KR20220094179A - Method and apparatus for predicting of mechanism of actions of novel compounds based on transcriptome phenotype - Google Patents

Abstract

According to an embodiment of the present invention, a transcriptome phenotype-based drug action mechanism predicting device comprises: a communication module; a memory in which a drug action mechanism predicting program is stored; and a processor for executing the drug action mechanism predicting program. The drug action mechanism prediction program inputs an embedding vector for a target material to a learning model built to locate an embedding vector for a compound and an embedding vector for transcriptome phenotype feature information induced by each compound in the same vector space and outputs ranking of a drum action mechanism matching at least one piece of transcriptome phenotype feature information outputted by the learning model.

Description

Transcriptome phenotype-based drug mechanism prediction apparatus and method

The present invention relates to a method and apparatus for predicting a drug action mechanism for a target chemical substance based on a transcriptome phenotype.

New drugs are developed by a process consisting of a new drug discovery stage and a new drug development stage. The new drug discovery phase includes target identification, candidate substance design, efficacy measurement, and drug candidate selection. The new drug development phase includes safety evaluation and clinical trials of drug candidates. It takes considerable time and money to commercialize a drug through the new drug discovery stage and new drug development stage, but it is known that the success rate is not high.

In the drug development pipeline, it is very important to identify a target protein suitable for a disease and find a molecule that binds to the target. Once a target for a disease is identified, a compound capable of binding to the target is found through high-efficiency screening, and structural analogues of the target-binding drug are also selected as drug candidates. Although more than 5,000 to 10,000 are selected as drug candidates, the success rate from testing and verification to sales is less than 0.02%, resulting in high cost and time-consuming development of new drugs.

The traditional method used to discover new drug candidates in the new drug discovery stage is a method of discovering new drug candidates based on a target. These include the discovery of target proteins, a process that reveals major factors related to disease, hit discovery, the process of finding compounds that can physically bind to the target protein and inhibit its function, and lead to structural optimization of the previously found active substances. It proceeds to the material optimization process (Lead optimization). Among the drug candidates discovered from the target protein, drugs that have effects on cells, tissues, and individuals are finally selected through the development stage. In this case, in addition to the target protein, if there are proteins having a structure similar to the target protein or having a similar binding pocket to the drug, the unintentionally developed drug binds to the corresponding proteins, resulting in an unintended effect (off-target effect). ), which can cause drug toxicity and side effects. If the protein and pathway that the drug affects, that is, the mechanism of action of the drug can be predicted, the structure of the drug can be designed to minimize unintended effects during the optimization process.

However, 1) a target protein hypothesis is essential, 2) even if the target protein hypothesis is found, drug development may be difficult if the compound has an undruggable target that cannot be bound to the target protein, 3) the target protein A limitation of the target-oriented drug development process is that it is difficult to experimentally verify a myriad of compounds even if the structure can be combined with a compound, and it takes about 5.5 years or more to derive candidate substances.

On the other hand, the development of new drugs based on the transcriptome phenotype first discovers drugs that can suppress diseases at the level of the transcriptome phenotype. Therefore, it has an advantage in that it can start the development of new drugs even for diseases for which there is no hypothesis, such as rare intractable diseases. In the process of developing new drugs based on the transcriptome phenotype, it is necessary to find out the mechanism by which the substance suppressed the disease by targeting which protein and pathway after discovering the hit substance.

Accordingly, in the present invention, it is intended to propose a method for predicting a drug action mechanism that can be used to reveal proteins and pathway targets of novel compounds.

Korean Patent Laid-Open Patent Publication No. 10-2018-0058648 (Title of the invention: a method for discovering a new drug candidate targeting a non-structural-structural transition site and an apparatus for discovering a new drug candidate)

In order to solve the above problems, according to an embodiment of the present invention, a drug action mechanism for a target compound can be output through a learning model built on the basis of information on the structure of the compound and information on the transcriptome phenotype characteristic. An object of the present invention is to provide an apparatus and method for predicting a drug mechanism of action.

However, the technical task to be achieved by the present embodiment is not limited to the technical task as described above, and other technical tasks may exist.

As a technical means for achieving the above technical problem, an apparatus for predicting a drug action mechanism according to an embodiment of the present invention includes a communication module; a memory in which a drug action mechanism prediction program is stored; a process of executing the drug mechanism of action prediction program, wherein the drug mechanism prediction program is constructed such that an embedding vector for a compound and an embedding vector for transcript phenotypic characteristic information induced by each compound are located in the same vector space An embedding vector for a target substance is input to the trained learning model, and a ranking of a drug action mechanism matched with at least one transcript phenotype characteristic information output by the learning model is output.

In addition, the transcript phenotype-based drug mechanism prediction method according to another embodiment of the present invention is constructed so that the embedding vector for the compound and the embedding vector for the transcript phenotype characteristic information induced by each compound are located in the same vector space providing a learned learning model; outputting at least one transcript phenotype characteristic information having a high degree of similarity to the embedding vector of the target material to which the learning model is input, as the embedding vector for the target material is input to the learning model; and outputting a ranking of a drug action mechanism that matches the transcriptome phenotype characteristic information.

According to the above-described means for solving the problems of the present invention, the time and cost consumed in the discovery step of discovering new drug candidates can be greatly reduced.

In addition, new drug development is possible even when a target protein hypothesis does not exist or when a target protein is known but a substance that actually binds to the protein cannot be made.

In addition, it identifies proteins and pathway targets of discovered new drug candidates, and uses the identified proteins and pathway targets to optimize the lead material in the process of developing new drugs based on transcriptome phenotype.

1 is a diagram illustrating the configuration of an apparatus for predicting a drug action mechanism according to an embodiment of the present invention.
2 is a flowchart for explaining the operation of a method for predicting a drug action mechanism according to an embodiment of the present invention.
3 shows the configuration of a learning model for predicting a drug action mechanism according to an embodiment of the present invention.
4 shows the configuration of an encoder for building a learning model according to an embodiment of the present invention.
5 is a diagram illustrating a configuration of a GP feature used when building a learning model according to an embodiment of the present invention.
6 is a diagram illustrating a drug action mechanism ranking output process according to an embodiment of the present invention.

Hereinafter, embodiments of the present application will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art to which the present application pertains can easily carry out. However, the present application may be implemented in several different forms and is not limited to the embodiments described herein. And in order to clearly explain the present application in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Throughout this specification, when a part is "connected" with another part, this includes not only the case where it is "directly connected" but also the case where it is "electrically connected" with another element interposed therebetween. do.

Throughout this specification, when a member is said to be located “on” another member, this includes not only a case in which a member is in contact with another member but also a case in which another member is present between the two members.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram illustrating the configuration of an apparatus for predicting a drug action mechanism according to an embodiment of the present invention.

As shown, the apparatus 100 for predicting a drug action mechanism may include a communication module 110 , a memory 120 , a processor 130 , and a database 140 . The drug action mechanism prediction apparatus 100 is configured based on a computing device, and further includes, not shown, a power supply unit, various input devices and output devices, and the like.

The communication module 110 may transmit/receive data about the chemical structure of the compound to and from an external computing device, or data regarding the transcript amount characteristic information induced by the compound as matching it. The communication module 110 may be a device including hardware and software necessary for transmitting and receiving signals such as control signals or data signals through wired/wireless connection with other network devices.

The memory 120 stores a drug action mechanism prediction program. The drug mechanism prediction program provides a learning model constructed so that the embedding vector for the chemical structure of the compound and the embedding vector for the transcript phenotypic characteristic information induced by each compound are located in the same vector space, and provides the learning model with the target substance Inputs an embedding vector for the chemical structure of , and outputs a ranking of a drug action mechanism that matches at least one transcript phenotypic characteristic information output by the learning model. In this case, the query input to the drug action mechanism prediction program may be data on the chemical structure of the new substance. In addition, the drug action mechanism prediction program may additionally perform a logic for building a drug learning model or a learning model update process for performing a new learning process on the built drug learning model.

In the memory 120 , various types of data generated during the execution of an operating system for driving the drug action mechanism prediction apparatus 100 or a missing data prediction program are stored.

In this case, the memory 120 collectively refers to a non-volatile storage device that continuously maintains stored information even when power is not supplied, and a volatile storage device that requires power to maintain the stored information.

In addition, the memory 120 may perform a function of temporarily or permanently storing data processed by the processor 130 . Here, the memory 120 may include magnetic storage media or flash storage media in addition to the volatile storage device that requires power to maintain stored information, but the scope of the present invention is limited thereto. it's not going to be

The processor 130 executes the program stored in the memory 120, but controls the entire process according to the execution of the drug action mechanism prediction program. Each operation performed by the processor 130 will be described in more detail later.

The processor 130 may include all kinds of devices capable of processing data. For example, it may refer to a data processing device embedded in hardware having a physically structured circuit to perform a function expressed as code or instructions included in a program. As an example of the data processing apparatus embedded in the hardware as described above, a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated (ASIC) circuit) and a processing device such as a field programmable gate array (FPGA), but the scope of the present invention is not limited thereto.

The database 140 stores or provides data necessary for the apparatus for predicting a drug action mechanism under the control of the processor 130 . The database 140 may be included as a component separate from the memory 120 , or may be built in a partial area of the memory 120 .

2 is a flowchart for explaining the operation of a method for predicting a drug action mechanism according to an embodiment of the present invention.

First, the apparatus 100 for predicting a drug action mechanism provides a learning model constructed so that the embedding vector for the chemical structure of the compound and the embedding vector for the transcript phenotype characteristic information induced by each compound are located in the same vector space ( S210).

Such a learning model may be composed of a multi-layered artificial neural network for learning chemical structure information of a compound and a multi-layered artificial neural network for learning transcript mass characteristic information.

Figure 3 shows the configuration of a learning model for predicting a drug action mechanism according to an embodiment of the present invention, Figure 4 shows the configuration of an encoder for building a learning model according to an embodiment of the present invention , FIG. 5 shows the configuration of a GP feature used when building a learning model according to an embodiment of the present invention.

As shown, the learning model has a technical feature in that the embedding vector for the chemical structure of the compound and the embedding vector for the characteristic information of the transcript amount generated when the compound is injected as a drug are located in the same vector space.

First, 2048 bits of extended-connectivity fingerprints (ECFP) data may be used to represent the chemical structure of a compound. It is binary data representing the chemical structure of an individual drug or compound, with each bit representing the presence or absence of the chemical structure. As such, ECFP corresponds to published data.

In addition, the transcript phenotype characteristic information is characteristic information indicating the total amount of RNA expressed by the compound, and data can be collected by using several known datasets. For example, an Integrated Network-based Cellular Signatures (LINCS) L1000 data set library may be used, and among them, characteristic information of 978 landmark genes may be utilized.

In this way, the encoder artificial neural network is applied to each of the data representing the chemical structure of the compound and the data representing the transcript phenotype characteristic information, respectively, as shown in FIG. 4 , and an embedding vector is generated for each.

Each encoder artificial neural network may be constructed in a form including a plurality of multilayer perceptron (MLP) layers. For example, a first encoder artificial neural network (f _str ) embedding the chemical structure of a compound may include 4 MLP layers embedding 2048 input data (X _str ) as 512-256-256, through which Create an embedding vector (Z _str ) representing the chemical structure of the compound.

In addition, the second encoder artificial neural network (f _sig ) for embedding transcriptome phenotypic feature information may include 4 MLP layers embedding as 512-512-256-256, based on 978 input data (X _sig ). and, through this, an embedding vector (Z _sig ) indicating transcript phenotype characteristic information is generated.

On the other hand, as shown in FIG. 5 as transcript phenotype characteristic information, gene pertubation information may be further included and embedded. At this time, the gene perturbation information includes information on which a specific gene is removed (knocked out) or the expression of a gene is suppressed (knocked down). Through this, gene perturbation information induced by the compound can be generated as an embedding vector.

For example, if an embedding vector for an EGFR inhibitor is input, the gene perturbation information expressed thereby is learned as an embedding vector, and in particular, it is placed in a position close to each other in the same vector space. At this time, the gene perturbation information is also provided as 978-bit data and can be utilized to generate an embedding vector.

On the other hand, for the construction of a learning model, an embedding vector for the chemical structure of a compound and an embedding vector for characteristic information on the amount of transcript generated when the compound is injected as a drug are constructed in the same vector space, and as a loss function for this, We use the triplet loss function. In this loss function, the embedding vector representing the transcript phenotype characteristic information is set as the anchor vector (Z _sig ), and the embedding vector of the compound that induces the transcript phenotypic characteristic corresponding to the anchor vector is a positive vector (Z _str ^p ), Set the embedding vector of a compound that does not induce a transcriptome phenotypic characteristic corresponding to the anchor vector as a negative vector (Z _str ⁿ ), minimize the distance between the anchor vector and the positive vector, and maximize the distance between the anchor vector and the negative vector. to do it The structure of the specific function is as shown in Equation 1.

[Equation 1]

)를 이용하여 계산한다.In this case, the distance of each vector is the cosine similarity (

) is used to calculate

According to this configuration, a learning model in which an embedding vector for a compound and an embedding vector for transcript phenotypic characteristic information induced by the compound are located in the same vector space can be constructed. Accordingly, when information on the structure of a specific compound is input, information on the phenotype characteristic of the transcriptome expressed by the information can be predicted or inferred.

Referring back to FIG. 2 , an embedding vector for a target material is input to the learning model built through the above process, and accordingly, the learning model receives at least one transcript phenotype characteristic information with high similarity to the embedding vector of the target material. output (S220).

As described above, in the learning model of the present invention, since the embedding vector for the compound and the embedding vector for the transcript phenotype characteristic information are built in one space, if the embedding vector for a drug or various compounds is input as input data, , at least one transcript phenotype characteristic information most relevant thereto may be output. In this case, the output transcript phenotype characteristic information may include the gene perturbation information described above.

In this case, the connectivity score may be used to calculate the degree of similarity, and the connectivity score may be calculated through the following equation.

[Equation 2]

In this case, the embedding vector (Z _c ) represents the embedding vector of the compound, and the embedding vector (Z _g ) represents the transcript phenotype characteristic information. If their cosine similarity is the greatest, the connectivity score is considered high. For example, if an embedding vector for an EGFR inhibitor is input, EGFR KD gene perturbation with a high degree of similarity is output as transcript phenotype characteristic information.

Next, the ranking of the drug action mechanism matching the transcript phenotype characteristic information output by the learning model is output (S230).

The transcript phenotype characteristic information output in the previous step (S220) is arranged in descending order according to the degree of similarity, and pathway analysis is performed on each of them.

6 is a diagram illustrating a drug action mechanism ranking output process according to an embodiment of the present invention.

When the transcript phenotype characteristic information highly related to the target substance is determined, a pathway analysis that recommends ranking information of the mechanism of action of the drug may be performed based on this.

The pathway DB includes various data on the mechanism of action of a drug, and a GSEA (Gene Set Enrichment Analysis) method is known as a method for its analysis. The GSEA method is a technique that simultaneously evaluates whether genes related to one biological characteristic are expressed with a statistically significant difference under two conditions (ie, normal cells vs. cancer cells, etc.) and whether the expressed characteristics are similar to each other.

Through such a process, in the present invention, information on transcript phenotypic characteristics or gene perturbation that is highly relevant to a new input compound can be inferred, and the mechanism of action of the drug in which the genes are enriched can be sequenced. output according to

An embodiment of the present invention may also be implemented in the form of a recording medium including instructions executable by a computer, such as a program module to be executed by a computer. Computer-readable media can be any available media that can be accessed by a computer and includes both volatile and nonvolatile media, removable and non-removable media. Also, computer-readable media may include computer storage media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data.

Although the methods and systems of the present invention have been described with reference to specific embodiments, some or all of their components or operations may be implemented using a computer system having a general purpose hardware architecture.

The above description of the present application is for illustration, and those of ordinary skill in the art to which the present application pertains will understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present application. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a dispersed form, and likewise components described as distributed may be implemented in a combined form.

The scope of the present application is indicated by the following claims rather than the above detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be construed as being included in the scope of the present application.

100: drug mechanism prediction device
110: communication module
120: memory
130: processor
140: database

Claims (7)

In the apparatus for predicting the mechanism of action of a drug based on a transcriptome phenotype,
communication module;
a memory in which a drug action mechanism prediction program is stored;
Including the process of executing the drug mechanism of action prediction program,
The drug action mechanism prediction program inputs the embedding vector for the target substance into the learning model constructed so that the embedding vector for the compound and the embedding vector for the transcript phenotypic characteristic information induced by each compound are located in the same vector space, The apparatus for predicting a drug action mechanism based on transcriptome phenotype to output a ranking of a drug action mechanism that matches at least one or more transcript phenotype characteristic information output by the learning model. The method of claim 1,
The learning model minimizes the distance between the first embedding vector representing the transcriptome phenotype feature information and the embedding vector of the compound inducing the transcript phenotype feature through a triplet loss function, and minimizes the distance between the first embedding vector and the transcript phenotype A device for predicting a mechanism of action of a drug, configured to maximize the distance from the embedding vector of a compound that does not induce a characteristic. The method of claim 1,
The transcript phenotype characteristic information will further include gene perturbation information, drug action mechanism prediction device. In a method for predicting a mechanism of action of a drug based on a transcriptome phenotype,
providing a learning model constructed so that an embedding vector for a compound and an embedding vector for transcript phenotypic characteristic information induced by each compound are located in the same vector space;
outputting at least one transcript phenotype characteristic information having a high degree of similarity to the embedding vector of the target material to which the learning model is input, as the embedding vector for the target material is input to the learning model; and
A method of predicting a drug mechanism of action based on a transcriptome phenotype, comprising outputting a ranking of a drug action mechanism matching the transcript phenotype characteristic information. 5. The method of claim 4,
The learning model minimizes the distance between the first embedding vector representing the transcriptome phenotype feature information and the embedding vector of the compound inducing the transcript phenotype feature through a triplet loss function, and minimizes the distance between the first embedding vector and the transcript phenotype A method for predicting a mechanism of action of a drug based on a transcriptome phenotype, wherein the compound that does not induce a characteristic is configured to maximize the distance from the embedding vector. 5. The method of claim 4,
The transcript phenotype characteristic information will further include gene perturbation information, a method for predicting a mechanism of action based on a transcriptome phenotype. A computer-readable recording medium in which a program for performing the method according to any one of claims 4 to 6 is recorded.

Images