Three-dimensional structure identification system, method, and program

Abstract

[Problem] To identify two or more three-dimensional structures composed of prescribed atoms by extracting feature quantities that are not affected by positions or rotations of three-dimensional structures such as molecules. [Solution] The three-dimensional structure identification system acquires spatial coordinate data of all atoms constituting each of the three-dimensional structures; calculates, on the basis of the acquired spatial coordinate data, a matrix having distances between all atoms of each of the three-dimensional structures as elements, as an interatomic distance matrix; constructs a proximity score matrix by converting the calculated interatomic distance matrix so as to assign a higher score as the distance decreases; converts the proximity score matrix into pixel values within a specific range; forms an image on the basis of the pixel values; and compares all or part of structures of the two or more three-dimensional structures by analyzing the image.

Description

Three-dimensional structure identification systems, methods, and programs

[0001] This invention relates to a stereostructure identification system, a stereostructure identification method, and a program for identifying two or more stereostructures composed of predetermined atoms.

[0002] For example, in the process of molecular design and molecular synthesis planning, the structures of novel molecules and existing molecules are written out, and the structural similarities between the novel and existing molecules are compared. In particular, in the drug discovery process in the pharmaceutical field, the structure of a candidate drug molecule is compared with the structure of an existing molecule to identify structural similarities.

[0003] Conventionally, one method of representing molecular structures is the SMILES (Simplified Molecular Input Line Entry System) notation, which converts the chemical structure of a molecule into a string of ASCII alphanumeric characters and represents it as a two-dimensional diagram or a three-dimensional model (for example, Non-Patent Documents 1 and 2).

[0004] There is also a notation that describes the coordinates of the atoms constituting a molecule in Euclidean space. The PDB format is widely used as a file format. These PDB files can be obtained from the Protein Data Bank (PDB), and the coordinates are read using software such as RasMol (e.g., Non-Patent Document 3), PyMOL (e.g., Non-Patent Document 4), and VMD (e.g., Non-Patent Document 5), and the molecule is displayed as a three-dimensional model. In addition to the PDB format, other formats such as XYZ format and mmCIF (Macromolecular Crystallographic Information File) format are also commonly used.

[0005] 3. SMILES-A Simplified Chemical Language, [online], [Retrieved November 28, 2024], Internet, <https: / / daylight.com / dayhtml / doc / theory / theory.smiles.html>OpenSMILES specification, [online], [Retrieved November 28, 2024], Internet, <http: / / opensmiles.org / opensmiles. html>Home Page for RasMol and OpenRasMol, [online], [searched November 29, 2024], Internet, <http: / / www.openrasmol.org / >PyMOL 3, [online], [searched November 28, 2024], Internet, <https: / / www.schrodinger.com / platform / products / pymol / >VMD Visual Molecular Dynamics, [online], [searched November 29, 2024], Internet, <https: / / www.ks.uiuc. edu / Research / vmd / >I. D. Brown & D. Altermatt, “Bond-Valence Parameters Obtained from a Systemic Analysis of the Inorganic Crystal Structure Database”, Acta Cryst. 1985, B41, 244-247.

[0006] However, the SMILES notation is specialized in describing the connection information of atoms and bonds within a molecule in two dimensions, and it has limitations in that it is difficult to fully reproduce the spatial arrangement of molecules or stereochemical characteristics (for example, information on chirality and stereoisomers), and cannot directly represent the three-dimensional structure or conformation of molecules.

[0007] Furthermore, while the PDB (Protein Data Bank) notation allows for a detailed description of a molecule's three-dimensional structure using coordinate data, it can result in different representations even for the same molecule if the orientation of the molecular coordinate system or the setting of the origin differs. Therefore, it has been pointed out that comparing molecular structures using PDB notation requires consideration of differences in coordinate systems, which can reduce computational efficiency.

[0008] These constraints often pose a significant obstacle when performing structural analysis and comparison of molecules. Therefore, in a wide range of fields involving molecular design and synthesis, particularly in the pharmaceutical field, there is a need to develop methods for identifying structural similarities or identicalities between different molecules while more accurately reflecting the spatial characteristics and stereochemical information of those molecules.

[0009] Incidentally, the minimum information necessary to define molecular structure is the types of atoms that make up the molecule and their coordinates. Generally, information recorded in PDB format is also based on the types of atoms and their coordinates. X-ray crystallography can obtain highly accurate coordinate information mainly for heavy atoms (e.g., carbon, nitrogen, oxygen, etc.), and this information is an important element in molecular structure analysis.

[0010] Therefore, in order to efficiently compare molecular structure information as image data, the interatomic distance matrix is ​​converted into a "Proximity Score Matrix (PSM)". In this conversion, in order to prevent the loss of information and accurately preserve the three-dimensional structural information of the molecule, it is necessary to not only comprehensively include interatomic distance information but also to appropriately reflect information about the types of atoms.

[0011] In view of these problems, the present invention aims to provide a three-dimensional structure identification system, a three-dimensional structure identification method, and a program that identify two or more three-dimensional structures composed of predetermined atoms by extracting characteristic quantities that are not affected by the position or rotation of the three-dimensional structure of molecules, etc.

[0012] The present invention provides the following solutions.

[0013] According to the invention relating to the first feature, a three-dimensional structure identification system is provided for identifying two or more three-dimensional structures composed of predetermined atoms, comprising: an acquisition unit that acquires spatial coordinate data of all atoms constituting each three-dimensional structure; a calculation unit that calculates an interatomic distance matrix based on the acquired spatial coordinate data, with the distances between all atoms of each three-dimensional structure as elements; a construction unit that constructs a proximity score matrix by transforming the calculated interatomic distance matrix so that a higher score is assigned to closer distances; an image forming unit that converts the proximity score matrix into pixel values ​​within a specific range and forms an image based on the pixel values; and a comparison unit that compares the structures of the two or more three-dimensional structures in whole or in part by analyzing the image.

[0014] According to the invention relating to the first feature, since feature quantities that are not affected by the position or rotation of three-dimensional molecules are extracted, accurate and efficient identification of three-dimensional structures becomes possible at all times. Furthermore, because the three-dimensional structure of molecules can be accurately represented as a matrix, it becomes possible to efficiently compare and generate molecular structures through image recognition.

[0015] The invention relating to the second feature is the invention relating to the first feature, wherein the construction unit constructs a proximity score matrix by converting each element of the interatomic distance matrix into its reciprocal power, thereby providing a three-dimensional structure identification system.

[0016] According to the invention relating to the second feature, by converting each element of the interatomic distance matrix to the reciprocal of its power, interactions between nearby atoms are emphasized, the influence between distant atoms is reduced, and it becomes possible to more accurately represent the physical interactions within the molecular structure.

[0017] The invention relating to the third feature is the invention relating to the first feature, wherein the image forming unit normalizes the proximity score matrix within a specific range and converts it into pixel values ​​within the range of 0 to 255, providing a three-dimensional structure identification system.

[0018] According to the invention relating to the third feature, by normalizing the proximity score matrix within a specific range (0 to 255), the scale of the data is unified, making it easier to compare different molecules and structures. Furthermore, by quantizing the data, it becomes possible to compress the amount of data while performing efficient computational processing.

[0019] The invention relating to the fourth feature is the invention relating to the second feature, wherein the construction unit further constructs N sorted matrices as a reference center proximity score matrix by sequentially targeting each atom (1, 2, 3, ..., N) and rearranging the elements of each row in descending order from the proximity score matrix, and the image forming unit converts the reference center proximity score matrix into pixel values ​​within a specific range and forms an image based on those pixel values, thereby providing a three-dimensional structure identification system.

[0020] According to the invention relating to the fourth feature, a set of matrices is obtained in which each atom is sorted in descending order of its distance score from other atoms when each atom is used as a reference point. By sorting in descending order of distance score, it becomes easier to extract characteristic patterns of molecular structures (e.g., proximity and bonding between specific atoms), and it becomes possible to efficiently search for data that matches specific conditions (e.g., pairs of atoms within a certain distance range).

[0021] The invention relating to the fifth feature is the invention relating to the fourth feature, wherein the construction unit rearranges the elements of each row in descending order by swapping rows and columns so that the atom of interest becomes an element in row 1 and column, thereby providing a three-dimensional structure identification system.

[0022] According to the invention relating to the fifth feature, by rearranging the matrix based on the atom of interest, the characteristics of the molecular structure can be emphasized, improving data consistency and analytical efficiency. Furthermore, by rearranging in descending order, important elements with large distances or scores are clustered at the top of the row, while noise and unnecessary elements with little influence are pushed to the back, thereby minimizing the impact of unnecessary data.

[0023] The invention relating to the fifth feature is the invention relating to the third feature, wherein the image forming unit normalizes the reference center proximity score matrix within a specific range and converts it into pixel values ​​within the range of 0 to 255, thereby providing a three-dimensional structure identification system.

[0024] According to the invention relating to the fifth feature, by normalizing the reference center proximity score matrix within a specific range (0 to 255), the scale of the data is unified, making it easier to compare different molecules and structures. Furthermore, by quantizing the data, it becomes possible to compress the amount of data while performing efficient computational processing.

[0025] The invention relating to the sixth feature is the invention relating to the fourth feature, wherein the image forming unit normalizes the reference center proximity score matrix within a specific range and converts it into pixel values ​​within the range of 0 to 255, thereby providing a three-dimensional structure identification system.

[0026] According to the invention relating to the sixth feature, by normalizing the reference center proximity score matrix within a specific range (0 to 255), the scale of the data is unified, making it easier to compare different molecules and structures. Furthermore, by quantizing the data, it becomes possible to compress the amount of data while performing efficient computational processing.

[0027] The invention relating to the seventh feature is the invention relating to the first feature, wherein the construction unit associates each element of the proximity score matrix with atomic identification information based on atomic number and molecular characteristic information including bond valency, polarity, bond orientation and interatomic angle, or one or more combinations thereof, and the image forming unit assigns the pixel values ​​to a first color channel, assigns the atomic identification information to a second color channel, assigns the molecular characteristic information to a third color channel, and forms the image based on the first color channel, the second color channel and the third color channel, thereby providing a three-dimensional structure identification system.

[0028] According to the invention relating to the seventh feature, by providing information that constitutes the elements of color to each element of the proximity score matrix, not only the distance between atoms but also the type of atom and molecular properties can be represented in the image. This makes it easier to visually compare the structural similarities between molecules more accurately, and enables similarity searches using image processing technology. Furthermore, because interatomic distance, atomic identification information, and molecular property information are simultaneously stored, it becomes possible to more accurately save the three-dimensional structure of molecules in PDB file data. In addition, since molecular similarity searches can be performed using existing image comparison algorithms, it becomes possible to improve the efficiency of searching for new compounds and screening drug candidates.

[0029] Although this invention falls under the category of computer systems, it exhibits similar functions and effects in other categories such as methods and programs, depending on the category.

[0030] According to the present invention, it is possible to provide a three-dimensional structure identification system, a three-dimensional structure identification method, and a program that can identify two or more three-dimensional structures composed of predetermined atoms by extracting characteristic quantities that are not affected by the position or rotation of the three-dimensional structure of molecules, etc.

[0031] This figure illustrates the outline of the three-dimensional structure identification system 1, which is the first embodiment of the present invention. This is a configuration diagram of the three-dimensional structure identification system 1 of this embodiment. This is a flowchart of the three-dimensional structure identification process performed by the three-dimensional structure identification system 1 of this embodiment. This figure shows an example of PDB file data in the three-dimensional structure identification system 1 of this embodiment. This figure shows the HETATM record of heavy atoms extracted from the PDB file data shown in Figure 4. This figure illustrates the interatomic distance matrix 200 in the three-dimensional structure identification system 1 of this embodiment. This figure illustrates the proximity score matrix 300 in the three-dimensional structure identification system 1 of this embodiment when the scaling index k = -2.15. This figure illustrates the molecular structure information represented by the proximity score matrix 300 in the three-dimensional structure identification system 1 of this embodiment when the scaling index k = -2.0. This figure illustrates the molecular structure information represented by the proximity score matrix 300 in the three-dimensional structure identification system 1 of this embodiment when the scaling index k = -1.95. This figure shows an example of a proximity score image 500 of molecular structures to be compared in the three-dimensional structure identification system 1 of this embodiment. This figure shows a proximity score image 500 corresponding to the converted pixel values ​​of the proximity score matrix 300 in the three-dimensional structure identification system 1 of this embodiment. This figure is for explaining the overview of the three-dimensional structure identification system 1, which is a second embodiment of the present invention. This is a flowchart of the three-dimensional structure identification process performed by the three-dimensional structure identification system 1 of this embodiment. This figure is for explaining the construction process of the reference center proximity score matrix 310 in the three-dimensional structure identification system 1 of this embodiment. This figure is for explaining the construction process of the reference center proximity score matrix 310 in the three-dimensional structure identification system 1 of this embodiment. This figure is for explaining the construction process of the reference center proximity score matrix 310 in the three-dimensional structure identification system 1 of this embodiment. This figure is for explaining the construction process of the reference center proximity score matrix 310 in the three-dimensional structure identification system 1 of this embodiment. This figure shows a reference center proximity score image 510 corresponding to the converted pixel values ​​of the reference center proximity score matrix 310 in the three-dimensional structure identification system 1 of this embodiment.

[0032] Hereinafter, the best mode for carrying out the present invention will be described with reference to the drawings. Note that this is merely an example, and the technical scope of the present invention is not limited thereto.

[0033] [First Embodiment] [Outline of the Three-dimensional Structure Identification System 1] The outline of the three-dimensional structure identification system 1 according to the first embodiment of the present invention will be described based on FIG. 1. FIG. 1 is a diagram for explaining the outline of the three-dimensional structure identification system 1 according to the first embodiment of the present invention. The three-dimensional structure identification system 1 is composed of a computer 2 and is a computer system for identifying two or more three-dimensional objects.

[0034] The computer 2 of the three-dimensional structure identification system 1 is, for example, a computer such as a desktop personal computer, a notebook personal computer, or a server, a mobile terminal such as a smartphone or a tablet terminal, or a wearable terminal such as a head-mounted display like smart glasses or a smartwatch.

[0035] [[ID=ID=11]] Further, the computer 2 of the three-dimensional structure identification system 1 may be realized by, for example, one terminal device, may be realized by a plurality of terminal devices, or may be realized by a virtual device such as a cloud computer.

[0036] Further, the three-dimensional structure identification system 1 may be composed of the above-described terminal devices instead of the computer 2.

[0037] The computer 2 of the three-dimensional structure identification system 1 is connected to the above-described terminal devices, other terminals and devices, etc. via a public communication network or the like so as to enable data communication, and performs transmission and reception of necessary data and information.

[0038] Next, an overview of the processing executed by the three-dimensional structure identification system 1 will be described. First, the computer 2 of the three-dimensional structure identification system 1 acquires spatial coordinate data of all atoms constituting each three-dimensional object (step S1). Specifically, the computer 2 extracts and acquires spatial coordinate data 100 of heavy atoms constituting a molecule from PDB file data of two or more molecules to be identified. In this specification, heavy atoms refer to atoms other than hydrogen. Regardless of the method of obtaining the PDB file data, for example, if it is an actual molecular structure, it may be downloaded from an external device such as the RCSB PDB website (https: / / www.rcsb.org / ), and if it is a new molecular structure, it may be RDB data obtained after the generation of the new molecule. Also, the PDB file data may be stored in advance in the computer 2.

[0039] Next, the computer 2 calculates, as an interatomic distance matrix 200, a matrix having, as elements, the distances between all pairs of atoms of each three-dimensional object based on the acquired spatial coordinate data 100 (step S2). Specifically, the computer 2 calculates, as an interatomic distance matrix 200, a matrix represented in matrix form with the Euclidean distances between heavy atoms of each molecule as elements based on the spatial coordinate data 100 represented by coordinates in the three-dimensional space acquired in step S1 described above.

[0040] Next, the computer 2 constructs a proximity score matrix (PSM) 300 by converting the calculated interatomic distance matrix 200 so as to assign a higher score to closer distances (step S3). Specifically, the computer 2 calculates a score for the distance by converting each element of the interatomic distance matrix 200 calculated in step S2 described above into the reciprocal of the power, and constructs the proximity score matrix 300. Therefore, each element of the proximity score matrix 300 represents the proximity score between the corresponding atoms, and the closer the interatomic distance, the larger the element of the proximity score matrix 300.

[0041] In step S3 described above, the computer 2 may associate each element of the proximity score matrix 300 with atomic identification information based on atomic number and molecular property information including bond valence, polarity, bond orientation and interatomic angle, or one or more combinations thereof. Specifically, the computer 2 may associate each element of the proximity score matrix 300 with a score for identifying combinations of atoms based on atomic number and molecular property information including bond valence, polarity, bond orientation and interatomic angle, or any one or more combinations thereof.

[0042] Next, computer 2 converts the proximity score matrix 300 into pixel values ​​within a specific range and forms an image based on these pixel values ​​(step S4). Specifically, computer 2 converts the values ​​of each element of the proximity score matrix 300 constructed in step S3 above into appropriate ranges for the pixel values ​​of the image, and forms a proximity score image 500 based on the converted pixel values. Therefore, the grayscale of the formed image is determined based on the corresponding matrix elements.

[0043] In step S4 described above, the computer 2 may assign the converted pixel values ​​to a first color channel, assign the atomic identification information associated with each element of the proximity score matrix 300 in step S3 to a second color channel, assign the molecular characteristic information associated with each element of the proximity score matrix 300 in step S3 to a third color channel, and form an image based on the first color channel, the second color channel, and the third color channel. Specifically, the computer 2 may, for example, assign pixel values ​​to the R channel, atomic identification information to the G channel, and molecular characteristic information to the B channel based on an RGB model, and generate a color image based on these channels. The color model may be a model other than the RGB model, such as a CMYK model or an HSV model, and is not particularly limited.

[0044] Next, computer 2 compares the structures of two or more three-dimensional objects, in whole or in part, by analyzing the images (step S5). Specifically, computer 2 searches for matching or similar wholes or parts of the two or more molecular structures represented in the proximity score image 500 by comparing the whole or in part of the proximity score image 500 obtained in step S4 on a pixel-by-pixel basis.

[0045] The above is an overview of the processes performed by the three-dimensional structure identification system 1.

[0046] [System Configuration of Three-Dimensional Structure Identification System 1] Based on Figure 2, the system configuration of the three-dimensional structure identification system 1 of this embodiment will be described. The three-dimensional structure identification system 1 consists of a computer 2 and is a computer system for identifying two or more three-dimensional objects.

[0047] The 3D structure identification system 1 may also include other terminals and devices. For example, each user may use a separate computer 2, in which case the 3D structure identification system 1 will perform the processes described later using one or more combinations of computer 2 and the other included terminals and devices.

[0048] Computer 2 may be implemented as, for example, a single terminal device, multiple terminal devices, or a virtual device such as a cloud computer.

[0049] Computer 2 includes, for example, computers such as desktop PCs, laptops, and servers; mobile devices such as smartphones and tablet devices; and wearable devices such as head-mounted displays like smart glasses and smartwatches.

[0050] Computer 2 includes a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), RAM (Random Access Memory), ROM (Read Only Memory), and other components as its control unit.

[0051] Computer 2 includes data storage as a memory unit, such as a hard disk, semiconductor memory, recording medium, or memory card. The data storage may be internal data storage and / or external data storage. The data may be stored in a cloud service or database.

[0052] Computer 2 includes a communication unit equipped with devices to enable communication with other terminals and equipment. The communication method may be wireless or wired.

[0053] Computer 2 shall have the necessary functions for operating Computer 2 as an input unit. Examples of input methods include a liquid crystal display with touch panel functionality, a keyboard, a mouse, a pen tablet, hardware buttons on the device, a microphone for voice recognition, etc. The present invention is not particularly limited by the input method.

[0054] Computer 2, as an output unit, shall be equipped with the functions necessary for a user of the 3D structure identification system 1 to operate Computer 2. Examples of output methods include display on a liquid crystal display, PC display, or projector, as well as audio output. The present invention is not particularly limited in function by the output method.

[0055] The control unit works in cooperation with the processing unit to realize the calculation unit 22 and the construction unit 23. The control unit also works in cooperation with the processing unit, the communication unit and the storage unit to realize the acquisition unit 21. Furthermore, the control unit works in cooperation with the processing unit and the output unit to realize the image forming unit 24 and the comparison unit 25.

[0056] The above describes the system configuration of the three-dimensional structure identification system 1.

[0057] [3D Structure Identification Process] Based on Figure 3, the 3D structure identification process performed by computer 2 will be described. Figure 3 is a flowchart of the 3D structure identification process performed by computer 2. As shown in Figure 3, the 3D structure identification process consists of steps S11 to S15 and is the process performed in steps S1 to S5 described above.

[0058] First, the acquisition unit 21 of the computer 2 acquires spatial coordinate data of all atoms constituting each stereochemistry (step S11). Specifically, the acquisition unit 21 extracts and acquires spatial coordinate data 100 of heavy atoms constituting a molecule from PDB file data of two or more molecules to be identified. The PDB file data is uniquely identified by a PDB ID that indicates specific molecular structure data such as proteins, DNA, RNA, and ligands. The method of obtaining the PDB file data is not limited; for example, if it is an existing molecular structure, it may be downloaded to the computer 2 via the communication unit from an external device such as the RCSB PDB website (https: / / www.rcsb.org / ), or if it is a new molecular structure, it may be PDB file data obtained after the new molecule has been generated. The PDB file data may also be stored in advance in the storage unit of the computer 2. A detailed explanation of the data items indicated by the PDB file data will be omitted.

[0059] Figure 4 shows an example of PDB file data, and Figure 5 shows the HETATM record of heavy atoms extracted from the PDB file data shown in Figure 4. The acquisition unit 21 extracts the HETATM record of heavy atoms from the PDB file data shown in Figure 4, as shown in Figure 5, and acquires the X, Y, Z coordinate data of the atoms contained in the HETATM record in three-dimensional space as spatial coordinate data 100.

[0060] Next, the calculation unit 22 of computer 2 calculates an interatomic distance matrix 200 based on the acquired spatial coordinate data 100, with the distances between all atoms in each solid as its elements (step S12). Specifically, the calculation unit 22 calculates an interatomic distance matrix 200 based on the spatial coordinate data 100, which is represented by coordinates in three-dimensional space acquired in step S11 above, with the Euclidean distances between heavy atoms of each molecule as its elements, expressed in matrix form.

[0061] Figure 6 is a diagram illustrating the interatomic distance matrix 200. For example, heavy atom i is coordinate (x ｉ , y ｉ , zｉ is located at (i), and the heavy atom j has coordinates (x ｊ , y ｊ , z ｊ ), the Euclidean distance r between the heavy atom i and the heavy atom j is calculated by the following formula. ｉｊ That is, it is calculated by the following formula.

[0062] Therefore, as shown in FIG. 6, for each heavy atom a, b, c, d,... constituting one molecule, the interatomic distance r obtained by the formula shown by Equation 1 above ａｂ , r ａｃ , r ａｄ ,... is used as elements to calculate the interatomic distance matrix 200.

[0063] Next, the construction unit 23 of the computer 2 constructs the proximity score matrix 300 by converting the calculated interatomic distance matrix 200 so as to assign a higher score to a closer distance (step S13). Specifically, the construction unit 23 calculates the score for the distance by converting each element of the interatomic distance matrix 200 calculated in step S12 above into its reciprocal, and constructs the proximity score matrix 300. Therefore, each element of the proximity score matrix 300 represents the proximity score between the corresponding atoms, and the closer the interatomic distance, the larger the element of the proximity score matrix 300.

[0064] FIG. 7 is a diagram for explaining the proximity score matrix 300. For example, by raising each element of the interatomic distance matrix 200 calculated in step S12 above, the interatomic distances r ａｂ , r ａｃ , r ａｄ ,... to the power of the scaling exponent k (k < 0), the interatomic distances r ａｂ , r ａｃ , r ａｄ ,... are converted to their reciprocals, and scores shown by the following formulas 2 to 4 etc. are calculated.

[0065] Therefore, for the interatomic distance matrix 200 as shown in FIG. 6, as shown in FIG. 7, a matrix with scores shown by Formulas 2 to 4 etc. as elements is calculated as the proximity score matrix 300.

[0066] The scaling exponent k can be any value as long as k < 0, but for ease of calculation, when the values ​​of the matrix elements are rounded to one decimal place, it is preferable that -2.15 ≤ k ≤ -1.95.

[0067] Figure 8 illustrates the molecular structure information represented by the proximity score matrix 300 when the scaling index k = -2.15. For example, for molecular structure A, the proximity score matrix 300 when K = -2.15 represents correlations up to two bonds, but does not represent distant correlations and, as shown in Figure 8, does not include steric information.

[0068] Figure 9 is a diagram illustrating the molecular structure information represented by the proximity score matrix 300 when the scaling index k = -2.0. Compared to the interatomic distance matrix 200 of molecular structure A, the structure shown by the proximity score matrix 300 when K = -2.0 yields a proximity score shown in bold, which indicates that long-range correlations are displayed and include near-range stereochemistry. However, because there is a threshold near the distance between the substituents at positions 1 and 4 within the molecule, it is difficult to distinguish the relative stereochemistry of these substituents.

[0069] Figure 10 is a diagram illustrating the molecular structure information represented by the proximity score matrix 300 when the scaling index k = -1.95. Compared to the interatomic distance matrix 200 of molecular structure A, the structure shown by the proximity score matrix 300 when K = -1.95 yields a proximity score indicated by the underline, which includes interpretable near-field stereochemistry. This allows us to distinguish whether the stereochemistry of the substituents at positions 1 and 4 within the molecule is 1,4-trans conformation or 1,4-gauche conformation.

[0070] In step S13 described above, the construction unit 23 of the computer 2 may associate each element of the proximity score matrix 300 with atomic identification information based on atomic number and molecular property information including bond valence, polarity, bond orientation and interatomic angle, or one or more combinations thereof. Specifically, the construction unit 23 may associate each element of the proximity score matrix 300 with a score for identifying combinations of atoms based on atomic number and molecular property information including bond valence, polarity, bond orientation and interatomic angle, or any one or more combinations thereof.

[0071] Atomic identification information may be, for example, obtained by assigning a three-digit atomic number to each combination of atoms that make up each element of the proximity score matrix 300, and then combining the two atomic numbers to obtain an identification number as a score. For example, the score between carbon and carbon would be "006006", the score between carbon and nitrogen would be "006007", the score between nitrogen and carbon would be "007006", and the score between carbon and oxygen would be "006008".

[0072] The bond valency, polarity, bond orientation, and interatomic angles that may be included in the molecular property information may be indirectly estimated or calculated from the PDB file data using a specific application. In particular, the bond valency may be derived from the interatomic distance matrix 200 calculated in step S12 using the Bond Valency Method based on Non-Patent Literature 6. The parameters that may be included in the molecular property information are not limited to bond valency, polarity, bond orientation, and interatomic angles. Furthermore, there are no particular limitations on the methods for estimating and calculating these parameters.

[0073] Next, the image forming unit 24 of computer 2 converts the proximity score matrix 300 into pixel values ​​within a specific range and forms an image based on these pixel values ​​(step S14). Specifically, the image forming unit 24 converts the values ​​of each element of the proximity score matrix 300 constructed in step S13 above into appropriate ranges for image pixel values, and forms a proximity score image 500 based on the converted pixel values. Therefore, the grayscale of the formed image is determined based on the elements of the corresponding matrix.

[0074] The appropriate range for the pixel values ​​mentioned above may be a numerical range represented by 8 × n (n: natural number) bits, but considering the memory efficiency of computer 2, the range of 0 to 255 is preferred. Figure 11 shows a proximity score image 500 corresponding to the converted pixel values ​​of a particular proximity score matrix 300. In Figure 11, the proximity score image 500 corresponding to the converted pixel values ​​of molecular structure B is shown, and in the proximity score matrix 300, elements with small scores are represented by "dark colors (e.g., black)", and elements with large scores are represented by "light colors (e.g., white)".

[0075] In step S14 described above, the image forming unit 24 of the computer 2 may assign the converted pixel values ​​to a first color channel, assign the atomic identification information associated with each element of the proximity score matrix 300 in step S13 to a second color channel, assign the molecular characteristic information associated with each element of the proximity score matrix 300 in step S13 to a third color channel, and form a proximity score image 500 based on the first color channel, the second color channel, and the third color channel. Specifically, the computer 2 may, for example, assign pixel values ​​to the R channel, atomic identification information to the G channel, and molecular characteristic information to the B channel based on an RGB model, and form a proximity score image 500 of a color image based on these channels. The assignment of information to each color channel is arbitrary. The color model for each color channel may be a model other than the RGB model, such as a CMYK model or an HSV model, and is not particularly limited.

[0076] Next, the comparison unit 25 of the computer 2 compares the structures of two or more three-dimensional objects, in whole or in part, by analyzing the images (step S15). Specifically, the comparison unit 25 searches for matching or similar wholes or parts in the two or more molecular structures represented in the proximity score image 500 by comparing the whole or in part of the proximity score image 500 obtained in step S14 using difference images on a pixel-by-pixel basis.

[0077] Figure 12 shows an example of proximity score images 500 of molecular structures to be compared. For example, as shown in Figure 12, by comparing the proximity score images 500 of molecular structure C and molecular structure D, the portion within the enclosed frame in the proximity score image 500 of molecular structure D is found to correspond to the entire pixel-level portion of the proximity score image 500 of molecular structure C.

[0078] The proximity score matrix 300 constructed in step S13 described above may be used in conjunction with a large-scale language model (LLM) to construct an AI system. Specifically, the constructed proximity score matrix 300 may be vectorized and used to train the LLM. Alternatively, important features may be extracted from the proximity score matrix 300 using principal component analysis (PCA) or a convolutional neural network (CNN) and used to train the LLM. The training process may be performed at any point after step S13 described above.

[0079] In step S14 described above, if a proximity score image 500 is created based on the color channels, the proximity score images 500 may be compared using existing image comparison algorithms other than the pixel-based image comparison algorithm, such as comparisons that consider structural similarity, such as SSIM (Structural Similarity Index), or feature-based comparisons, such as histogram comparisons. The image comparison algorithm used is not particularly limited.

[0080] The above describes the three-dimensional structure identification process.

[0081] Therefore, the three-dimensional structure identification system 1 extracts features that are not affected by the position or rotation of the three-dimensional structure of molecules, enabling consistently accurate and efficient identification of three-dimensional structures. Furthermore, because the three-dimensional structure of a molecule can be accurately represented as a matrix, it becomes possible to efficiently compare and generate molecular structures through image recognition.

[0082] According to the three-dimensional structure identification system 1, by further converting each element of the interatomic distance matrix to the reciprocal of its power, interactions between nearby atoms are emphasized, the influence of distant atoms is reduced, and it becomes possible to more accurately represent the physical interactions within the molecular structure.

[0083] According to the 3D structure identification system 1, further normalizing the proximity score matrix within a specific range (0 to 255) unifies the data scale, making it easier to compare different molecules and structures. Furthermore, quantization allows for efficient computation while compressing the data volume.

[0084] [Second Embodiment] [Overview of the Three-Dimensional Structure Identification System 1] An overview of the three-dimensional structure identification system 1, which is a second embodiment of the present invention, will be described based on Figure 13. Figure 13 is a diagram illustrating the overview of the three-dimensional structure identification system 1, which is a second embodiment. The three-dimensional structure identification system 1 consists of a computer 2 and is a computer system for identifying two or more three-dimensional objects. The same reference numerals are used for functions and components that are the same as those in the first embodiment described above, and their descriptions are omitted. The difference from the first embodiment described above is that after step S3, a reference center proximity score matrix 310 is constructed (step S31), an image is formed based on the reference center proximity score matrix 310 (step S41), and the structure of the whole or a part of the two or more three-dimensional objects is compared in step S51 by analyzing the image (step S51).

[0085] The computer 2 of the three-dimensional structure identification system 1 is the same as that of the first embodiment described above, so its description will be omitted.

[0086] Next, an overview of the processes performed by the three-dimensional structure identification system 1 will be described. First, the computer 2 of the three-dimensional structure identification system 1 acquires spatial coordinate data of all atoms constituting each three-dimensional structure (step S1). This step is the same as in the first embodiment described above, so its explanation will be omitted.

[0087] Next, the computer 2 calculates an interatomic distance matrix 200 based on the acquired spatial coordinate data 100, with the distances between all atoms in each solid as its elements (step S2). This step is the same as in the first embodiment described above, so its explanation is omitted.

[0088] Next, computer 2 constructs a proximity score matrix 300 by transforming the calculated interatomic distance matrix 200 so that closer distances are assigned higher scores (step S3). This step is the same as in the first embodiment described above, so its explanation is omitted.

[0089] Next, computer 2 takes each atom (1, 2, 3, ..., N) in turn from the proximity score matrix 300 and sorts the elements of each row in descending order to construct N sorted matrices, which are then constructed as the reference center proximity score matrix 310 (step S31). Specifically, computer 2 takes each atom (1, 2, 3, ..., N) in turn from the proximity score matrix 300 constructed in step S3 above and sorts the elements of each row in descending order to construct N sorted matrices, which are then constructed as the reference center proximity score matrix 310.

[0090] Next, computer 2 converts the reference center proximity score matrix 310 into pixel values ​​within a specific range and forms an image based on these pixel values ​​(step S41). Specifically, computer 2 converts the values ​​of each element of the reference center proximity score matrix 310 constructed in step S31 above into appropriate ranges for the pixel values ​​of the image, and forms a reference center proximity score image 510 using the converted reference center proximity scores as pixel values. Therefore, the grayscale of the formed image is determined based on the corresponding matrix elements.

[0091] Next, computer 2 compares the structures of two or more three-dimensional objects, in whole or in part, by analyzing the images (step S51). Specifically, computer 2 searches for matching or similar wholes or parts of two or more molecular structures represented in the reference center proximity score image 510 by comparing the whole or in part of the reference center proximity score image 510 obtained in step S41 on a pixel-by-pixel basis.

[0092] In this embodiment, the processes of steps S4-S5 of the first embodiment described above may be performed, and the timing of performing these steps may be any timing after step S3.

[0093] The above is an overview of the processes performed by the three-dimensional structure identification system 1.

[0094] [System Configuration of the Three-Dimensional Structure Identification System 1] The system configuration of the three-dimensional structure identification system 1, which is the second embodiment of the present invention, is the same as that of the first embodiment described above, so its description is omitted.

[0095] [3D Structure Identification Process] Based on Figure 14, the 3D structure identification process performed by computer 2 will be described. Figure 3 is a flowchart of the 3D structure identification process performed by computer 2. As shown in Figure 14, the 3D structure identification process consists of steps S11 to S151 and is the process performed in steps S1 to S51 described above.

[0096] First, the acquisition unit 21 of the computer 2, the computer 2 of the three-dimensional structure identification system 1, acquires spatial coordinate data of all atoms constituting each three-dimensional structure (step S11). This step is the same as in the first embodiment described above, so its explanation is omitted.

[0097] Next, the calculation unit 22 of the computer 2 calculates an interatomic distance matrix 200 based on the acquired spatial coordinate data 100, with the distances between all atoms in each solid as its elements (step S12). This step is the same as in the first embodiment described above, so its explanation is omitted.

[0098] Next, the construction unit 23 of the computer 2 constructs a proximity score matrix 300 by transforming the calculated interatomic distance matrix 200 so that closer distances are assigned higher scores (step S13). This step is the same as that of the first embodiment described above, so its explanation is omitted.

[0099] Next, the construction unit 23 of computer 2 takes each atom (1, 2, 3, ..., N) in order from the proximity score matrix 300 and further constructs N sorted matrices by rearranging the elements of each row in descending order, which are then constructed as the reference center proximity score matrix 310 (step S131). Specifically, for example, the construction unit 23 moves the i-th row and the i-th column of the proximity score matrix 300, which consists of N rows and N columns (N: natural number), to the first row and the i-th column to the first column. Next, it removes the column where the value in the first row is 0, sorts the second column and subsequent columns in descending order of the values ​​in the first row, removes the column where the value in the first row is 0, and sorts the second column and subsequent columns in descending order of the values ​​in the first row, thereby constructing the reference center proximity score matrix 310. By repeating this process from i=1 to N, N reference center proximity score matrices 310 are constructed.

[0100] Figures 15 to 18 illustrate the process of constructing the reference center proximity score matrix 310, which consists of 13 rows and 13 columns. In Figure 15, as shown in bold, the 13th row is moved to the 1st row, and in Figure 16, as shown in bold, the 13th column is moved to the 1st column. Next, in Figure 17, as shown in bold, the construction unit 23 deletes the column where the value in the 1st row is 0, and sorts the columns from the 2nd column onward in descending order of the values ​​in the 1st row. Furthermore, in Figure 18, as shown in bold, the column where the value in the 1st row is 0 is deleted, and the columns from the 2nd column onward are sorted in descending order of the values ​​in the 1st row. As a result, the reference center proximity score matrix 310 for the case i = 13 is constructed. For the sake of explanation, i = 13 was used as an example, but in reality, the processing for i = 1 to 12 is performed sequentially before the processing for i = 13, and the reference center proximity score matrix 310 for i = 1 to 12 is constructed accordingly.

[0101] Next, the image forming unit 24 of computer 2 converts the reference center proximity score matrix 310 into pixel values ​​within a specific range and forms an image based on these pixel values ​​(step S141). Specifically, the image forming unit 24 converts the values ​​of each element of the N reference center proximity score matrices 310 constructed in step S131 above into appropriate ranges for the pixel values ​​of the image, and forms a reference center proximity score image 510 using the converted reference center proximity scores as pixel values. Therefore, the grayscale of the formed image is determined based on the elements of the corresponding matrices.

[0102] The appropriate range for the pixel values ​​mentioned above may be a numerical range represented by 8 × n (n: natural number) bits, but considering the memory efficiency of computer 2, the range of 0-255 is preferred. Figure 19 shows a reference center proximity score image 510 corresponding to the converted pixel values ​​of the reference center proximity score matrix 310. In Figure 19, proximity score images 500 are shown corresponding to the converted pixel values ​​of N (N=13) reference center proximity score matrices 310 of molecular structure B, where elements with small scores are represented by "dark colors (e.g., black)" and elements with large scores are represented by "light colors (e.g., white)".

[0103] In step S141 described above, the image forming unit 24 of the computer 2 may assign the converted pixel values ​​to a first color channel, assign the atomic identification information associated with each element of the reference center proximity score matrix 310 to a second color channel, assign the molecular characteristic information associated with each element of the reference center proximity score matrix 310 to a third color channel, and form a reference center proximity score image 510 based on the first color channel, the second color channel, and the third color channel. Specifically, the computer 2 may, for example, assign pixel values ​​to the R channel, atomic identification information to the G channel, and molecular characteristic information to the B channel based on an RGB model, and form a reference center proximity score image 510 of a color image based on these channels. The assignment of information to each color channel is arbitrary. The color model for each color channel may be a model other than the RGB model, such as a CMYK model or an HSV model, and is not particularly limited.

[0104] Next, the comparison unit 25 of the computer 2 compares the structures of two or more three-dimensional objects, in whole or in part, by analyzing the images (step S151). Specifically, the comparison unit 25 searches for matching or similar wholes or parts of two or more molecular structures represented by the reference center proximity score images 510 by comparing all or part of the N (N: natural number) reference center proximity score images 510 obtained in step S141 above on a pixel-by-pixel basis using difference images.

[0105] In this embodiment, the processes of steps S14-S15 of the first embodiment described above may be performed, and the timing of performing these steps may be any timing after step S13.

[0106] The proximity score matrix 300 constructed in step S131 described above may be used in conjunction with a large-scale language model (LLM) to construct an AI system. Specifically, the constructed proximity score matrix 300 may be vectorized and used to train the LLM. Alternatively, important features may be extracted from the proximity score matrix 300 using principal component analysis (PCA) or a convolutional neural network (CNN) and used to train the LLM. This training process may be performed at any point after step S13 described above.

[0107] In step S141 described above, if a reference center proximity score image 510 is created based on the color channels, the reference center proximity score image 510 may be compared using existing image comparison algorithms other than a pixel-based image comparison algorithm, such as a comparison that considers structural similarity, such as SSIM (Structural Similarity Index), or a feature-based comparison, such as a histogram comparison. The image comparison algorithm used is not particularly limited.

[0108] Furthermore, regarding the reference center proximity score matrix 310 constructed in step S131 described above, the reference center proximity score matrix 310 for molecular structures with a known three-dimensional structure may be stored in the storage unit of computer 2 by constructing an Atomic Configuration of real molecules (ACRM) database. This storage process of the reference center proximity score matrix 310 can be performed at any point after step S131 described above. Therefore, for novel molecular structures, it is possible to optimize them as possible molecular structures by searching the ACRM database.

[0109] The above describes the three-dimensional structure identification process.

[0110] Therefore, the three-dimensional structure identification system 1 provides a set of matrices sorted in descending order of distance score between each atom and other atoms, with each atom acting as a reference point. By sorting in descending order of distance score, it becomes easier to extract characteristic patterns of molecular structures (e.g., proximity and bonding between specific atoms), and it becomes possible to efficiently search for data that matches specific conditions (e.g., pairs of atoms within a certain distance range).

[0111] According to the three-dimensional structure identification system 1, by rearranging the matrix based on the atom of interest, the characteristics of the molecular structure can be emphasized, improving data consistency and analysis efficiency. Furthermore, by rearranging in descending order, important elements with large distances or scores are clustered at the top of the row, while unnecessary elements with little noise or influence are pushed to the back, thus minimizing the impact of unnecessary data.

[0112] According to the three-dimensional structure identification system 1, further normalizing the reference center proximity score matrix within a specific range (0 to 255) unifies the data scale, making it easier to compare different molecules and structures. Furthermore, quantization allows for efficient computation while compressing the data volume.

[0113] According to the 3D structure identification system 1, by providing information that constitutes the color of each element of the proximity score matrix, not only the distance between atoms but also the type of atom and molecular properties can be represented in the image. This makes it easier to visually compare the structural similarities between molecules more accurately, and enables similarity searches using image processing technology. Furthermore, because interatomic distances, atomic identification information, and molecular property information are simultaneously preserved, it becomes possible to more accurately store the three-dimensional structure of molecules in PDB file data. In addition, since molecular similarity searches can be performed using existing image comparison algorithms, it becomes possible to improve the efficiency of discovering new compounds and screening drug candidates.

[0114] According to the three-dimensional structure identification system 1, further normalizing the reference center proximity score matrix 310 within a specific range (0 to 255) unifies the scale of the data, making it easier to compare different molecules and structures. Furthermore, quantization allows for efficient computation while compressing the amount of data.

[0115] The means and functions described above are realized by a computer (including the CPU, information processing unit, and various terminals) reading and executing a predetermined program. The program is provided, for example, from one or more computers via a network (cloud service, SaaS: Software as a Service). Alternatively, the program may be provided, for example, recorded on a computer-readable recording medium. In this case, the computer reads the program from the recording medium, transfers it to an internal or external recording device, records it, and executes it. Alternatively, the program may be pre-recorded on a recording device (recording medium) such as a magnetic disk, optical disk, or magneto-optical disk, and provided to the computer from that recording device via a communication line.

[0116] Although embodiments of the present invention have been described above, the present invention is not limited to these embodiments. Furthermore, the effects described in the embodiments of the present invention are merely a list of the most preferred effects arising from the present invention, and the effects of the present invention are not limited to those described in the embodiments.

[0117] 1. Three-dimensional structure identification system, 2. Computer, 21. Acquisition unit, 22. Calculation unit, 23. Construction unit, 24. Image formation unit, 25. Comparison unit, 100. Spatial coordinate data, 200. Interatomic distance matrix, 300. Proximity score matrix, 310. Reference center proximity score matrix, 500. Proximity score image, 510. Reference center proximity score image

Claims

1. A three-dimensional structure identification system for identifying two or more three-dimensional structures composed of predetermined atoms, comprising: an acquisition unit that acquires spatial coordinate data of all atoms constituting each three-dimensional structure; a calculation unit that calculates an interatomic distance matrix based on the acquired spatial coordinate data, with the distances between all atoms of each three-dimensional structure as elements; a construction unit that constructs a proximity score matrix by transforming the calculated interatomic distance matrix so that a higher score is assigned to closer distances; an image forming unit that converts the proximity score matrix into pixel values ​​within a specific range and forms an image based on the pixel values; and a comparison unit that compares the structures of the two or more three-dimensional structures in whole or in part by analyzing the image.

2. The stereochemical structure identification system according to claim 1, wherein the construction unit constructs a proximity score matrix by converting each element of the interatomic distance matrix into its reciprocal power.

3. The stereoscopic structure identification system according to claim 1, wherein the image forming unit normalizes the proximity score matrix within a specific range and converts it into pixel values ​​within the range of 0 to 255.

4. The construction unit further constructs N sorted matrices as a reference center proximity score matrix by sequentially targeting each atom (1, 2, 3, ..., N) and rearranging the elements of each row in descending order from the proximity score matrix; and the image forming unit converts the reference center proximity score matrix into pixel values ​​within a specific range and forms an image based on the pixel values, the three-dimensional structure identification system according to claim 2.

5. The stereostructure identification system according to claim 4, wherein the constructing unit rearranges the elements of each row in descending order by swapping rows and columns so that the atom of interest becomes an element in row 1 and column.

6. The stereoscopic structure identification system according to claim 4, wherein the image forming unit normalizes the reference center proximity score matrix within a specific range and converts it into pixel values ​​within the range of 0 to 255.

7. The construction unit associates each element of the proximity score matrix with atomic identification information based on atomic number and molecular characteristic information including bond valency, polarity, bond orientation and interatomic angle, or one or more combinations thereof; the image forming unit assigns the pixel values ​​to a first color channel, assigns the atomic identification information to a second color channel, assigns the molecular characteristic information to a third color channel, and forms the image based on the first color channel, the second color channel and the third color channel; the three-dimensional structure identification system according to claim 1.

8. A method for identifying three-dimensional structures, comprising: a computer, for identifying two or more three-dimensional structures composed of predetermined atoms, the method comprising: acquiring spatial coordinate data of all atoms constituting each three-dimensional structure; calculating an interatomic distance matrix based on the acquired spatial coordinate data, with the distances between all atoms of each three-dimensional structure as elements; constructing a proximity score matrix by transforming the calculated interatomic distance matrix so that a higher score is assigned to closer distances; converting the proximity score matrix into pixel values ​​within a specific range, and forming an image based on the pixel values; and comparing the structures of the two or more three-dimensional structures in whole or in part by analyzing the image.

9. A computer-readable program for causing a computer to perform the following steps: acquiring spatial coordinate data of all atoms constituting each solid; calculating an interatomic distance matrix based on the acquired spatial coordinate data, with the distances between all atoms of each solid as its elements; constructing a proximity score matrix by transforming the calculated interatomic distance matrix so that closer distances are assigned higher scores; converting the proximity score matrix into pixel values ​​within a specific range and forming an image based on those pixel values; and comparing the structures of two or more solids, in whole or in part, by analyzing the image.

Images