JPWO2019235567A1 - Protein interaction analyzer and analysis method - Google Patents

Abstract

Accurately analyze specific interactions between proteins and ligands. By machine learning using the surface shape data of a predetermined ligand binding site, the electrostatic potential distribution data of the ligand binding site, and the three-dimensional structure data of the ligand interacting with the ligand binding site as teacher data, it is related to the analysis target. Generate data on the protein interactions that occur.

Description

The present invention relates to a protein interaction analyzer and an analysis method used when analyzing the interaction between a ligand and a protein.

Enzymes have so-called substrate specificity, which recognizes a substrate having a specific structure. In addition, the receptor specifically binds to a physiologically active substance having a specific structure and exerts its action (for example, signal transduction activity and transcription promoting activity). Thus, proteins such as enzymes and receptors function through specific binding to so-called ligands such as substrates and bioactive substances.

As a result of research on proteins, base sequence information, amino acid sequence information and three-dimensional structure information on genes encoding them are accumulated daily. Of these, for sequence information, for example, NCBI (National Center of Biotechnology Information) Genbank, DNA Data Bank of Japan (DDBJ), and EMBL have been constructed. For information on the three-dimensional structure of proteins, the Protein Data Bank, which includes the Japan Protein Data Bank Japan (PDBj), has been constructed. Furthermore, KEGG has been constructed as a database that integrates information on intermolecular networks such as metabolism and signal transduction.

Among various efforts using such various data, "synthetic biotechnology" that predicts metabolic pathways and gene sequences that natural microorganisms do not have by computational science and artificially designs them is drawing attention. In "synthetic biotechnology", for example, in order to synthesize a substance for production purpose, a metabolic pathway for biosynthesizing the final target substance from a starting substance is constructed using the above-mentioned various data, and genome editing or the like is performed. A host organism is prepared by the method. Here, the metabolic pathway can be designed by combining a plurality of enzymatic reactions consisting of a substrate and an enzyme.

In addition, a method for high-throughput screening of lead compounds against target proteins has been proposed in the field of drug discovery by using the various data described above. In this method, for example, the basic structure of a lead compound capable of interacting with a ligand binding site is designed based on the three-dimensional structure data of the ligand binding site in the target protein.

As described above, the knowledge about the specific interaction of a protein with a ligand (interaction between a substrate and an enzyme, interaction between a target protein and a lead compound) is very useful in the fields of synthetic biotechnology and drug discovery. It turns out that the data is highly valuable.

In addition, Patent Document 1 discloses a method for identifying the function of a protein whose function is unknown in consideration of protein-protein interaction. In the method disclosed in Patent Document 1, teacher data is obtained from the three-dimensional structure of a plurality of receptors having known functions and the three-dimensional structure of a plurality of ligands. The teacher data disclosed in Patent Document 1 includes, for each receptor, a plurality of shape complementarity evaluation values when each receptor docks with a plurality of ligands, and the total total charge and the total surface charge of each receptor. The charge information including the difference between the two is included. Then, in the method disclosed in Patent Document 1, a plurality of shape complementarity evaluation values when a protein of unknown function docks with a plurality of ligands and charge information of the protein of unknown function are input, and the above teacher data is learned. Identify the function of a protein of unknown function.

On the other hand, in Non-Patent Document 1, the conformation formed between these domains is analyzed based on the electrostatic potential distribution of the adenylation domain (AdD) and the oligonucleotide binding domain (OBD) in DNA ligase, and the enzymatic reaction is performed. We are verifying the relationship with. Further, Patent Document 2 discloses that a mutation is introduced into the domain by a site-specific mutation method to identify an amino acid residue that is deeply involved in a ligation reaction. From these Patent Documents 1 and 2, it can be understood that the functional analysis of the protein is possible by the conformation analysis based on the electrostatic potential distribution of the surface of the protein.

Japanese Patent No. 5170630

Tanabe M., Ishino S., Yohda M., Morikawa K., Ishino Y., Nishida H. (2012) Structure-based mutational study of an archaeal DNA ligase towards improvement of ligation activity. ChemBioChem 13, 2575-2582. Tanabe M., Ishino Y., Nishida H. (2015) From structure-function analyzes to protein engineering for practical applications of DNA ligase. Archaea ID 267570.

Problems to be solved by the invention

As described above, even if the knowledge about the specific interaction between the protein and the ligand is derived based on the novel protein-related information accumulated daily, there is a problem that the desired substance production is not achieved in the synthetic biotechnology at present. In the field of drug discovery, there is a problem that a lead compound having high binding activity cannot be designed.

Therefore, in view of the above-mentioned circumstances, the present invention provides a protein interaction analysis device and an analysis method that output protein interaction-related data capable of accurately analyzing a specific interaction between a protein and a ligand. The purpose.

As a result of diligent studies by the present inventors in order to achieve the above-mentioned object, the ligand binding site association including at least the three-dimensional structure data regarding the ligand binding site and the electrostatic potential distribution at the ligand binding site based on the protein-related information. By using the data, it was found that the specific interaction between the ligand and the protein can be accurately analyzed, and the present invention has been completed.

The present invention includes the following.
(1) A data input unit for inputting information about the analysis target,
The surface shape of the ligand binding site for a given protein, generated based on the amino acid sequence data and three-dimensional structure data of the protein stored in the external storage unit and the three-dimensional structure data of the ligant that specifically interacts with the protein. A data storage unit that stores data, electrostatic potential distribution data of the ligand binding site, and three-dimensional structure data of a ligand that interacts with the ligand binding site in association with each other.
The surface shape data of a predetermined ligand binding site, the electrostatic potential distribution data of the ligand binding site, and the three-dimensional structure data of the ligand interacting with the ligand binding site, which are stored in the data storage unit, are used as teacher data. A protein interaction analysis apparatus including a calculation processing unit that generates data related to protein interactions related to an analysis target input by the data input unit by the above-mentioned machine learning.

(2) The information regarding the analysis target is information regarding the structure of the ligand.
The protein interaction analysis apparatus according to (1), wherein the calculation processing unit generates data on a protein or a ligand binding site that interacts with the ligand.

(3) The information regarding the analysis target is information regarding the structure of the protein or ligand binding site.
The protein interaction analysis apparatus according to (1), wherein the calculation processing unit generates data on a compound or ligand that interacts with the protein or ligand binding site.

(4) The calculation processing unit calculates an evaluation value indicating the similarity between the analysis target input by the data input unit and the analysis target included in the generated data for the data related to the protein interaction generated by machine learning. The protein interaction analysis apparatus according to (1), wherein the evaluation value calculation unit is provided.

(5) The calculation processing unit calculates a suitability score that quantitatively indicates the binding stability when the analysis target input by the data input unit interacts with respect to the data related to the protein interaction generated by machine learning. The protein-protein interaction analyzer according to (1), which comprises a protein-ligand compatibility score calculation unit.

(6) The data storage unit calculates the center coordinates of the ligand binding site based on the atomic coordinates of the atoms constituting the ligand binding site, and creates a three-dimensional grid space containing atoms within a predetermined distance from the center coordinates. The protein interaction analyzer according to (1), which is set and stores surface shape data generated based on the three-dimensional grid space.

(7) The three-dimensional grid space has a plurality of grid points by a grid set at predetermined intervals, and a specific character is assigned to the grid point closest to each atom within a predetermined distance from the center coordinates. The protein-protein interaction analyzer according to (6), wherein the data is the data in which another character is given to the lattice point to which the specific character is not given.

(8) The protein-protein interaction analysis apparatus according to (6), wherein each atom within a predetermined distance from the center coordinates is a plurality of non-hydrogen atom species.

(9) The protein interaction analysis apparatus according to (6), wherein the data storage unit stores electrostatic potential distribution data calculated for grid points in the three-dimensional grid space.

(10) The data storage unit has positive electrostatic potential distribution data consisting of positive values calculated for the grid points in the three-dimensional grid space and negative values calculated for the grid points in the three-dimensional grid space. The protein interaction analyzer according to (6), which stores negative electrostatic potential distribution data comprising.

(11) The process of inputting information about the analysis target by the input device and
Based on the amino acid sequence data and three-dimensional structure data of the protein stored in the external storage unit and the three-dimensional structure data of the ligant that specifically interacts with the protein, the arithmetic unit determines the ligand binding site for a predetermined protein. The surface shape data, the electrostatic potential distribution data of the ligand binding site, and the three-dimensional structure data of the ligand interacting with the ligand binding site are generated, and these surface shape data, the electrostatic potential distribution data, and the three-dimensional structure of the ligand are generated. The process of associating data with the storage device and storing it in the storage device,
The arithmetic unit stores the surface shape data of a predetermined ligand binding site, the electrostatic potential distribution data of the ligand binding site, and the three-dimensional structure data of the ligand interacting with the ligand binding site stored in the storage device. A protein interaction analysis method comprising a step of generating data on a protein interaction related to an analysis target input by the input device by machine learning as teacher data.

(12) The information regarding the analysis target is information regarding the structure of the ligand.
The protein interaction analysis method according to (11), wherein the arithmetic unit generates data on a protein or a ligand binding site that interacts with the ligand.

(13) The information regarding the analysis target is information regarding the structure of the protein or ligand binding site.
The protein interaction analysis method according to (11), wherein the arithmetic unit generates data on a compound or ligand that interacts with the protein or ligand binding site.

(14) A step in which the calculation device calculates an evaluation value indicating the similarity between the analysis target input by the input device and the analysis target included in the generated data for the data related to the protein interaction generated by machine learning. (11) The protein interaction analysis method according to (11).

(15) A step of calculating a suitability score that quantitatively indicates the binding stability when the analysis target input by the input device interacts with the data related to the protein interaction generated by the machine learning by the calculation device. (11) The protein interaction analysis method according to (11).

(16) The arithmetic device calculates the center coordinates of the ligand binding site based on the atomic coordinates of the atoms constituting the ligand binding site, and sets a three-dimensional grid space including atoms within a predetermined distance from the center coordinates. The protein interaction analysis method according to (11), wherein the surface shape data generated based on the three-dimensional grid space is stored in the data storage unit.

(17) The three-dimensional grid space has a plurality of grid points by a grid set at predetermined intervals, and a specific character is assigned to the grid point closest to each atom within a predetermined distance from the center coordinates. The protein-protein interaction analysis method according to (11), wherein the data is the data in which another character is given to the lattice point to which the specific character is not given.

(18) The protein interaction analysis method according to (11), wherein each atom within a predetermined distance from the center coordinates is a plurality of non-hydrogen atom species.

(19) The protein interaction analysis method according to (11), wherein the arithmetic unit stores electrostatic potential distribution data calculated for grid points in the three-dimensional grid space in the data storage unit.

(20) The calculation device is based on positive electrostatic potential distribution data consisting of positive values calculated for the grid points in the three-dimensional grid space and negative values calculated for the grid points in the three-dimensional grid space. The protein interaction analysis method according to (11), wherein the negative electrostatic potential distribution data is stored in the data storage unit.

This specification includes the disclosure of Japanese Patent Application No. 2018-108362, which is the basis of the priority of the present application.

According to the protein interaction analyzer and analysis method according to the present invention, the specific interaction between a protein and a ligand can be accurately analyzed. For example, according to the protein interaction analyzer and analysis method according to the present invention, a protein or ligand that has a high possibility of specifically interacting with a ligand or a protein specified by a user is analyzed with high accuracy by machine learning. be able to.

It is a block diagram which shows an example of the protein interaction analysis apparatus to which this invention was applied. It is a block diagram which shows an example of the calculation processing part in the protein interaction analysis apparatus to which this invention was applied. It is a conceptual diagram about the extraction method of the ligand binding part in the protein interaction analysis apparatus to which this invention was applied, and the production method of Voxel which becomes learning data. It is a conceptual diagram which arranges the coordinates and the electrostatic potential value of each atom of a protein in the protein interaction analysis apparatus to which this invention is applied to the close lattice point in a voxel. It is a block diagram which shows another example of the calculation processing part in the protein interaction analysis apparatus to which this invention was applied.

Hereinafter, the present invention will be described in detail with reference to the drawings.

The protein interaction analyzer to which the present invention is applied can be used for three-dimensional structure analysis of a site (ligand binding site) that interacts with a ligant in the protein to be analyzed based on amino acid sequence data and three-dimensional structure data of the protein. Data (hereinafter referred to as ligand binding site surface texture data) is generated, and the interaction between the ligand and the protein is analyzed through machine learning using the data. The ligand binding site surface property data is data obtained by combining the surface shape data of the ligand binding site and the electrostatic potential distribution data of the ligand binding site.

As an example, the protein interaction analyzer 1 shown in FIG. 1 is connected to an external storage unit 2 that stores amino acid sequences and three-dimensional structure data related to a protein and data related to a ligand for the protein, and is connected to an external storage unit 2 that stores a ligand binding site in a predetermined protein. A surface shape data generation unit 3 that generates surface shape data, an electrostatic potential distribution data generation unit 4 that generates an electrostatic potential distribution of the ligand binding site, and a ligand three-dimensional structure that generates three-dimensional structure data regarding a ligand for the protein. It includes a data generation unit 5. In addition, the protein interaction analysis apparatus 1 includes surface shape data and electrostatic potential distribution data (ligand binding site surface texture data) regarding a predetermined ligand binding site generated by the surface shape data generation unit 3 and the electrostatic potential distribution data generation unit 4. ) And a data storage unit 6 that stores three-dimensional structure data relating to the ligand interacting with the ligand binding site as teacher data. Further, the protein interaction analysis device 1 includes a data input unit 7 for inputting data to be analyzed by the user.

Furthermore, the protein interaction analysis device 1 uses the teacher data stored in the data storage unit 6 and generates data on the protein interaction by machine learning for the analysis target input by the data input unit 7. And an output unit 9 for outputting the result calculated by the calculation processing unit 8.

The calculation processing unit 8 analyzes the interaction between the ligand and the protein according to the user's specification, which will be described in detail later. As an example, as shown in FIG. 2, the calculation processing unit 8 applies to the machine learning unit 10 that performs machine learning using the teacher data stored in the data storage unit 6 and the analysis target input by the data input unit 7. On the other hand, the evaluation value calculation unit 11 for calculating the evaluation value indicating the similarity to the protein or ligand contained in the teacher data, the result of machine learning performed by the machine learning unit 10, and the evaluation value calculated by the evaluation value calculation unit 11. It is provided with a list generation unit 12 that generates a list including the above.

Here, the protein interaction analysis device 1 shown in FIG. 1 is configured to be connected to one external storage unit 2 that stores the above-mentioned data. However, although not shown, the protein interaction analysis device 1 may be connected to a plurality of external storage units in which the above-mentioned data is distributed and stored. For example, the protein interaction analysis device 1 may be capable of connecting to an external storage unit that stores amino acid sequence and three-dimensional structure data related to a protein and an external storage unit that stores data related to a ligand for a protein. ..

The data stored in the external storage unit 2 is data on the amino acid sequence, three-dimensional structure data, and ligand of a predetermined protein. Here, the ligand broadly means a substrate for an enzyme, a low molecular weight compound that interacts with a receptor protein, a substance that specifically interacts with a protein, such as a coenzyme or a regulator. The ligand may be interpreted as being limited to a substance that binds to a receptor existing on the cell membrane or an intracellular receptor. However, the term "ligand" is used in a broad sense to include substances that specifically interact with proteins, including substrates for enzymes, coenzymes, regulators, substances that bind to receptors, and the like. Used in. Therefore, the ligand may be either a low molecular weight compound or a high molecular weight compound, or may mean a partial region of the compound. That is, the molecular structure and atomic coordinates of the ligand may be the molecular structure and atomic coordinates of the entire compound that interacts with the protein, or may be the molecular structure and atomic coordinates of at least a partial region of the compound that interacts with the protein.

The protein means a polymer compound having an amino acid sequence as a primary structure, and may be any of a monomer, a homomultimer, and a heteromultimer. In addition, the protein may have post-translational chemical modifications such as sugar chain addition, functional group addition, and phosphorylation. Therefore, the three-dimensional structure data based on the atomic coordinates at the ligand binding site may be the three-dimensional structure data based on the atomic coordinates obtained by the protein having no post-translation chemical modification described above, or the above-mentioned post-translation chemical modification. It may be three-dimensional structure data based on the atomic coordinates obtained by the protein possessed. The three-dimensional structure data based on the atomic coordinates at the ligand binding site is the atomic coordinates obtained from the above-mentioned post-translation protein without chemical modification, modified so as to be the atomic coordinates of the protein having a predetermined chemical modification. It may be three-dimensional structure data based on the (corrected) atomic coordinates.

Atomic coordinates mean data showing the coordinates of the atoms that make up a protein. Atomic coordinates can be obtained for various proteins by one or both of an X-ray crystallography method mainly using a protein single crystal and a nuclear magnetic resonance method targeting a protein solution. Atomic coordinates can also be obtained by nuclear magnetic resonance technology using stable isotopes called stereo-array isotope labeling.

The atomic coordinates of a protein are not particularly limited to a format, but each atom constituting the protein may be in a format in which x-coordinates, y-coordinates and z-coordinates are shown as a combination. The unit of each coordinate can be, for example, [Å].

As an example of the external storage unit 2 that stores the above-mentioned data, a Protein Data Bank (hereinafter, PDB) including a Japan Protein Data Bank Japan (PDBj) can be mentioned. That is, the protein interaction analysis device 1 can be configured to be able to connect to the PDB as the external storage unit 2. In PDB, atomic coordinates are displayed as one line of data for each atomic number under a predetermined record name (standard amino acid is ATOM), for example. As an example, for a predetermined atomic number, the atomic name (main chain angstrom: N, α carbon: CA, β carbon: CB), residue name (amino acid three-letter notation), Chain ID, residue number, each of the atoms x-coordinate [Å], y-coordinate [Å], z-coordinate [Å], occupancy (sample to be analyzed, for example, the proportion of the atom present in its place in the crystal, occupancy, usually 1.00) and temperature factor B [ Å ² ] (if determined by X-ray crystallography) can be included in the data.

In addition to the above-mentioned data on amino acids, the data stored in PDB includes data on protein molecule types, registered names and accession numbers (HEADER line), and title names when published on PDB (TITLE). (Line), information about protein molecules (COMPND line), information about protein host (SOURCE line), information about experiments during stereoanalysis (REMARK line), amino acid sequence information (SEQRES line), α-helix Contains information on the amino acids that make up the protein (HELIX line) and position information on intramolecular disulfide bonds (SSBOND).

In particular, the data stored in the PDB contains data on the molecular structure and atomic coordinates of the ligand interacting with the ligand binding site described above. Specifically, the PDB contains information about the molecular structure of the ligand (HETATM line) and information about ligand binding (CONECT line). In the data stored in the PDB, the HETATM line contains information that identifies an atom that constitutes a ligand and the coordinates of that atom. In addition, the HETATM line contains information indicating the type of conformer when the ligand has a conformer.

The protein interaction analysis device 1 generates surface shape data regarding a predetermined protein including a ligand binding site in the surface shape data generation unit 3 using the data stored in the external storage unit 2. The surface shape data generation unit 3 may generate the entire surface shape data of the protein, or may generate the surface shape data for the partial region including the ligand binding site in the protein. In particular, the surface shape data generation unit 3 preferably generates surface shape data for a partial region including the entire ligand binding site in the protein. The surface shape data can be data in which the surface of the protein to which the ligand binds is the xy plane and the unevenness in the xy plane is indicated by a value in the z-axis direction.

Here, the xy plane in the surface shape data can be a plane in which the ligand occupies the maximum area in a state where the ligand specifically interacts with the protein. That is, it is preferable that the plane on which the ligand can be projected most is the xy plane in the surface shape data in the state where the ligand specifically interacts with the protein. Alternatively, the xy plane in the surface shape data can be a plane in which the ligand binding site in the protein occupies the maximum area. That is, in the three-dimensional structure of the protein, it is preferable that the plane having the largest ligand binding site is the xy plane in the surface shape data.

The value in the z-axis direction in the xy plane constituting the surface shape data is obtained as a function (z = f (x, y)) shown over the entire xy plane using the data stored in the external storage unit 2. Can be done. Further, the value in the z-axis direction in the xy plane can be obtained as a discrete value at the mesh point (intersection) by using the xy plane as mesh data. For example, when the xy plane is used as mesh data, for example, mesh data at intervals of 0.05 to 1.0 Å, preferably mesh data at intervals of 0.1 to 0.5 Å, and more preferably mesh data at intervals of 0.2 Å can be used. It is also possible to obtain the value in the z-axis direction in the xy plane as a discrete value at the point (intersection). Further, the value in the z-axis direction in the xy plane can be obtained as the average value of the values in the z-axis direction in the xy plane calculated for the region in each mesh by using the xy plane as the mesh data as described above.

Further, the surface shape data generation unit 3 may generate a plurality of surface shape data for one ligand binding site. The surface shape data generation unit 3 may generate a pair of surface shape data which are so-called stereograms as a plurality of surface shape data. The surface shape data generation unit 3 makes a predetermined angle, for example, ± 0.5 to 10 degrees around the x-axis or the y-axis in the xy plane in which the ligand occupies the maximum area in a state where the ligand specifically interacts with the protein. A plurality of planes inclined in the range of ± 1 to 5 degrees may be set, and surface shape data may be generated for each of the plurality of planes. More specifically, the surface shape data generation unit 3 is tilted ± 5 degrees with respect to the xy plane in which the ligand occupies the maximum area and the y-axis in the xy plane in a state where the ligand specifically interacts with the protein. Surface shape data (total, three surface shape data) can be generated for each of the two planes. Alternatively, the surface shape data generation unit 3 has surface shape data (total,) for each of the xy plane in which the ligand binding site in the protein occupies the maximum area and the two planes tilted ± 5 degrees about the y-axis in the xy plane. Three surface shape data) can be generated.

Further, the surface shape data generation unit 3 is in the xy plane in which the ligand occupies the maximum area in a state where the ligand specifically interacts with the protein, and one side is in the range of, for example, 30 to 100 Å, preferably 40. It is possible to generate xy planes in the range of ~ 80 Å, more preferably in the range of 45-60 Å. Alternatively, the surface shape data generation unit 3 is an xy plane in which the ligand binding site in the protein occupies the maximum area, and one side is, for example, in the range of 30 to 100 Å, preferably in the range of 40 to 80 Å, and more preferably in the range of 45 to 45 Å. It is possible to generate xy planes in the range of 60 Å.

On the other hand, the protein interaction analysis device 1 generates electrostatic potential distribution data of a ligand binding site in a predetermined protein in the electrostatic potential distribution data generation unit 4 using the data stored in the external storage unit 2. The electrostatic potential distribution data generation unit 4 can calculate the electrostatic potential distribution of the ligand binding site by applying a known method for calculating the surface charge of the protein.

The electrostatic potential distribution data generation unit 4 can appropriately use a conventionally known method for calculating the electrostatic potential (surface charge) of a protein. Here, the electrostatic potential (surface charge) can be defined as the Coulomb energy that a positive charge having a unit electric amount receives at an arbitrary point.

To calculate the electrostatic potential (surface charge) of a protein, first, carbon (C) and oxygen (O) are obtained from the amino acid sequence and atomic coordinate data of the protein stored in the external storage unit 2 such as PDB. , Nitrogen (N), Sulfur (S) and other non-hydrogen atom species and their coordinates are read. Next, the hydrogen atom bonded to each non-hydrogen atom species read and its coordinates are calculated. Next, by using these coordinates together, it is possible to obtain information on all atoms coordinated on the surface in the molecule, that is, information on all electrons. Then, using this information, it is possible to calculate the electrostatic charge at an arbitrary position inside and outside the protein molecule by assuming a constant dielectric constant. In particular, the positive charge calculated on the surface of the protein can be used as the electrostatic potential (surface charge) of the protein. It is also possible to obtain the electrostatic potential distribution by normalizing the calculated electrostatic potential (surface charge) of the protein to, for example, +5 to −5. Further, this electrostatic potential distribution can be calculated as a function of spatial coordinates (c = f (x, y, z) with the electrostatic potential value as c) over the entire space assuming a constant permittivity.

Then, the value on the curved surface obtained by the surface shape data generation unit 3 is extracted from this spatially continuous value, and the (x, y, z) value of the surface shape data and the electrostatic potential (surface charge) value c are combined. It is desirable to store it as 4-dimensional data (as (x, y, z, c)).

As a conventionally known method for calculating the electrostatic potential (surface charge) of a protein, for example, Rocchia et al. Vol. 23, No. 1 Journal of Computational Chemistry, 128-137, 2002 can be mentioned. In addition, examples of software that can be used to calculate the electrostatic potential (surface charge) include GRASP, Chimera, APBS, and QUANTA.

The electrostatic potential distribution data generation unit 4 may generate an electrostatic potential distribution for the entire region of the xy plane created by the surface shape data generation unit 3, or may generate an electrostatic potential distribution for a partial region of the xy plane. It may be generated. The partial region of the xy plane created by the surface shape data generation unit 3 includes a region including a ligand binding site contained in the xy plane, for example, a spatial region within 10 Å, preferably within 5 Å from the ligand binding site, from the surface charge. An electrostatic potential distribution may be generated.

Further, when the surface shape data generation unit 3 creates a plurality of xy planes for one ligand binding site, the electrostatic potential distribution data generation unit 4 generates an electrostatic potential distribution for all xy planes. Alternatively, an electrostatic potential distribution may be generated for some xy planes.

The value of the electrostatic potential in the xy plane can also be obtained as a discrete value at the mesh point (intersection point) by using the xy plane as mesh data. For example, the xy plane can be, for example, mesh data at intervals of 0.05 to 1.0 Å, preferably mesh data at intervals of 0.1 to 0.5 Å, more preferably mesh data at intervals of 0.2 Å, discrete at mesh points (intersections). It is also possible to obtain the value of the surface charge as a target value. Further, the value of the surface charge in the xy plane can be obtained as the average value of the electrostatic potentials calculated for the regions in each mesh by using the xy plane as the mesh data as described above.

By the way, as described above, the surface shape data generation unit 3 sets the surface of the protein or the surface of the ligand binding site as the xy plane, and generates data showing the unevenness in the xy plane by the value in the z-axis direction, and generates the electrostatic potential. The distribution data generation unit 4 generated an electrostatic potential distribution for the entire region or a partial region of the xy plane created by the surface shape data generation unit 3, but the protein interaction analyzer 1 is not limited to this form. .. That is, in the protein interaction analysis apparatus 1, the surface shape data generation unit 3 defines, for example, a three-dimensional grid space generated based on the atomic coordinates of the atoms constituting the protein, and generates the entire surface shape data of the protein. Alternatively, surface shape data may be generated for a partial region containing a ligand binding site in the protein. Then, the electrostatic potential distribution data generation unit 4 may generate an electrostatic potential distribution for this surface shape data.

More specifically, from the data set stored in the PDB, data related to atomic coordinates other than the residue name (three-letter notation of amino acid) is extracted from the data associated with each atomic number. That is, from the data set stored in the PDB, data on the atomic coordinates of the entire protein or the ligand binding site in the protein is extracted.

Next, the center coordinates of the entire protein or the ligand binding site are calculated. The method for calculating the center coordinates is not particularly limited, but for example, the x-coordinate [Å], is obtained from the atomic coordinates of the atoms constituting the entire protein extracted as described above or the atomic coordinates of the atoms constituting the ligand binding site. The arithmetic averages of the y-coordinate [Å] and the z-coordinate [Å] can be calculated, and the obtained average value can be used as the center coordinate. When calculating the central coordinates of the ligand binding site, after extracting the atomic coordinates of the entire protein, only the atomic coordinates of the atoms constituting the ligand binding site are further extracted and the arithmetic average is calculated as described above. You may.

Next, atoms within a predetermined distance are extracted from the calculated center coordinates of the entire protein or the center coordinates of the ligand binding site. In other words, a sphere having a predetermined radius is given from the calculated center coordinates, and all atoms located inside the sphere are extracted. At this time, the distance from the center coordinates, that is, the radius of the spherical surface can be arbitrarily set, for example, in the range of 15 to 50 Å, preferably in the range of 20 to 40 Å, and more preferably in the range of 23 to 30 Å. ..

Next, a cube inscribed or circumscribed with respect to a sphere having a predetermined radius from the center coordinates is given, and a three-dimensional grid space is given by dividing each side of the cube at a predetermined interval. The predetermined interval is not particularly limited, but may be, for example, 0.25 Å or 0.5 Å. Then, the coordinates of the grid points of each division in the three-dimensional grid space should be defined as a coordinate system common to the atomic coordinates of the atoms constituting the entire protein for which the central coordinates have been calculated or the atomic coordinates of the atoms constituting the ligand binding site. Can be done.

As described above, the surface shape data generation unit 3 can define the three-dimensional grid generated based on the atomic coordinates of the atoms constituting the protein and generate the surface shape data of the entire protein.

More specifically, as shown in FIG. 3, a three-dimensional grid space can be defined. In the example shown in FIG. 3, a sphere 14 having a predetermined radius (for example, 10 to 20 Å) can be given from the center coordinates, and the first division 15 (solid line in FIG. 3) is given to the cube inscribed in the sphere 14. , 0.5 Å) can be set. Further, in order to make the resolution finer, a second division 16 (broken line in FIG. 3) that further divides the first division 15 into two can be set. Thereby, a predetermined grid positions counts 17 can be defined on each side of the cube inscribed in the spherical surface 14. A three-dimensional grid space composed of cubes inscribed in a spherical surface 14 divided by a predetermined number of grids in this way is called a Voxel.

In FIG. 3, the cube inscribed in the spherical surface 14 is a three-dimensional grid space, Voxel, and the vicinity of eight corners of the cube is a space in which atomic coordinates do not exist. Although not shown, when the cube circumscribing the sphere 14 is a three-dimensional grid space, the atomic coordinates are included in all of the Voxel.

Next, with respect to the surface shape data represented as the three-dimensional grid space generated as described above, the electrostatic potential distribution data generation unit 4 uses the data stored in the external storage unit 2 to perform capacitance. Generate potential distribution data. The electrostatic potential distribution data generation unit 4 reads information on non-hydrogen atom species such as carbon (C), oxygen (O), nitrogen (N), sulfur (S) and their coordinates, and for each non-hydrogen atom species. Set Voxel, which is a three-dimensional grid space. Then, with respect to a predetermined non-hydrogen atom species, a specific character such as "1" is given to the lattice points closest to the non-hydrogen atom species among the lattice points in Voxel, and the lattice not given "1". Give other characters such as "0" to the dots. As an example, for carbon (C), for each carbon atom located inside a sphere (sphere 14 in FIG. 3) having a predetermined radius from the center coordinates, to the closest grid point in Voxel based on the coordinate data. On the other hand, "1" is given, and "0" is given to the lattice points where there are no adjacent carbon atoms. This process can generate Voxel "C" data for carbon atoms. By performing this treatment on all non-hydrogen atom species such as all oxygen (O), nitrogen (N), sulfur (S), Voxel "O" data on oxygen atom, Voxel "N" on nitrogen atom Voxel data can be generated for each non-hydrogen atom species, such as data and Voxel "S" data for sulfur atoms. The data set obtained in this way is referred to as three-dimensional convolution data. In FIG. 5A, as an example, for carbon (C), oxygen (O), nitrogen (N), and sulfur (S), each atom is assigned to the nearest lattice point based on the atomic coordinate data (the value of the lattice point is assigned. 1) is schematically shown.

Next, the electrostatic potential distribution data generation unit 4 is described by, for example, the method disclosed in Rocchia et al. Vol. 23, No. 1. Journal of Computational Chemistry, 128-137, 2002, GRASP, Chimera, APBS and QUANTA. The electrostatic potential distribution can be generated by using commercially available software such as.

Next, the electrostatic potential distribution data generation unit 4 can store the calculated electrostatic potential values having a positive value and those having a negative value as different data. That is, the electrostatic potential distribution data generation unit 4 can generate Voxel “positive” data and Voxel “negative” data based on the calculated electrostatic potential value. FIG. 5B schematically shows that different data are stored depending on whether the electrostatic potential value is “positive” or “negative”. In FIG. 5B, the inside of the circle is represented by shading according to the absolute values of the “positive” and “negative” values. And these Voxel "positive" data and Voxel "negative" data are three-dimensional together with the above-mentioned Voxel "O" data regarding oxygen atom, Voxel "N" data regarding nitrogen atom, Voxel "S" data regarding sulfur atom, etc. It can be convolution data (3D Convolution data).

On the other hand, the protein interaction analysis device 1 generates three-dimensional structure data related to the ligand in the ligand structure data generation unit 5 using the data stored in the external storage unit 2. The ligand structure data generation unit 5 can obtain the three-dimensional structure data of the ligand by appropriately using a conventionally known method.

When obtaining the three-dimensional structure data of the ligand, the ligand structure data generation unit 5 uses the data on the ligand stored in the external storage unit 2, that is, the molecular formula, the structural formula, the compound name, and the information on the atom interacting with the protein. The three-dimensional structure of the ligand is extracted, and the three-dimensional structure data of the ligand is generated based on these. For example, when using PDB, the residue name is searched using the _nonpolymer flag as an index from the same file from which the atomic coordinate data used when generating the surface shape data described above is extracted, and the residue name is not an amino acid. And / or its residue name can be determined to be a ligand molecule because it refers to an organic compound other than a protein. In PDB, the compound name is associated with the _nonpolymer flag as a three-letter code. Therefore, for the compound judged as the ligand molecule, the three-letter code can be used as a clue to extract the three-dimensional coordinates (x, y, z) of all or some of the atoms constituting the ligand from the coordinate data description part at the rear of the file. it can. At this time, the three-dimensional structure data of the ligand generated by the ligand structure data generation unit 5 may be the three-dimensional structure data of the entire compound interacting with the protein, or the three-dimensional structure data relating to the partial region interacting with the protein in the compound. Is also good.

The ligand structure data generation unit 5 may use the molecular compound structure description method as two-dimensional graph structure data when obtaining the three-dimensional structure data of the ligand. Examples of the molecular compound structure description method include a SMILES (simplified molecular input line entry system) description method, a SMARTS (Smiles Arbitrary Target Specification) description method, and an InChI (International Chemical Identifier) description method. In particular, it is preferable to generate the three-dimensional structure data of the ligand by the SMILES description method. The ligand structure data generation unit 5 can be used as graph convolution data for machine learning two-dimensional graph structure data using a molecular compound structure description method such as the SMILES description method.

Further, it is preferable that the ligand structure data generation unit 5 generates an electrostatic potential distribution for the ligand in addition to the three-dimensional structure data of the ligand. The electrostatic potential distribution for the ligand can be determined, for example, according to the method described in Rocchia et al. Journal of Computational Chemistry, Vol. 23, No. 1, pages 128-137. The electrostatic potential distribution for the ligand may be the electrostatic potential distribution of the entire compound that interacts with the protein, or may be the electrostatic potential distribution for the partial region that interacts with the protein in the compound.

Then, the protein interaction analyzer 1 generates surface shape data and electrostatic potential distribution data 4 of the ligand binding site generated by the surface shape data generation unit 3 regarding the combination of the ligand binding site and the ligand in a predetermined protein. The electrostatic potential distribution data of the ligand binding site generated in 1 and the three-dimensional structure data related to the ligand generated by the ligand structure data generation unit 5 are associated and stored in the data storage unit 6. That is, the data storage unit 6 contains "surface shape data", "electrostatic potential distribution data", and "three-dimensional structure data on the ligand" for a plurality of combinations of the ligand binding site and the ligand in a predetermined protein. Stores property data. In addition to these data, the data storage unit 6 may store electrostatic potential distribution data related to the ligand in association with each other, or juxtapose the three-dimensional structure information of a plurality of three-dimensional structural isomers (rotors) possessed by the ligand. You may memorize it.

The protein interaction analysis device 1 generates data on the protein interaction desired by the user by machine learning using a plurality of ligand binding site surface texture data stored in the data storage unit 6 as teacher data. The user inputs information about the analysis target to the data input unit 7 in the protein interaction analysis device 1. The information regarding the analysis target is information about a protein that interacts with a predetermined compound or its ligand binding site when it is analyzed, and is a compound that interacts with a predetermined protein or its ligand binding site. When analyzing a protein or a ligand, it is information on the protein or its ligand binding site.

Examples of the information regarding the compound input to the data input unit 7 include information regarding the three-dimensional structural formula or molecular formula and three-dimensional structure of the compound, and information regarding the three-dimensional structural formula or molecular formula and three-dimensional structure of a partial region of the compound. Based on the information on these compounds, a candidate protein or a candidate ligand binding site that interacts with the compound can be analyzed by a process described in detail later.

For example, when analyzing a candidate protein (candidate enzyme) involved in an enzymatic reaction for synthesizing a target compound (product) from a predetermined compound (substrate) using a protein interaction analyzer 1, a substrate. As information on the compound to be used, information on the three-dimensional structural formula or molecular formula and three-dimensional structure of the substrate compound, and information on the three-dimensional structural formula or molecular formula and three-dimensional structure of the region on which the enzyme acts in the substrate compound is input to the data input unit 7. In this example, the type of the enzymatic reaction from the substrate to the product and the name of the enzyme involved in the enzymatic reaction may be input to the data input unit 7.

Further, examples of the information regarding the protein or ligand binding site input to the data input unit 7 include the amino acid sequence, atomic coordinates, and three-dimensional structure of the protein or ligand binding site. Based on the information on these proteins or ligand binding sites, a candidate compound (candidate ligand) that interacts with the protein or ligand binding site can be analyzed by a process described in detail later.

For example, when a compound (ligand compound) that interacts with a predetermined protein (for example, a receptor protein) is selected by using the protein interaction analyzer 1, information on the protein includes an amino acid sequence and a steric structure of the protein. Structural data, the amino acid sequence of the ligand binding site, or the three-dimensional structure data of the ligand binding site is input to the data input unit 7.

In the calculation processing unit 8 of the protein interaction analysis apparatus 1, the ligand binding site surface texture data stored in the data storage unit 6 and the ligand binding site interact with each other based on the information regarding the analysis target input by the data input unit 7. Generate data on protein interactions related to the analysis target, including analysis results by machine learning using a plurality of sets of three-dimensional structure data on acting ligands as teacher data.

For example, when the information about the analysis target input by the data input unit 7 is the information about the compound, the calculation processing unit 8 generates a candidate protein or a candidate ligand binding site that may interact with the compound. More specifically, when the information about the analysis target input by the data input unit 7 is the information about the compound that is the substrate in the predetermined enzyme reaction, the calculation processing unit 8 is the candidate enzyme that may use the compound as the substrate. To generate. At this time, the calculation processing unit 8 may generate one candidate protein, a candidate ligand binding site, or a candidate enzyme that is most likely to interact with the compound, or interacts with the compound. A group of probable candidate proteins or candidate ligand binding sites or candidate enzymes may be produced.

Further, when the information regarding the analysis target input by the data input unit 7 is the information regarding the protein or ligand binding site, the calculation processing unit 8 is a candidate compound or a candidate compound that may interact with the protein or ligand binding site. Generate a candidate ligand. At this time, the calculation processing unit 8 may generate one candidate compound or candidate ligand that is most likely to interact with the protein or ligand binding site, or may generate one candidate compound or ligand binding site with respect to the protein or ligand binding site. A group of candidate compounds or ligands that are likely to interact may be produced.

In the example shown in FIG. 2, in the machine learning unit 10 in the calculation processing unit 8, a plurality of data sets of the above-mentioned ligand binding site surface texture data and the three-dimensional structure data relating to the ligand interacting with the ligand binding site are used as the teacher data. Perform analysis by machine learning. Further, in the example shown in FIG. 2, the evaluation value calculation unit 11 in the calculation processing unit 8 shows the similarity to the protein or ligand contained in the teacher data with respect to the analysis target input by the data input unit 7. Is calculated. In the example shown in FIG. 2, the list generation unit 12 in the calculation processing unit 8 generates a list in which the result of machine learning performed by the machine learning unit 10 and the evaluation value calculated by the evaluation value calculation unit 11 are combined.

Further, in the calculation processing unit 8, the machine learning teacher data processed by the machine learning unit 10 is not particularly limited, but includes, for example, an evaluation value for evaluating the interaction between the protein or the ligand binding site and the ligand molecule. Is preferable. As an example of the evaluation value, a higher score is obtained for a short distance between a ligand binding site and a ligand molecule in a protein from "ligand binding site surface texture data" and "ligand-related three-dimensional structure data". It is possible to use a three-dimensional shape unevenness homology evaluation value that gives. The three-dimensional shape unevenness homology evaluation value can be an n-dimensional vector having an common n number in each data set consisting of "ligand binding site surface texture data" and "three-dimensional structure data related to the ligand".

Here, the n-dimensional vector indicates the three-dimensional shape unevenness homology evaluation value at n predetermined sites in the ligand molecule. These n predetermined sites can be arbitrarily defined for each ligand molecule. As an example, as the order and sequence order in the n-dimensional vector, each non-hydrogen (carbon, nitrogen, oxygen) constituting the ligand molecule follows the ranking of carbon atoms according to the IUPAC (International Union of Pure and Applied Chemistry) nomenclature. , Sulfur, selenium, etc.). Thereby, for a predetermined ligand molecule, it is possible to determine the three-dimensional shape unevenness homology evaluation value at each site on the three-dimensional arrangement. The n number in the three-dimensional shape unevenness homology evaluation value may be a value common to each data set consisting of "ligand binding site surface texture data" and "three-dimensional structure data related to the ligand", but is a value different for each data set. But it's okay.

Further, the machine learning teacher data processed by the machine learning unit 10 is not particularly limited, and may include, for example, an evaluation value for evaluating the binding energy related to the electrostatic binding between the protein or the ligand binding site and the ligand molecule. preferable. As an example of the evaluation value, from "electrostatic potential distribution data" and "three-dimensional structure data on the ligand", regarding the electrostatic bond between the protein and the ligand molecule, the bond energy (enthalpy change) of each site in the ligand molecule. A bond energy rating that gives a higher score depending on the size can be used. The binding energy evaluation value can be an m-dimensional vector having a common m number in each data set consisting of "electrostatic potential distribution data" and "three-dimensional structure data on ligand".

Here, the m-dimensional vector indicates the binding energy evaluation value at m predetermined sites in the ligand molecule. These m predetermined sites can be arbitrarily defined for each ligand molecule as in the n-dimensional vector described above. The number of m in the binding energy evaluation value may be a different value for each set consisting of a protein or a ligand binding site and a ligand molecule, or may be a common value. The n number in the three-dimensional shape unevenness homology evaluation value may be a value common to each data set consisting of "electrostatic potential distribution data" and "three-dimensional structure data related to ligand", but may be a value different for each data set. good.

The values of the n number and the m number can be arbitrary. For example, as the values of n number and m number, after machine learning is performed using a part of the above-mentioned data set, the appropriateness of the answer to other data sets not used for machine learning becomes high. Can be set as.

On the other hand, when a predetermined compound or ligand is input as an analysis target in the data input unit 7, the evaluation value calculation unit 11 mutually refers to the compound or ligand extracted as a result of machine learning in the machine learning unit 10. Calculate the evaluation value for the candidate protein or candidate ligand binding site that may act. This evaluation value is a value indicating the similarity between the compound or ligand input by the data input unit 7 and the compound or ligand to which the extracted candidate protein or candidate ligand binding site is associated.

Specifically, in the evaluation value calculation unit 11, the compound or ligand input in the data input unit 7 is maximally matched with the extracted candidate protein or the compound or ligand to which the candidate ligand binding site is associated, and the degree of compatibility is achieved. A high evaluation value can be given to a molecule having a high degree of (matching). This evaluation value is obtained, for example, when the proportion of atoms contained in the input compound or ligand within a predetermined distance (for example, 1 Å) from the corresponding atom in the "ligand structure data" of the collation destination is high. It can be specified to be a high number. Furthermore, this evaluation value matches the type and position of oxygen and nitrogen atoms, which are likely to cause local electrostatic bias when collating the input compound or ligand with "ligand structure data". If is high, it can be specified to be a higher value. By defining the evaluation value as described above, the structural similarity between the input compound or ligand and the compound or ligand to which the candidate protein or candidate ligand binding site is associated can be evaluated more accurately. ..

Further, the evaluation value calculation unit 11 is a compound or ligand associated with a predetermined compound or ligand input as an analysis target by the data input unit 7 and a candidate protein or candidate ligand binding site extracted by the machine learning unit 10. It is preferable to calculate the evaluation value for the similarity of the structure to the structure of the region sufficiently close to the ligand binding site. Here, as a region sufficiently close to each other, for example, a region within 5 Å from the ligand binding site in a state where the compound or the ligand interacts with the ligand binding site can be mentioned. By this processing, the evaluation value is mutual with the predetermined compound or ligand input as the analysis target in the data input unit 7 and the compound or ligand associated with the candidate protein or candidate ligand binding unit extracted by the machine learning unit 10. The similarity with the region involved in the action can be evaluated.

Further, when the data input unit 7 inputs the type of enzyme reaction from the substrate to the product and the name of the enzyme involved in the enzyme reaction in addition to the predetermined compound or ligand as the analysis target, the evaluation value calculation unit In No. 11, it is possible to give an evaluation value indicating the degree of agreement or similarity with the input enzyme reaction or enzyme name for the extracted candidate protein or candidate ligand binding site.

Alternatively, when the evaluation value calculation unit 11 inputs a predetermined protein or ligand binding site as an analysis target in the data input unit 7, the protein or ligand binding site extracted as a result of machine learning in the machine learning unit 10 The evaluation value is calculated for the candidate compound or the candidate ligand that may interact with the protein. This evaluation value is a value indicating the similarity between the protein or ligand binding site input by the data input unit 7 and the protein or ligand binding site to which the extracted candidate compound or candidate ligand is associated.

Specifically, in the evaluation value calculation unit 11, the degree of coincidence of the amino acid sequence between the protein or ligand binding site input in the data input unit 7 and the protein or ligand binding site to which the extracted candidate compound or candidate ligand is associated. Can be calculated and a high evaluation value can be given to the candidate compound or the candidate ligand associated with the protein having a high degree of agreement or the ligand binding site. Further, in the evaluation value calculation unit 11, the degree of conformity in terms of three-dimensional structure between the protein or ligand binding site input by the data input unit 7 and the protein or ligand binding site to which the extracted candidate compound or candidate ligand is associated. Can be calculated and a high evaluation value can be given to a candidate compound or a candidate ligand associated with the protein having a high similarity or a ligand binding site. Further, in the evaluation value calculation unit 11, the electrostatic potential distribution is similar between the protein or ligand binding site input by the data input unit 7 and the protein or ligand binding site to which the extracted candidate compound or candidate ligand is associated. The degree can be calculated and a high evaluation value can be given to the candidate compound or the candidate ligand associated with the protein having a high similarity or the ligand binding site. By defining the evaluation value as described above, the structural similarity between the input protein or ligand binding site and the protein or ligand binding site to which the candidate compound or the candidate ligand is associated, and the similarity of the electrostatic potential distribution Gender can be evaluated more accurately.

As described above, the list generation unit 12 generates a list in which the data related to the protein interaction generated by the machine learning unit 10 and the evaluation value calculated by the evaluation value calculation unit 11 are integrated. When a predetermined compound is input as an analysis target in the data input unit 7, a list of candidate proteins or candidate ligand binding sites that may interact with the compound and the evaluation values calculated for these is generated. .. Alternatively, when a predetermined protein or ligand binding site is input as an analysis target in the data input unit 7, candidate compounds or candidate ligands that may interact with the protein or ligand binding site and the evaluation calculated for these. Generate a list of associated values.

In the example shown in FIG. 2, the calculation processing unit 8 includes a machine learning unit 10, an evaluation value calculation unit 11, and a list generation unit 12, but as shown in FIG. 3, further protein- It may be provided with a ligand compatibility score calculation unit 13. When the analysis target input by the data input unit 7 is a predetermined compound or ligand, the protein-ligand compatibility score calculation unit 13 interacts with the compound or ligand extracted by the machine learning unit 10. Calculate the compatibility score for the binding stability of a possible candidate protein or candidate ligand binding site to the compound or ligand to be analyzed. Alternatively, when the analysis target input by the data input unit 7 is a predetermined protein or ligand binding site, the protein-ligand compatibility score calculation unit 13 applies to the protein or ligand binding site extracted by the machine learning unit 10. The compatibility score for the binding stability of the candidate compound or the candidate ligand that may interact with the protein or the ligand binding site to be analyzed is calculated.

Here, the compatibility score can be a value calculated based on the binding enthalpy between the ligand and the ligand binding site. The ligand alone is in a state in which the water molecule is coordinated, and the potential energy 1 of the ligand alone is calculated from the bond enthalpy with the water molecule. Next, the potential energy 2 is calculated by calculating the amount of enthalpy in a state where the ligand binding site and the ligand are bound (ionic bond, hydrophobic bond, etc.). When the difference between the potential energy 2 and the potential energy 1 is positive, it means that the ligand is likely to bind to the ligand binding site. Therefore, the above-mentioned bond stability can be quantitatively evaluated by calculating the compatibility score in consideration of the difference between the potential energy 2 and the potential energy 1.

In the example shown in FIG. 3, as described above, the list generation unit 12 calculates the data on the protein interaction generated by the machine learning unit 10, the evaluation value calculated by the evaluation value calculation unit 11, and the protein-ligand compatibility score calculation. A list that integrates the suitability scores calculated in Part 13 is generated. When a predetermined compound or ligand is input as an analysis target in the data input unit 7, the candidate protein or candidate ligand binding site that may interact with the compound or ligand and the evaluation value calculated for these and the compatibility with the candidate protein or the candidate ligand binding site. Generate a list associated with the score. Alternatively, when a predetermined protein or ligand binding site is input as an analysis target in the data input unit 7, candidate compounds or candidate ligands that may interact with the protein or ligand binding site and the evaluation calculated for these. Generate a list of values associated with the fitness score.

As described above, the list generation unit 12 shown in FIG. 2 or 3 generates a list of candidate proteins or candidate ligand binding sites, or candidate compounds or candidate ligands extracted by the machine learning unit 10. At this time, the list generation unit 12 further adds a candidate protein or a candidate ligand binding site, or a candidate compound or a candidate ligand included in the list extracted by the machine learning unit 10 based on the above-mentioned evaluation value and / or compatibility score. It may be limited. That is, the list generation unit 12 has an evaluation value and / or a compatibility score of a predetermined value or less among the candidate proteins or candidate ligand binding sites, or the candidate compounds or candidate ligands included in the list extracted by the machine learning unit 10. You may remove things from the list.

Specifically, when the information regarding the analysis target input by the data input unit 7 is information about a compound that serves as a substrate in a predetermined enzyme reaction, the machine learning unit 10 selects a candidate enzyme that may use the compound as a substrate. Extract. In this case, the list generation unit 12 may exclude from the list the candidate enzymes whose evaluation value and / or compatibility score are equal to or less than a predetermined value among the candidate enzymes extracted by the machine learning unit 10. Further, in this case, the list generation unit 12 may exclude from the list the candidate enzymes extracted by the machine learning unit 10 that are not related to the enzyme reaction input by the user.

Then, the output unit 9 of the protein interaction analysis device 1 outputs the list generated by the list generation unit 12. Here, the list output by the output unit 9 may be the list generated by the list generation unit 12 as it is, or may be a list generated by the list generation unit 12 with further information added.

For example, when a predetermined compound is input as an analysis target in the data input unit 7, a list including a candidate protein or a candidate ligand binding site that may interact with the compound is generated in the list generation unit 12. However, it is possible to output a list to which the candidate proteins included in this list, functional information on the proteins including the candidate ligand binding site, and the like are added.

Further, although not shown in FIGS. 1 to 3, a process of analyzing engineering information based on the above-mentioned evaluation value may be performed before the output unit 9 outputs the list. For example, when a predetermined compound is input as an analysis target in the data input unit 7, the cause of inhibition of the interaction between the candidate protein or the candidate ligand binding site and the input compound is identified based on the evaluation value. , Analyze engineering information that facilitates the interaction of the above compounds.

Specifically, first, a region that contributes to a decrease in the evaluation value of the input compound is specified from a structural comparison between the input compound and the compound associated with the candidate protein or the candidate ligand binding site. Next, in the candidate protein or candidate ligand binding site, the position where the region contributing to the decrease in the evaluation value of the compound interacts is specified. Next, based on the three-dimensional structure and electrostatic potential distribution at the specified position, the input compound is introduced into the candidate protein or candidate ligand binding site so as to have a three-dimensional structure and electrostatic potential distribution that facilitate interaction. Identify mutations and modifications. Mutations and modifications thus identified can be generated as engineering information for candidate proteins or candidate ligand binding sites.

On the other hand, when a predetermined protein is input as an analysis target in the data input unit 7, the cause of inhibition of the interaction between the candidate compound or the candidate ligand and the input protein is identified based on the evaluation value, and the above It is also possible to analyze engineering information that facilitates protein interaction.

Specifically, first, a region of the input protein that contributes to a decrease in the evaluation value is specified from a structural comparison between the input protein and the protein associated with the candidate compound or the candidate ligand. Next, in the candidate compound or the candidate ligand, the position where the region contributing to the decrease in the evaluation value of the protein interacts is specified. Next, based on the three-dimensional structure and electrostatic potential distribution at the specified position, the structural modification (functionality) for the candidate compound or candidate ligand is made so that the input protein has a three-dimensional structure or electrostatic potential distribution that facilitates interaction. Identify removals, changes and additions of groups, etc.). The structural modification thus identified can be generated as engineering information for a candidate compound or a candidate ligand.

The protein interaction analysis device 1 described above can also be realized by a general computer device. That is, the protein interaction analysis device 1 includes an arithmetic unit such as a CPU, a storage device such as a hard disk, RAM and ROM, an input device such as a keyboard and a pointing device, and an output device such as a display and a printer. .. The protein interaction analysis device 1 may include, for example, a communication device for connecting an external storage device such as an external storage unit 2 via a network such as the Internet. In the protein interaction analysis device 1, this communication device functions as an input device for various data and an output device to the outside. In the protein interaction analysis device 1, a storage device such as a hard disk, RAM, and ROM stores a program for causing a computer device to perform the above-mentioned various processes. That is, the protein interaction analysis device 1 can be realized by executing the program stored in the storage device with the hardware described above. The protein interaction analysis device 1 may be composed of one computer device, or may be composed of a plurality of computer devices that are physically different but can communicate with each other.

All publications, patents and patent applications cited herein are incorporated herein by reference in their entirety.

Claims (20)

A data input unit for inputting information about the analysis target,
The surface shape of the ligand binding site for a given protein, generated based on the amino acid sequence data and three-dimensional structure data of the protein stored in the external storage unit and the three-dimensional structure data of the ligant that specifically interacts with the protein. A data storage unit that stores data, electrostatic potential distribution data of the ligand binding site, and three-dimensional structure data of a ligand that interacts with the ligand binding site in association with each other.
The surface shape data of a predetermined ligand binding site, the electrostatic potential distribution data of the ligand binding site, and the three-dimensional structure data of the ligand interacting with the ligand binding site, which are stored in the data storage unit, are used as teacher data. A protein interaction analysis apparatus including a calculation processing unit that generates data related to protein interactions related to an analysis target input by the data input unit by the above-mentioned machine learning. The information regarding the analysis target is information regarding the structure of the ligand.
The protein interaction analysis apparatus according to claim 1, wherein the calculation processing unit generates data on a protein or a ligand binding site that interacts with the ligand. The information regarding the analysis target is information regarding the structure of the protein or ligand binding site.
The protein interaction analysis apparatus according to claim 1, wherein the calculation processing unit generates data on a compound or ligand that interacts with the protein or ligand binding site. The calculation processing unit calculates an evaluation value indicating the similarity between the analysis target input by the data input unit and the analysis target included in the generated data for the data related to the protein interaction generated by machine learning. The protein interaction analysis apparatus according to claim 1, further comprising a calculation unit. The calculation processing unit calculates a compatibility score that quantitatively indicates the binding stability when the analysis target input in the data input unit interacts with respect to the data related to the protein interaction generated by machine learning. The protein interaction analysis apparatus according to claim 1, further comprising a compatibility score calculation unit. The data storage unit calculates the center coordinates of the ligand binding site based on the atomic coordinates of the atoms constituting the ligand binding site, sets a three-dimensional grid space including atoms within a predetermined distance from the center coordinates, and sets the three-dimensional grid space. The protein interaction analyzer according to claim 1, wherein the surface shape data generated based on the three-dimensional grid space is stored. The three-dimensional grid space has a plurality of grid points by a grid set at predetermined intervals, and gives a specific character to the closest grid point for each atom within a predetermined distance from the center coordinates. The protein-protein interaction analyzer according to claim 6, wherein the data is data in which another character is given to a lattice point to which the specific character is not given. The protein-protein interaction analysis apparatus according to claim 6, wherein each atom within a predetermined distance from the center coordinates is a plurality of non-hydrogen atom species. The protein-protein interaction analysis device according to claim 6, wherein the data storage unit stores electrostatic potential distribution data calculated for grid points in the three-dimensional grid space. The data storage unit is composed of positive electrostatic potential distribution data consisting of positive values calculated for the grid points in the three-dimensional grid space and negative values calculated for the grid points in the three-dimensional grid space. The protein interaction analyzer according to claim 6, wherein the electrostatic potential distribution data of the above is stored. The process of inputting information about the analysis target using an input device,
Based on the amino acid sequence data and three-dimensional structure data of the protein stored in the external storage unit and the three-dimensional structure data of the ligant that specifically interacts with the protein, the arithmetic unit determines the ligand binding site for a predetermined protein. The surface shape data, the electrostatic potential distribution data of the ligand binding site, and the three-dimensional structure data of the ligand interacting with the ligand binding site are generated, and these surface shape data, the electrostatic potential distribution data, and the three-dimensional structure of the ligand are generated. The process of associating data with the storage device and storing it in the storage device,
The arithmetic unit stores the surface shape data of a predetermined ligand binding site, the electrostatic potential distribution data of the ligand binding site, and the three-dimensional structure data of the ligand interacting with the ligand binding site stored in the storage device. A protein interaction analysis method comprising a step of generating data on a protein interaction related to an analysis target input by the input device by machine learning as teacher data. The information regarding the analysis target is information regarding the structure of the ligand.
The protein interaction analysis method according to claim 11, wherein the arithmetic unit generates data on a protein or a ligand binding site that interacts with the ligand. The information regarding the analysis target is information regarding the structure of the protein or ligand binding site.
The protein interaction analysis method according to claim 11, wherein the arithmetic unit generates data on a compound or ligand that interacts with the protein or ligand binding site. The arithmetic device has a step of calculating an evaluation value indicating the similarity between the analysis target input by the input device and the analysis target included in the generated data for the data related to the protein interaction generated by machine learning. 11. The protein interaction analysis method according to claim 11. The computing device has a step of calculating a suitability score that quantitatively indicates the binding stability when the analysis target input by the input device interacts with respect to the data related to the protein interaction generated by machine learning. The protein interaction analysis method according to claim 11, which is characterized. The arithmetic device calculates the center coordinates of the ligand binding site based on the atomic coordinates of the atoms constituting the ligand binding site, sets a three-dimensional grid space including atoms within a predetermined distance from the center coordinates, and sets the relevant The protein interaction analysis method according to claim 11, wherein the surface shape data generated based on the three-dimensional grid space is stored in the data storage unit. The three-dimensional grid space has a plurality of grid points by a grid set at predetermined intervals, and gives a specific character to the closest grid point for each atom within a predetermined distance from the center coordinates. The protein interaction analysis method according to claim 11, wherein the data is data in which another character is given to a lattice point to which the specific character is not given. The protein interaction analysis method according to claim 11, wherein each atom within a predetermined distance from the center coordinates is a plurality of non-hydrogen atom species. The protein interaction analysis method according to claim 11, wherein the arithmetic unit stores electrostatic potential distribution data calculated for grid points in the three-dimensional grid space in the data storage unit. The arithmetic device has a negative electrostatic potential distribution data consisting of positive values calculated for the grid points in the three-dimensional grid space and negative values calculated for the grid points in the three-dimensional grid space. The protein interaction analysis method according to claim 11, wherein the electrostatic potential distribution data is stored in the data storage unit.

Images