Method and device for obtaining binding pose between protein receptor and ligand by using artificial intelligence model

Abstract

A device according to an aspect comprises: at least one memory; and at least one processor, wherein the at least one processor extracts interaction information of a protein and a ligand to input the extraction information to an artificial intelligence model, obtains a candidate binding pose of the protein and the ligand as output data of the artificial intelligence model, and obtains a final binding pose on the basis of a charge of the ligand bound in the candidate binding pose.

Description

Method and device for obtaining binding poses between protein receptors and ligands using artificial intelligence models

[0001] The present disclosure relates to a method and device for obtaining a binding pose between a protein receptor and a ligand using an artificial intelligence model.

[0002] In recent drug development programs, docking simulation is a frequently used method to discover candidate substances by predicting the interaction between target proteins and ligands.

[0003] In ligand screening for a specific protein, docking simulations must find the precise binding pose between the protein receptor and the ligand, which allows for the determination of the relative orientation of the ligand within the binding site.

[0004] However, because proteins and ligands have structural characteristics, it has been difficult to identify binding patterns between proteins and ligands.

[0005] Therefore, the present disclosure aims to provide a method for more accurately and efficiently deriving binding postures between proteins and ligands and calculating binding energy at that time by applying artificial intelligence models and quantum mechanics (QM) and molecular mechanics (MM) calculations to docking.

[0006] The present invention provides a method and device for obtaining binding poses between protein receptors and ligands using an artificial intelligence model. Furthermore, the present invention provides a computer-readable recording medium containing a program for executing the method on a computer. The technical challenges to be addressed are not limited to the technical challenges described above, and other technical challenges may exist.

[0007] According to one aspect of the present disclosure, a method for obtaining a binding pose between a protein receptor and a ligand using an artificial intelligence model may be provided, the method comprising: a step of extracting interaction information between a protein and a ligand and inputting the extracted information into an artificial intelligence model; a step of obtaining a candidate binding pose of the protein and the ligand as output data of the artificial intelligence model; and a step of obtaining a final binding pose based on a charge of the ligand determined as the candidate binding pose.

[0008] According to another aspect of the present disclosure, a device includes a memory storing at least one program; and at least one processor executing the at least one program; wherein the at least one processor extracts interaction information between a protein and a ligand and inputs it into an artificial intelligence model, obtains candidate binding poses of the protein and the ligand as output data of the artificial intelligence model, and obtains a final binding pose based on the charge of the ligand determined as the candidate binding pose.

[0009] A computer-readable recording medium according to another aspect of the present disclosure includes a recording medium having recorded thereon a program for executing the above-described method on a computer.

[0010] Using artificial intelligence models, binding poses between proteins and ligands with geometric structures can be accurately identified, allowing for accurate calculation of binding affinity.

[0011] In addition, by introducing an artificial intelligence model, the number of QM calculations can be reduced compared to the existing QM / MM docking method, thereby shortening the required time.

[0012] However, the effects of the embodiments are not limited to the effects mentioned above, and other effects not mentioned can be clearly understood by a person having ordinary skill in the art from the description of the present invention.

[0013] FIG. 1 is a diagram illustrating an example of a method for obtaining a binding pose between a protein receptor and a ligand using an artificial intelligence model in one embodiment.

[0014] FIG. 2 is a schematic diagram illustrating an example of a device for obtaining a binding pose between a protein receptor and a ligand using an artificial intelligence model according to one embodiment.

[0015] FIG. 3 is a flowchart illustrating an example of a method for obtaining a binding pose between a protein receptor and a ligand using an artificial intelligence model according to one embodiment.

[0016] FIG. 4 is a diagram illustrating an example of a method for generating multiple binding poses of a protein and a ligand according to one embodiment.

[0017] FIG. 5 is a diagram illustrating an example of a method for determining a candidate combination pose according to one embodiment.

[0018] FIG. 6 is a diagram illustrating an example of a method for calculating QM / MM values ​​of candidate combined poses according to one embodiment.

[0019] A device according to one aspect comprises at least one memory; and at least one processor; wherein the at least one processor extracts interaction information between a protein and a ligand and inputs it into an artificial intelligence model, obtains candidate binding poses of the protein and the ligand as output data of the artificial intelligence model, and obtains a final binding pose based on the charge of the ligand bound to the candidate binding pose.

[0020] The terms used in the examples are selected from widely used, current terms, as much as possible. However, these terms may vary depending on the intentions of those skilled in the art, precedents, the emergence of new technologies, etc. Furthermore, in certain cases, the applicant may arbitrarily select terms, in which case their meanings will be described in detail in the relevant description. Therefore, the terms used in the specification should be defined based on their intended meaning and the overall content of the specification, rather than simply their names.

[0021] When a part of the specification is said to "include" a component, this does not exclude other components, but rather implies the inclusion of other components, unless otherwise specifically stated. Furthermore, terms such as "unit" and "module" used throughout the specification refer to a unit that processes at least one function or operation, which may be implemented in hardware, software, or a combination of hardware and software.

[0022] Additionally, terms including ordinal numbers, such as "first" or "second," used in the specification may be used to describe various components, but the components should not be limited by the terms. The terms may be used to distinguish one component from another.

[0023] The present disclosure will be described in detail with reference to the attached drawings. However, the embodiments may be implemented in various different forms and are not limited to the examples described herein.

[0024] FIG. 1 is a diagram illustrating an example of a method for obtaining a binding pose between a protein receptor and a ligand using an artificial intelligence model in one embodiment.

[0025] Referring to Fig. 1, the binding pose (100) between a protein (10) and a ligand (20) can be derived using an artificial intelligence model (1).

[0026] For example, an artificial intelligence model (1) can derive a binding pose most similar to the native pose by using interaction information between a protein (10) and a ligand (20).

[0027] During the new drug development process, developers can discover new drug candidates by performing docking simulations using information on the interaction between a target protein (10) and a ligand (20). Here, docking generally refers to a computational simulation of the binding between a receptor (e.g., a protein) and a candidate ligand (20). In the field of molecular modeling, docking is used as a method to predict the preferred orientation of a first molecule relative to a second molecule when two molecules bind to each other to form a stable complex.

[0028] At this time, the developer needs to perform a docking simulation between a specific receptor, such as a protein (10), and a ligand (20) to derive an accurate binding pose (100) of the receptor-ligand (20) complex. Here, the binding pose (100) may refer to a binding posture of the receptor and ligand (20) generated as docking is performed.

[0029] Accordingly, the developer can extract interaction information between a protein (10) and a ligand (20) and use the extracted interaction information as input data for an artificial intelligence model (1). In addition, the developer can obtain candidate binding poses of a protein (10) and a ligand (20) as output data of the artificial intelligence model (1).

[0030] Additionally, the developer can calculate the QM / MM values ​​of the candidate binding poses of the protein (10) and the ligand (20) and use them to derive a new charge value of the ligand (20).

[0031] Therefore, the developer can obtain the final binding pose by replacing the ligand (20) charge value with a new charge value, and through this, the binding energy of the protein (10) and the ligand (20) can also be calculated.

[0032]

[0033] FIG. 2 is a schematic diagram illustrating an example of a device for obtaining a binding pose between a protein receptor and a ligand using an artificial intelligence model according to one embodiment.

[0034] Referring to FIG. 2, a device (hereinafter referred to as "device") (200) for obtaining a binding pose between a protein receptor and a ligand using an artificial intelligence model may include a communication unit (210), a processor (220), and a memory (230). Only components related to the embodiment are illustrated in the device (200) of FIG. 2. Therefore, it will be apparent to those skilled in the art that other general components may be included in addition to the components illustrated in FIG. 2.

[0035] The communication unit (210) may include one or more components that enable wired / wireless communication with an external server or external device. For example, the communication unit (210) may include a short-range communication unit (not shown) and a mobile communication unit (not shown) for communication with an external server or external device.

[0036] The processor (220) controls the overall operation of the device (200). For example, the processor (220) can control the input unit (not shown), the display (not shown), the communication unit (210), the memory (230), etc., by executing programs stored in the memory (230).

[0037] The processor (220) may be implemented using at least one of application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, micro-controllers, microprocessors, and other electrical units for performing functions.

[0038] The processor (220) can control the operation of the device (200) by executing programs stored in the memory (230). As an example, the processor (220) can perform at least a part of the method of obtaining a binding pose between a protein receptor and a ligand using an artificial intelligence model, which is described with reference to FIGS. 3 to 6.

[0039] The memory (230) is hardware that stores various data processed within the device (200), and can store a program for processing and controlling the processor (220).

[0040] For example, the memory (230) may store various data, such as candidate binding poses of proteins and ligands, interaction information, learning data of an artificial intelligence model, and data generated according to the operation of the processor (220). In addition, the memory (230) may store an operating system (OS) and at least one program (e.g., a program required for the processor (220) to operate).

[0041] The memory (230) may include random access memory (RAM) such as dynamic random access memory (DRAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), CD-ROM, Blu-ray or other optical disk storage, hard disk drive (HDD), solid state drive (SSD), or flash memory.

[0042]

[0043] FIG. 3 is a flowchart illustrating an example of a method for obtaining a binding pose between a protein receptor and a ligand using an artificial intelligence model according to one embodiment.

[0044] Referring to FIG. 3, a method for obtaining a binding pose between a protein receptor and a ligand using an artificial intelligence model may include steps 310 to 330. However, the method is not limited thereto, and other general steps may be further included in the method for obtaining a binding pose between a protein receptor and a ligand using an artificial intelligence model, in addition to the steps illustrated in FIG. 3. Furthermore, as described above with reference to FIGS. 1 and 2, at least one of the steps in the flowchart illustrated in FIG. 3 may be processed by a processor.

[0045] At step 310, the processor can extract interaction information between proteins and ligands and input it into an artificial intelligence model.

[0046] For example, the processor may generate multiple binding poses of a protein and a ligand, extract interaction information for each of the generated multiple binding poses, and input the extracted interaction information into an artificial intelligence model as input data for the artificial intelligence model. Here, the artificial intelligence model may be a model trained using interaction information for each of the multiple protein receptors and the multiple ligands as training data.

[0047] In step 320, the processor can obtain candidate binding poses of the protein and ligand as output data of the artificial intelligence model.

[0048] For example, the processor can generate binding poses of a protein and a ligand, calculate a similarity between the generated binding pose and the native pose of the protein and the ligand, and determine a binding pose whose calculated similarity is less than or equal to a preset value as a candidate binding pose.

[0049] In step 330, the processor can obtain a final binding pose based on the charge of the ligand determined as the candidate binding pose.

[0050] For example, the processor can use the QM / MM values ​​of the candidate binding poses to calculate a replacement charge value for the ligand, change the charge value of the ligand to the calculated replacement charge value, and dock the ligand with the changed charge value to the replacement charge value and the protein to obtain the final binding pose.

[0051] Additionally, the processor can produce the binding energy of the final binding pose.

[0052]

[0053] Hereinafter, with reference to FIGS. 4 to 6, a method for obtaining a binding pose between a protein receptor and a ligand using the artificial intelligence model described above with reference to FIG. 3 will be described in more detail.

[0054] FIG. 4 is a diagram illustrating an example of a method for generating multiple binding poses of a protein and a ligand according to one embodiment.

[0055] Hereinafter, with reference to FIG. 4, an example of a method for a processor to generate multiple combined poses is described.

[0056] Referring to FIG. 4, the processor can generate multiple binding poses (410, 420) of a protein and a ligand.

[0057] For example, the processor can generate multiple binding poses (410, 420) of a protein and a ligand.

[0058] For example, the processor can determine the binding site at which a protein and a ligand will bind. In other words, the processor can determine each binding site based on the location of the ligand as a result of surface analysis of the protein.

[0059] Additionally, the processor can generate a plurality of binding poses (410, 420) in which the protein and ligand are bound based on the determined binding site.

[0060] For example, the processor can extract interaction information for each of the generated multiple combined poses (410, 420).

[0061] For example, the processor can obtain interaction information of multiple binding poses (410, 420) of a protein and a ligand using a molecular dynamics simulation (MD Simulation) method.

[0062] For example, the processor can set up a simulation environment for obtaining interaction information of a plurality of combined poses (410, 420). That is, the processor can obtain interaction information of each of the plurality of combined poses (410, 420) by variously changing the simulation environment, such as ion concentration and force field.

[0063] Additionally, the processor can store the extracted interaction information in memory as a Protein-Ligand Interaction Fingerprint (PLIF). PLIF is one of the methods for quantitatively representing interaction information between proteins and ligands, and can be a concise representation of specific interaction patterns between two molecules.

[0064] For example, the processor can express the extracted interaction information as a vector. Specifically, the processor can extract the interaction information in a 9-bit format and express it as a binary vector by concatenating the extracted bits into a single string. In addition, each bit can represent, but is not limited to, an interaction contact, a backbone interaction, a sidechain interaction, polar residues, hydrophobic residues, an H-bond acceptor, an H-bond donor, an aromatic residue, and a charge residue.

[0065] Additionally, the processor can extract ligands bound to proteins in the form of a 1024-bit Morgan fingerprint vector and perform principal component analysis (PCA) of the extracted vector. Here, the processor can classify ligand and interaction information based on the sum of the square elbow points within a cluster using the K-means algorithm.

[0066] Accordingly, the processor can input the extracted interaction information as input data into the AI ​​model. Here, the AI ​​model may be a model trained using interaction information for each of multiple protein receptors and multiple ligands as training data.

[0067] For example, an artificial intelligence model refers to a set of machine learning algorithms that use a hierarchical algorithmic structure based on a deep neural network in machine learning technology and cognitive science.

[0068] For example, an artificial intelligence model may be composed of an input layer that receives input signals or data from the outside, an output layer that outputs output signals or data corresponding to the input data, and at least one hidden layer located between the input layer and the output layer that receives signals from the input layer, extracts characteristics, and transmits them to the output layer. The output layer receives signals or data from the hidden layer and outputs them to the outside.

[0069] For example, the processor can train artificial intelligence models using TensorFlow and Keras.

[0070] For example, a processor can train an AI model to solve a binary classification problem of positive or negative.

[0071] For example, a processor can train an AI model to solve a binary classification problem using a binary cross entropy loss function.

[0072] Additionally, the processor can add non-linearity using a Leaky ReLU activation function and train the AI ​​model to output a value between 0 and 1 as output data using a Sigmoid function.

[0073] For example, the processor can validate a trained artificial intelligence model. For example, the processor can validate the artificial intelligence model using a first validation data set, a predetermined percentage of which is randomly selected from the entire data set. As another example, the processor can validate the artificial intelligence model using a second validation data set extracted using ligand clustering. However, without limitation, the processor can also validate the artificial intelligence model using both the first validation data set and the second validation data set.

[0074] For example, the processor can verify the true or false classification performance of a learned artificial intelligence model using a predetermined evaluation metric.

[0075] For example, a positive result is when the RMSD between the actual value and the value predicted by the AI ​​model is 2.0. It can mean that the voice has an RMSD of 2.0 This may mean that there is an excess. Here, RMSD (Root Mean Square Deviation) is the average distance between corresponding atoms in the three-dimensional structures of two proteins. A lower value indicates greater structural similarity, while a higher value indicates less structural similarity. In other words, a lower RMSD value may indicate greater structural similarity.

[0076] For example, the processor can display a TP (true positive) if the learned artificial intelligence model accurately predicts positive, a TN (true negative) if it accurately predicts negative, a FP (false positive) if it fails to predict positive, and a FN (false negative) if it fails to predict negative, and perform verification using a predetermined formula.

[0077] As an example, the processor may evaluate the accuracy of an AI model using the first formula. Here, the first formula for calculating accuracy may be "Accuracy = (TP + TN) / (TP + TN + FP + FN)". As another example, the processor may evaluate the accuracy of an AI model using the second formula. Here, the second formula for calculating accuracy may be "Accuracy = (TP) / (TP + FP)".

[0078] Accordingly, the processor can obtain candidate binding poses of proteins and ligands whose structural similarity satisfies a preset value as output data of the artificial intelligence model.

[0079] For example, the processor can obtain candidate binding poses of proteins and ligands as output data of an artificial intelligence model.

[0080]

[0081] FIG. 5 is a diagram illustrating an example of a method for determining a candidate combination pose according to one embodiment.

[0082] Hereinafter, with reference to FIG. 5, an example of a method by which a processor obtains candidate binding poses of a protein and a ligand is described.

[0083] Referring to FIG. 5, the processor can determine a candidate binding pose based on the similarity between the native pose (510) and the binding pose (520) of the protein and the ligand. Here, the native pose may refer to a binding pose that the protein and the ligand naturally have when binding to a specific binding site.

[0084] For example, the processor can generate multiple binding poses (520) of a protein and a ligand. In other words, the processor can generate binding poses (520) of a protein and a ligand from which interaction information, which is input data of an artificial intelligence model, is extracted.

[0085] Additionally, the processor can calculate the similarity between the native pose (510) of the protein and ligand and the generated binding pose (520).

[0086] For example, the processor can calculate the RMSD value of the native pose (510) and the combined pose (520) as described above.

[0087] Additionally, the processor can determine a combined pose (520) in which the similarity between the produced native pose (510) and the combined pose (520) is less than or equal to a preset value as a candidate combined pose.

[0088] For example, the processor has an RMSD value of 2.0 between the produced native pose (510) and the combined pose (520). In the following cases, the binding pose (520) of the protein and ligand can be determined as a candidate binding pose.

[0089] Therefore, the processor can obtain the final binding pose based on the charge of the ligand determined as the candidate binding pose.

[0090] For example, the processor can use the QM / MM values ​​of the candidate binding poses to derive alternative charge values ​​for the ligand.

[0091]

[0092] FIG. 6 is a diagram illustrating an example of a method for calculating QM / MM values ​​of candidate combined poses according to one embodiment.

[0093] Hereinafter, with reference to FIG. 6, an example of a method for a processor to calculate QM / MM values ​​of a candidate combination pose is described.

[0094] First, referring to FIG. 6, the processor can set the QM region (620) with only the ligand.

[0095] For example, the processor can produce a QM / MM value of a candidate binding pose. Here, the QM / MM (Quantum Mechanics / Molecular Mechanics) value may refer to a value produced by calculating a portion of a molecular structure using quantum mechanics (QM) and the remaining portion using molecular mechanics (MM).

[0096] For example, the processor can determine a QM region (620) and an MM region (610) in which to perform QM / MM calculations.

[0097] Specifically, the processor can determine the ligand portion in the candidate binding pose in which the protein and the ligand are bound as the QM region (620) and the other protein portion as the MM region (610). In other words, the processor can determine the ligand portion in the candidate binding pose as the MM region (610). The area within can be determined as the QM area (620) and all other areas can be determined as the MM area (610), but 3 is not limited to this as an example.

[0098] For example, the processor can use the QM / MM values ​​of the generated candidate binding poses to calculate the alternative charge values ​​of the ligand. In other words, the processor can use the QM / MM values ​​to calculate the alternative charge values ​​of the ligand, i.e., (atomic electrostatic potential charge) can be produced.

[0099] Additionally, the processor can change the charge value of the ligand to an alternative charge value.

[0100] For example, the processor can perform docking of a ligand with a protein whose charge values ​​have been changed to alternative charge values ​​to obtain a final binding pose.

[0101] That is, the processor can obtain an optimal final binding pose in which the protein and ligand are bound close to the native pose.

[0102] Additionally, the processor can produce the binding energy of the final binding pose.

[0103] Therefore, the processor can produce optimized binding energies of proteins and ligands, which can be usefully utilized in drug screening during new drug development.

[0104] Meanwhile, the above-described method can be written as a program that can be executed on a computer, and can be implemented on a general-purpose digital computer that runs the program using a computer-readable recording medium. In addition, the structure of the data used in the above-described method can be recorded on a computer-readable recording medium through various means. The computer-readable recording medium includes storage media such as magnetic storage media (e.g., ROM, RAM, USB, floppy disk, hard disk, etc.) and optical reading media (e.g., CD-ROM, DVD, etc.).

[0105] Those skilled in the art will appreciate that the present invention can be implemented in modified forms without departing from the essential characteristics of the above-described invention. Therefore, the disclosed methods should be considered illustrative rather than restrictive. The scope of the claims, not the foregoing description, is defined by the scope of the patent, and should be interpreted to encompass all differences within the scope equivalent thereto.

Claims

1. A step of extracting interaction information between proteins and ligands and inputting it into an artificial intelligence model; A step of obtaining candidate binding poses of the protein and the ligand as output data of the artificial intelligence model; and A step of obtaining a final binding pose based on the charge of the ligand determined as the candidate binding pose; comprising; A method for obtaining binding poses between protein receptors and ligands using artificial intelligence models.

2. In paragraph 1, The steps to input into the above artificial intelligence model are: A step of generating multiple binding poses of the protein and the ligand; A step of extracting the interaction information of each of the generated plurality of combined poses; and A method comprising: a step of inputting the extracted interaction information into the artificial intelligence model as input data of the artificial intelligence model.

3. In paragraph 1, The above artificial intelligence model is a method in which a model is learned using interaction information of each of a plurality of protein receptors and a plurality of ligands as learning data.

4. In paragraph 1, The step of obtaining the above candidate combination pose is: A step of generating a binding pose of the above protein and the above ligand; A step of calculating the similarity between the native pose of the protein and the ligand and the generated binding pose; and A method comprising: a step of determining a combined pose having a similarity lower than or equal to a preset value as the candidate combined pose.

5. In paragraph 1, The steps for obtaining the above final combined pose are: A step of calculating the replacement charge value of the ligand using the QM / MM values of the candidate binding pose; a step of changing the charge value of the ligand to the calculated replacement charge value; and A method comprising: performing docking of the ligand and the protein, the charge value of which has been changed to the alternative charge value, to obtain the final binding pose.

6. In paragraph 1, The above method, A method further comprising: a step of calculating binding energy of the final binding pose.

7. A computer-readable recording medium having recorded thereon a program for executing the method of Article 1 on a computer.

8. Memory in which at least one program is stored; and comprising at least one processor executing at least one program; At least one processor of the above, Extracting interaction information of a protein and a ligand and inputting it into an artificial intelligence model, obtaining candidate binding poses of the protein and the ligand as output data of the artificial intelligence model, and obtaining a final binding pose based on the charge of the ligand bound to the candidate binding pose. A device that obtains binding poses between protein receptors and ligands using an artificial intelligence model.

Images