic database search. The collected information is processed and input to the input nodes of the compound design model.

In this case, the search range may vary based on the learning history of the protein structure obtained through genome analysis. For example, if there is experience in which the obtained protein structure or similar structure is used as learning data to perform learning, it can be expected that the accuracy of inference through the compound design model is high. Therefore, the system can search the obtained protein structure in the past learning history data, and determine the amount or range of data to collect through crawling according to the search result. Specifically, the system determines whether at least the number of keywords for data retrieval, the content of the keyword, and the size of data to be collected, based on whether there is an experience in which the same or similar protein structure has been learned, and the degree of similarity to the protein structure with the learning experience. decide one Here, the degree of similarity of the protein structure may be determined based on the type, linkage structure, size, etc. of amino acids constituting the protein. To this end, the system is a database (not shown) storing information on learning data used for learning of artificial neural networks (eg, artificial neural networks in the first artificial neural network set 710 and the third artificial neural network set 730). city) may be included. The above-described data collection unit may also perform a function of collecting learning data.

As in the above-described embodiments, the system according to an embodiment of the present invention performs genome analysis, molecular diagnosis, protein structure prediction, characteristic compound design, compound property inspection, etc. based on input genome data. At this time, since the amount of input genome data is huge, in order to increase the efficiency of calculation, the system can perform necessary functions while increasing the complexity of the algorithm step by step.

According to an embodiment, the system may use a plurality of artificial neural networks capable of performing the same function. For example, there may be a plurality of artificial neural networks for performing protein analysis (proteomics), and the plurality of artificial neural networks may have different structures in reasoning accuracy and computational complexity/time. In this case, the system infers the operation with one artificial neural network that matches the initially set accuracy and computational complexity/time, and if the result does not converge, then using the next artificial neural network with higher inference accuracy. Inference can be performed again. Here, the convergence of the results may be determined based on feedback from the user (eg, input of evaluation for the result), whether a predefined criterion is met based on the determined attribute (eg, comparison of quantified attribute values and thresholds). there is. Here, whether a predefined criterion is satisfied based on the determined attribute may be fed back from the design unit 240 to the genetic information analysis unit 220 .

In this case, based on the accumulated reasoning result, the system may set initial accuracy and computational complexity/time differently for each function. In addition, the system may adjust the initial values of accuracy and computation time according to the number of functions the user wants to perform. Alternatively, the system may receive basic data for accuracy and calculation time from a user, and determine and apply an initial value based on the input basic data.

As in the above-described embodiments, the system according to an embodiment of the present invention performs functions necessary for analysis and development using an artificial neural network. For accurate inference using an artificial neural network, it is required that the artificial neural network is in a properly trained state. To this end, the system may further include a learning unit. However, in the initial stage of the system and platform according to an embodiment of the present invention, it may be difficult to expect accurate inference due to a lack of learning amount. In particular, when a system and a platform are built based on a local device rather than a server-based one, the decrease in accuracy of inference due to lack of learning may be greater.

As a method to solve this problem, transfer learning may be applied. Transfer learning is a learning method that applies variables (eg, weight values) from another artificial neural network that have already been trained to the corresponding artificial neural network. That is, during the initial installation process of the system, according to a user's command or satisfaction of a given condition, transfer learning using an artificial neural network of another system according to an embodiment of the present invention may be performed. Transfer learning may be performed through a procedure as shown in FIG. 8 below.

8 is a diagram illustrating a procedure for performing transfer learning in a genome analysis and drug substance development system according to an embodiment of the present invention. In FIG. 8 , each of system A and system B is independent systems that provide a platform according to embodiments of the present invention, and may be a system built on a local device basis or a server basis.

Referring to FIG. 8 , in step 801, the system B registers at least one learned artificial neural network in a sharable state. That is, the user of system B registers allowing the learned artificial neural network to be used for transfer learning of other systems (eg, system A). Although not shown in FIG. 8 , information on whether to register may be stored in a separate support server. For registration, system B displays a list of artificial neural networks through a display means, confirms selection and registration commands by the user, and then carries out the processing for registration (e.g., creating a shared whitelist or notifying a separate support server). can be done

In step 803, system A sends a request for shared information. In other words, system A requests information on at least one sharable artificial neural network. In step 805, the system B transmits the shared information. The shared information includes a list of at least one sharable artificial neural network. The list may include information on a structure, information on a function, and information on a learning amount in at least one shareable artificial neural network.

In step 807, the system A selects at least one artificial neural network to use for transfer learning. That is, the system A selects at least one artificial neural network to be used for transfer learning of the artificial neural network to be learned in the system A from among the artificial neural networks included in the shared information provided from the system B. For example, the system A may select the artificial neural network of the system B having the same function as the artificial neural network to be learned. In the example of FIG. 8 , shared information was received only from system B, but system A receives shared information from other systems and selects an artificial neural network to be used for transfer learning by comprehensively considering the shared information provided from a plurality of systems. can

In step 809, the system A transmits a request for information related to the selected artificial neural network. In step 811, the system B transmits information related to the requested artificial neural network. Information related to the artificial neural network may include information necessary for transfer learning (eg, weight values).

In step 813, the system A performs transfer learning. In other words, system A performs transfer learning based on information related to the artificial neural network provided from system B. In this case, the system A may reuse all of the weight values included in the received information in its own artificial neural network as it is, or selectively reuse only some of the weight values. Whether or not to reuse a part may be determined based on the amount of learning of the artificial neural network provided from the system B, the similarity of the structure, and the like.

As in the embodiment described with reference to FIG. 8 , transfer learning may be performed through interaction between different systems. In this case, according to another embodiment, a separate support system supporting interaction between systems may exist. In this case, each system registers information about artificial neural networks that allow sharing in the support system, and the system that wants transfer learning performs request-response signaling with the support system to obtain data and information necessary for transfer learning. can do.

Furthermore, the support system provides information on the learning performance of the artificial neural network in another system (eg, system B) after transfer learning of the corresponding system (eg, system A), and the system that received the information (eg, system A) ) can display information about the further learning progress of the artificial neural network in other systems to the user through the interface. For example, the system may display information on the progress of additional learning in real time through a separate notification, or display information on the progress of additional learning when execution of a corresponding function is commanded. Accordingly, since the user may request to perform transfer learning again, interaction between the systems may be increased. According to another embodiment, information on the learning performance of the artificial neural network may be directly received from another system.

Exemplary methods of the present invention are expressed as a series of actions for clarity of description, but this is not intended to limit the order in which the steps are performed, and each step may be performed simultaneously or in a different order if necessary. In order to implement the method according to the present invention, other steps may be included in addition to the illustrated steps, steps may be excluded from some steps, and/or other steps may be included except for some steps.

Various embodiments of the present invention do not list all possible combinations but are intended to illustrate representative aspects of the present invention, and the details described in various embodiments may be applied independently or in combination of two or more.

In addition, various embodiments of the present invention may be implemented by hardware, firmware, software, or a combination thereof. For implementation by hardware, one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), general purpose It may be implemented by a processor (general processor), a controller, a microcontroller, a microprocessor, and the like.

The scope of the present invention includes software or machine-executable instructions (eg, operating system, application, firmware, program, etc.) that cause operation according to the method of various embodiments to be executed on a device or computer, and such software or and non-transitory computer-readable media in which instructions and the like are stored and executable on a device or computer.

Claims (20)

In a system that provides a platform for analyzing genome and medical information and developing a drug substance,
an input unit for obtaining first data representing a nucleotide sequence of a genome and second data including medical information;
The first data is analyzed through at least one of genomic analysis methods to confirm the mutation of a specific sequence on the genome, and based on the first data, an activity analysis at the intracellular level is performed, and for the first data a first analysis unit for predicting a structure of a disease-causing protein generated by a protein-encoding gene in the genome based on the analysis result;
a second analysis unit for confirming information on symptoms occurring in the subject by analyzing the second data;
a collection unit that collects data on the compound related to the structure of the protein and provides it to the design unit;
Generating the structure of a compound for use as a pharmaceutical substance by determining the structures of the compound capable of interacting with the structure of the protein based on the structure of the protein, the analysis result for the first data, and information on the symptom, Analyze the properties of the compound, including its pharmacokinetics, through simulations to predict the action of the structure of the compound in the cells of the subject and absorption, distribution, metabolism, excretion and toxicity tests for the structure of the compound, and , the design unit excluding the structure of the compound from a candidate material when the property of the compound has a side effect based on a result of analyzing the property of the compound; and
and an output unit for outputting analysis results for the first data and the second data and information related to structures and properties of the remaining compounds not excluded from the candidate material,
Analysis of molecular level activity in the cell includes nucleic acid analysis, protein analysis, and intracellular metabolite analysis of DNA (deoxyribonucleic acid) or RNA (ribonucleic acid),
The first analysis unit, the second analysis unit, and the design unit operate using at least one artificial neural network,
The collection unit crawls the data on the protein-to-compound interaction by performing a web search and an academic database search through an external communication network, and processes the crawled data according to the structure of the input layer of the artificial neural network, ,
The collection range and amount of the crawled data is determined by whether there is an experience in which a protein structure identical or similar to the structure of the compound exists, and if the learned experience exists, between the same or similar protein structure and the structure of the compound. It is determined based on the similarity,
The degree of similarity is determined based on at least one of the type, linkage structure and size of amino acids constituting the same or similar protein structure,
The output unit, epigenomics analysis item, exome analysis item, GWAS (genome-wide association study) analysis item, metabolite analysis (metabolomics) item, pangenomic analysis (metagenomics) item, protein analysis (proteomics) ) item, target sequencing item, transcriptome analysis item, WGS(Whole-Genome Sequencing) analysis item, ADME(Absorption, Distribution, Metabolism, Excretion) analysis item, Basic Science Research B(Basic Science Research) B) item, biomarker development item, lead identification item, lead optimization item, protein interaction analysis item, target validation item and toxicity analysis (toxicity analysis) display the interface including items,
The system of claim 1, wherein the simulation includes a simulation of a reactivity test for a structure of the compound, a side effect test, and a prediction of activity.
The method according to claim 1,
The first data is a system including genome information of a subject who is to take a drug containing the pharmaceutical substance and genome information of a virus or bacteria causing a target disease.
The method according to claim 1,
Extracting a lesion region from the medical image data included in the medical information, securing learning data by performing data augmentation on the extracted lesion region data, and using the secured learning data to the second analysis unit A system further comprising a learning unit for learning artificial neural networks used by the
The method according to claim 1,
The compound is a substance that specifically reacts with the protein, a compound that destroys viruses or bacteria, a compound that inhibits the activity of a protein produced by the virus or the bacteria, and protein production by the virus or the bacteria A system comprising at least one of an inhibitory compound.
delete The method according to claim 1,
The number of functions specified by the user, the accuracy requested by the user, and the calculation time required by the user, among a plurality of artificial neural networks in which the first analysis unit and the second analysis unit have different inference accuracy and computational complexity A system for selecting at least one artificial neural network to be used for analysis of the first data and the second data based on the data.
The method according to claim 1,
The design unit determines whether the compound meets a predefined criterion based on the attribute, and feeds back whether the compound meets the predefined criterion to the first analyzer and the second analyzer,
When it is fed back that the compound does not meet the predefined criteria, the first analyzer and the second analyzer use at least one other artificial neural network having higher inference accuracy than the last used at least one artificial neural network. A system that re-runs the analysis.
The method according to claim 1,
Further comprising a learning unit for performing learning of the at least one artificial neural network,
The learning unit receives a list of artificial neural networks allowing sharing for transfer learning from another system, selects at least one artificial neural network to be used for transfer learning based on the list, and selects the selected at least one artificial neural network A system for receiving information related to , and performing transfer learning on an artificial neural network of the system based on the received information.
9. The method of claim 8,
The learning unit determines a range to reuse the weight values of the selected at least one artificial neural network based on a learning amount of the selected at least one artificial neural network and the similarity of the structure.
9. The method of claim 8,
The learning unit receives information on performing additional learning of the selected at least one artificial neural network in the other system after the transfer learning,
The information on performing the additional learning is displayed to the user through an interface.
A method for analyzing genomic and medical information and developing a drug substance, the method comprising:
acquiring, from the input unit, first data representing the nucleotide sequence of the genome and second data including medical information;
In the first analysis unit, the first data is analyzed through at least one of the genome analysis methods to confirm the variation of a specific nucleotide sequence on the genome, and based on the first data, an activity analysis at the intracellular level is performed, and , predicting the structure of a disease-causing protein generated by a gene encoding a protein in the genome based on the analysis result for the first data;
In the second analysis unit, by analyzing the second data, confirming information about the symptoms occurring in the subject;
In the collecting unit, collecting data on the compound related to the structure of the protein and providing it to the design unit;
In the design unit, the structure of the compound for use as a pharmaceutical substance is determined by determining the structures of the compound capable of interacting with the structure of the protein based on the structure of the protein, the analysis result for the first data, and information on the symptoms. generating;
In the design section, absorption, distribution, metabolism, excretion, and toxicity tests for the structure of the compound and simulation for predicting the action of the structure of the compound in the cell of the subject, including pharmacokinetics of the compound analyzing the properties, and excluding the structure of the compound from a candidate material when the properties of the compound have side effects based on a result of analyzing the properties of the compound; and
outputting, in the output unit, the analysis result for the first data, the analysis result for the second data, and information related to the structure and properties of the remaining compounds not excluded from the candidate material,
Analysis of molecular level activity in the cell includes nucleic acid analysis, protein analysis, and intracellular metabolite analysis of DNA (deoxyribonucleic acid) or RNA (ribonucleic acid),
Analysis of the first data, analysis of the second data, confirmation of the structure of the protein, generation of the structure of the compound, and analysis of properties of the compound are performed using at least one artificial neural network,
In the collecting unit, the data on the protein-to-compound interaction is crawled by performing a web search and an academic database search through an external communication network, and the crawled data is processed according to the structure of the input layer of the artificial neural network. becomes,
The collection range and amount of the crawled data is determined by whether there is an experience in which a protein structure identical or similar to the structure of the compound exists, and if the learned experience exists, between the same or similar protein structure and the structure of the compound. It is determined based on the similarity,
The degree of similarity is determined based on at least one of the type, linkage structure and size of amino acids constituting the same or similar protein structure,
In the output unit, epigenomics analysis item, exome analysis item, GWAS (genome-wide association study) analysis item, metabolite analysis (metabolomics) item, pangenomic analysis (metagenomics) item, protein analysis ( proteomics item, target sequencing item, transcriptome analysis item, WGS(Whole-Genome Sequencing) analysis item, ADME(Absorption, Distribution, Metabolism, Excretion) analysis item, Basic Science B(Basic Science) Research B) item, biomarker development item, lead identification item, lead optimization item, protein interaction analysis item, target validation item, and toxicity An interface including items of toxicity analysis is displayed;
The simulation is a method comprising a simulation for the reactivity test, side effect test, activity prediction for the structure of the compound.
12. The method of claim 11,
The first data may include genome information of a subject who is to take a drug containing the pharmaceutical substance and genome information of a virus or bacteria causing a target disease.
12. The method of claim 11,
extracting, by the learning unit, a lesion region from medical image data included in the medical information;
securing, in the learning unit, learning data by performing data augmentation on the extracted data on the lesion region; and
The method further comprising the step of learning, in the learning unit, the artificial neural networks used by the second analysis unit by using the secured learning data.
12. The method of claim 11,
The compound is a substance that specifically reacts with the protein, a compound that destroys viruses or bacteria, a compound that inhibits the activity of a protein produced by the virus or the bacteria, and protein production by the virus or the bacteria A method comprising at least one of an inhibitory compound.
delete 12. The method of claim 11,
In the first analysis unit and the second analysis unit, analyzing the first data and the second data includes:
Among a plurality of artificial neural networks having different inference accuracy and computational complexity, the first data and the second data based on the number of functions specified by the user, the accuracy requested by the user, and the computation time requested by the user selecting at least one artificial neural network to use for analysis of
12. The method of claim 11,
determining, in the design unit, whether the compound meets a predefined criterion based on the attribute;
If the compound does not meet the predefined criteria, the analysis is repeated using at least one other artificial neural network having higher inference accuracy than the at least one artificial neural network last used in the first and second analyzers. The method further comprising the step of performing.
12. The method of claim 11,
receiving, in the learning unit, a list of artificial neural networks allowing sharing for transfer learning from other systems;
selecting, in the learning unit, at least one artificial neural network to be used for the transfer learning based on the list;
receiving, in the learning unit, information related to the selected at least one artificial neural network;
The method further comprising the step of performing, in the learning unit, transfer learning on the artificial neural network of the method based on the received information.
19. The method of claim 18,
In the learning unit, performing the transfer learning includes:
and determining a range to reuse the weight values of the selected at least one artificial neural network based on a learning amount of the selected at least one artificial neural network and the similarity of the structure.
19. The method of claim 18,
Further comprising the step of receiving, in the learning unit, information on performing additional learning of the selected at least one artificial neural network in the other system after the transfer learning,
The information on performing the additional learning is displayed to the user through an interface.

Images