JP5170630B2 - Protein function identification device - Google Patents

Description

  The present invention relates to a technique for identifying protein functions based on the three-dimensional structure of a protein by bioinformatics.

  Conventionally, in the field of bioinformatics, attempts have been actively made to distinguish functions from protein structures. In this type of technology, the structure of a protein with unknown function is compared with the structure of a protein with unknown function in some way to identify the function of a protein with unknown function.

  The most widely used method at present is primary structure information, ie, sequence similarity search, with a protein of known function. In this method, the “sequence-affinity-function” correlation is utilized and an effort is made to detect the distant relationships as far as possible. However, there is a limit in principle to estimate the function from the distant relationship. The principle limit is caused by the omission of the three-dimensional structure in the first principle of protein function (sequence-stereostructure-function). In fact, only 50% of gene functions are estimated by sequence analysis alone, and the remaining gene functions remain unknown.

  As a function identification method using tertiary structure information, the method of Non-Patent Document 1 is known. This conventional technique proposes an approach for predicting a binding site by regarding the function of a protein as binding, that is, molecular recognition, which is the first step of expression of the function. This prior art discriminates functions assuming that proteins with similar shape, electrostatic potential and hydrophilicity / hydrophobicity in a certain molecular surface region bind similar small molecule ligands. Non-Patent Document 2 proposes a method of function identification by paying attention to a partial structure of a protein called a module.

  Here, many of proteins (1) take the interaction between proteins as a trigger, and (2) cause a chemical reaction between the protein and the protein or low molecular chemical substance. Non-Patent Document 1 regards the chemical reaction of (2) as a protein function, and does not identify the function based on the protein-protein interaction of (1). Non-patent document 2 also does not perform function identification based on the protein-protein interaction of (1). Non-Patent Document 2 refers to protein-protein interaction, but protein interaction cannot be used for function identification.

  Thus, the conventional function identification technique cannot reflect the protein-protein interaction. In other words, the interaction between a protein and another protein is not used to identify the function of a protein. In fact, although protein-protein interactions have been studied, there have been no successful examples of incorporating protein interactions into functional discrimination.

  In Non-Patent Documents 1 and 2, function identification is performed by paying attention to partial features of the three-dimensional structure. Therefore, when the structure of the part different from the target part is actually related to the function, the function cannot be accurately identified. Further, since it is assumed that the part related to the function is known in advance, the function cannot be identified when the part related to the function is not known.

  The current state of function identification using a three-dimensional structure is as described above. Although there is a great expectation for function identification by a three-dimensional structure, it cannot be said that sufficient identification ability is obtained at present, and improvement of identification ability is always desired.

  As another related technique, ZDOCK of Non-Patent Document 3 is known. ZDOCK is a technique for obtaining a complex structure in which two proteins are docked (coupled) from the three-dimensional structure of two proteins. ZDOCK evaluates the shape complementarity between the three protein conformations. That is, it is evaluated how well the surface irregularities of the two proteins are matched. According to ZDOCK, protein-protein interaction can be determined. However, ZDOCK does not provide a method for identifying the function of a protein.

Another related technique is Non-Patent Document 4. Non-Patent Document 4 provides a sample of a receptor and a ligand.
Kengo Kinoshita, Haruki Nakamura, Identification of protein biochemical functions by similarity search using the molecular surface database eF-site, Protein Science, Cold Spring Harbor Laboratory Press, 2003, 12: 1589-1595 K. Yura, M. Shionyu, K. Kawaktani, M Go, Repetitive use of a phosphate-binding module in DNA polymerase, Oct-1 POU domain and phage repressors, CMLS Cellular and Molecular Life Science, Birkhauser Verlag, Basel, 1999, 55: 472-486 Rong Chen, Zhiping Weng, A Novel Shape Complementarity Scoring Function for Protein-Protein Docking, PROTEINS: Structure, Function, and Genetics, WILEY-LISS, INC., 2003, 51: 397-408 Julian Mintseris, Kevin Wiehe, Brian Pierce, Robert Anderson, Rong Chen, Joel Janin, and Zhiping Weng, Protein-Protein Docking Benchmark 2.0: An Update, PROTEINS: Structure, Function, and Genetics, WILEY-LISS, INC., 2005, 60: 214-216

  The present invention has been made under the background described above, and an object of the present invention is to provide a technique capable of performing functional identification considering protein-protein interaction and capable of performing functional identification with high discrimination ability. It is in.

  One embodiment of the present invention is a protein function identification device that identifies the function of a protein with an unknown function based on the three-dimensional structure of a protein with a known function and the three-dimensional structure of a protein with an unknown function. A receptor storage unit for storing the three-dimensional structure of the receptor, a ligand storage unit for storing the three-dimensional structure of a plurality of ligands, a teacher for function identification based on the three-dimensional structure of the plurality of receptors and the three-dimensional structure of the plurality of ligands For each of a plurality of receptors whose functions are known, teacher data generation for calculating a plurality of shape complementarity evaluation values when each receptor is docked with each of the plurality of ligands and calculating charge information of each receptor Part, an unknown protein input part for inputting the three-dimensional structure of the unknown protein, and the unknown function Based on the three-dimensional structure of the protein and the three-dimensional structure of the plurality of ligands, a plurality of shape complementarity evaluation values when the unknown protein is docked with the plurality of ligands are calculated as identification input data for function identification. And an identification input data generation unit that calculates charge information of the function-unknown protein, and a learning identification unit that learns the teacher data and identifies the function of the function-unknown protein. Based on similarity of the plurality of shape complementarity evaluation values and the charge information of docking with the plurality of ligands in a plurality of common receptors, the plurality of shape complementarity evaluation values and the charge information are the functions unknown By determining the function of a receptor similar to a protein, the function of the unknown protein is identified.

  According to the present invention, teacher data is obtained from the three-dimensional structures of a plurality of receptors having known functions and the three-dimensional structures of a plurality of ligands as described above. The teacher data includes, for each receptor, a plurality of shape complementarity evaluation values when each receptor is docked with a plurality of ligands, and charge information of each receptor. In addition, identification input data is obtained from the three-dimensional structure of the unknown protein and the three-dimensional structure of a plurality of ligands. The identification input data includes a plurality of shape complementarity evaluation values when the unknown protein is docked with a plurality of ligands, respectively. And charge information of proteins of unknown function. And this invention learns said teacher data, and identifies the function of a function unknown protein. In this way, by using the shape complementarity evaluation value and the charge information, it is possible to perform function identification considering protein-protein interaction, and it is possible to perform function identification with high discrimination ability.

  The teacher data generation unit performs a docking simulation to match the unevenness of the surface of the compatible body with respect to the combination of the three-dimensional structure of each receptor and the three-dimensional structure of each ligand, and expresses how much the unevenness of the surface of the compatible body matches. A shape complementarity evaluation value may be calculated, and the identification input data generation unit performs the docking simulation on a combination of the three-dimensional structure of the function-unknown protein and the three-dimensional structure of each ligand to obtain the shape complementarity evaluation value. It may be calculated. According to the present invention, shape complementarity can be appropriately evaluated by performing a docking simulation.

  The teacher data generation unit and the identification input data generation unit may perform a process of specifying the protein surface from the three-dimensional structure of the protein before the docking simulation, and the thickness of the protein surface may be adjustable. According to the present invention, the surface thickness of the three-dimensional structure used in the docking simulation can be adjusted, and thereby the identification ability can be improved.

  The charge information calculated by the teacher data generation unit and the identification input data generation unit may include a difference between the total charges of the proteins that are the receptors or the proteins of unknown function and the total charges of the surface. According to the present invention, as described above, by using the charge information including the difference between the total charge of the entire protein and the total charge of the surface as the teacher data, the discrimination ability can be improved.

  The charge information calculated by the teacher data generation unit and the identification input data generation unit is the solvent exposure area of the protein that is each receptor or the function-unknown protein, the number of positive charge residues of the whole protein, The number of negatively charged residues of the whole protein, the number of histidine residues of the whole protein, the total charge of the whole protein, the number of positively charged residues of the protein surface, and the negative charge residue of the protein surface. It may include the number of groups, the number of histidine residues on the protein surface, the total charge on the protein surface, and the difference between the total charge on the whole protein and the total charge on the surface. According to the present invention, the identification capability can be improved by using the charge information including the various parameters as described above as the teacher data.

  The learning identification unit classifies the charge information and the shape complementarity evaluation value included in the teacher data into a plurality of functional categories by a receptor function, and the charge information of the unknown protein included in the identification input data and the It may be determined to which functional category the shape complementarity evaluation value belongs. According to the present invention, it is possible to appropriately learn information on a plurality of receptors with known functions, and it is possible to identify the function of a function-unknown protein on the basis of information on known receptors.

  The plurality of shape complementarity evaluation values of each receptor and the plurality of ligands and the charge information of each receptor constitute a receptor vector of each receptor, and the function-unknown protein and the plurality of ligands A plurality of shape complementarity evaluation values and the charge information of the function-unknown protein constitute an identification input vector of the function-unknown protein, and the learning identification unit includes a plurality of receptor vectors corresponding to the plurality of receptors. The separation plane divided into functional categories may be detected, and the functional category to which the identification input vector belongs is determined from the positional relationship between the separation plane and the identification input vector. According to the present invention, by using vector data, information on a plurality of receptors with known functions can be appropriately learned, and the function of a function-unknown protein can be identified based on the information on receptors with known functions. A support vector machine may be used for learning identification.

  The plurality of functional categories may be three categories of antibodies, enzymes, and other functions. According to the present invention, large-scale functions such as antibodies, enzymes, and other functions can be appropriately identified.

  Another aspect of the present invention is a protein function identification device for identifying the function of a protein with unknown function based on the three-dimensional structure of a protein with known function and the three-dimensional structure of a protein with unknown function, and comprising teacher data for function identification As for each of a plurality of receptors as proteins having a known function, information on the function of each receptor, a plurality of shape complementarity evaluation values when the three-dimensional structure of each receptor is respectively docked with a three-dimensional structure of a plurality of ligands, and A teacher data storage unit for storing charge information of each receptor, and a plurality of shape complementarity when the three-dimensional structure of the function-unknown protein is docked with the three-dimensional structure of the plurality of ligands as identification input data for function identification An identification input data generation unit that calculates an evaluation value and calculates charge information of the function-unknown protein A learning identification unit that learns the teacher data and identifies the function of the unknown protein, and the learning identification unit includes a plurality of docking with the plurality of ligands in a plurality of receptors having a common function. Based on the shape complementarity evaluation value and the similarity of the charge information, the function of the function unknown protein is obtained by obtaining the function of a receptor whose plurality of shape complementarity evaluation values and the charge information are similar to the function unknown protein. Identify. This aspect also provides the above-described advantages of the present invention.

  Another aspect of the present invention is a protein function identification method for identifying the function of a protein with unknown function based on the three-dimensional structure of a protein with known function and the three-dimensional structure of a protein with unknown function. The three-dimensional structure of each receptor is read from the receptor storage unit, the three-dimensional structure of a plurality of ligands is read from the ligand storage unit, and the teacher data for function identification based on the three-dimensional structure of the plurality of receptors and the three-dimensional structure of the plurality of ligands As for each of a plurality of receptors having known functions, a plurality of shape complementarity evaluation values are calculated when each receptor is docked with the plurality of ligands, and charge information of each receptor is calculated, and the function unknown protein The three-dimensional structure of the protein of unknown function and the plurality of regans Based on the three-dimensional structure, as the identification input data for function identification, a plurality of shape complementarity evaluation values when the unknown protein is docked with the plurality of ligands are calculated, and the charge information of the unknown protein is calculated. Calculating and learning learning to identify the function of the unknown protein by learning the teacher data, the learning identification is the plurality of shapes of docking with the plurality of ligands in a plurality of receptors having a common function Based on the complementarity evaluation value and the similarity of the charge information, the function of the function unknown protein is determined by obtaining the function of the receptor whose shape complementarity evaluation value and the charge information are similar to the function unknown protein. Identify. This aspect also provides the above-described advantages of the present invention.

  Another aspect of the present invention is a protein function identification program for causing a computer to execute a process of identifying a function of a protein whose function is unknown based on the three-dimensional structure of a protein whose function is known and the three-dimensional structure of a protein whose function is unknown. Reads the three-dimensional structure of a plurality of receptors as known proteins from the receptor storage unit, reads the three-dimensional structure of a plurality of ligands from the ligand storage unit, and functions based on the three-dimensional structure of the plurality of receptors and the three-dimensional structure of the plurality of ligands As teacher data for identification, for each of a plurality of receptors with known functions, a plurality of shape complementarity evaluation values are calculated when each receptor is docked with the plurality of ligands, and the charge information of each receptor is obtained. Calculate and input the three-dimensional structure of the function unknown protein, Based on the three-dimensional structure of the protein and the three-dimensional structure of the plurality of ligands, a plurality of shape complementarity evaluation values when the unknown protein is docked with the plurality of ligands are calculated as identification input data for function identification. And calculating the charge information of the function-unknown protein and learning the teacher data to identify the function of the function-unknown protein, and causing the computer to execute the process. The plurality of shape complementarity evaluation values and the charge information are based on the similarity of the plurality of shape complementarity evaluation values and the charge information of docking with the plurality of ligands in the plurality of receptors. The function of the unknown protein is identified by determining the function of the receptor similar to the above. This aspect also provides the above-described advantages of the present invention.

  According to the present invention, by identifying the function of a protein with unknown function using shape complementarity and charge information with a plurality of ligands as described above, it is possible to perform function identification in consideration of protein-protein interaction, Functional identification can be performed with high identification ability.

  DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments of the invention will be described with reference to the drawings.

  FIG. 1 shows a protein function identification device according to an embodiment of the present invention. The protein function identification device 1 is a computer device, a CPU that is an arithmetic device, a storage device such as a RAM and a ROM, an input device such as a keyboard and a pointing device, an output device such as a display and a printer, and a hard disk. An external storage device is provided. The protein function identification device 1 may have a communication function with a network, and this communication function may function as an information input / output device. In addition, the configuration for reading and writing data with respect to the external recording medium may also function as an input / output device. The storage device stores a program that causes a computer to perform various processes of the present invention, and the protein function identification device 1 is realized by executing the program. The protein function identification device 1 may be composed of one computer device or a plurality of computer devices, and they may be arranged in a distributed manner.

  The protein function identification device 1 is a device that generally identifies the function of a protein whose function is unknown based on the three-dimensional structure of a protein whose function is known and the three-dimensional structure of a protein whose function is unknown. Three-dimensional structure data of a plurality of receptors is used as information on proteins with known functions. Furthermore, three-dimensional structure data of a plurality of ligands is used. Receptors and ligands are specifically receptor proteins and ligand proteins, but are simply referred to herein as receptors and ligands. The protein function identification device 1 identifies the function from the three-dimensional structure of the unknown protein based on these three-dimensional structure data.

  As shown in FIG. 1, the protein function identification device 1 includes a receptor storage unit 3 that stores a plurality of receptor three-dimensional structures, a ligand storage unit 5 that stores a plurality of ligand three-dimensional structures, a plurality of receptors, and a plurality of ligands. Teacher data section 7 for generating teacher data from the three-dimensional structure data, unknown protein input section 9 for inputting the three-dimensional structure of the unknown protein, and identification input data from the three-dimensional structure of the unknown protein and the three-dimensional structure of a plurality of ligands An identification input data unit 11 that performs processing of the teacher data and the identification input data to identify the function of the unknown protein and an output unit 15 that outputs information on the identified function.

  As described above, the receptor storage unit 3 stores the three-dimensional structure data of a plurality of receptors whose functions are known, and the ligand storage unit 5 stores the three-dimensional structure data of a plurality of ligands. As for the ligand, data of a ligand whose function is known may be stored. As the three-dimensional structure data, data in PDB (Protein Data Bank) format is preferably used.

  The teacher data unit 7 includes a teacher data generation unit 21 and a teacher data storage unit 23. The teacher data generation unit 21 generates teacher data from the three-dimensional structure of the plurality of receptors in the receptor storage unit 3 and the three-dimensional structure of the plurality of ligands in the ligand storage unit 5, and the generated teacher data is stored in the teacher data storage unit 23. Is done. The following teacher data generation process will be described in detail.

  FIG. 2 shows a configuration of the teacher data generation unit 21. As shown in FIG. 2, the teacher data generation unit 21 includes a teacher docking evaluation unit 31, a teacher docking data storage unit 33, a teacher residue charge evaluation unit 35, and a teacher residue charge data storage unit 37. The teacher docking evaluation unit 31 calculates a shape complementarity evaluation value by performing a docking simulation for each combination formed from a plurality of receptors and a plurality of ligands, and the calculated shape complementarity evaluation value is used as the teacher docking data storage unit 33. To store. The teacher residue charge evaluation unit 35 calculates charge information for each of the plurality of receptors, and stores the calculated charge information in the teacher residue charge data storage unit 37.

  The teacher docking evaluation unit 31 includes grid conversion units 41 and 43, PSC conversion units 45 and 47, and a docking simulation unit 49 as illustrated.

  The three-dimensional structure of the receptor is input from the receptor storage unit 3 to the grid conversion unit 41. The grid conversion unit 41 converts the three-dimensional structure of the receptor expressed in the PDB format into grid data. Similarly, the three-dimensional structure of the ligand is input from the ligand storage unit 5 to the grid conversion unit 43, and the grid conversion unit 43 converts the three-dimensional structure of the ligand expressed in the PDB format into grid data. The data in the PDB format is composed of coordinate data of each atom in the protein. On the other hand, the grid data is composed of data of a large number of grid points at equal intervals, and each grid point is located in a structural attribute value (structurally, in which part of the protein (surface, internal, external)) Information). As the grid data, data in a Voxel expression format is preferably used. The Voxel expression will be described later. The converted grid data of each receptor is referred to as receptor grid data, and the converted grid data of each ligand is referred to as ligand grid data. The receptor grid data is input to the PSC converter 45, and the ligand grid data is input to the PSC converter 47.

  The PSC conversion units 45 and 47 convert the input grid data of the protein three-dimensional structure into a PSC expression format necessary for executing the docking simulation. PSC is an abbreviation for Pair wise Shape Complementarity contact surface score function defined in ZDOCK (see Non-Patent Document 3) developed by Weng et al. Of Boston University. The PSC converter 45 converts the grid data of each receptor into a PSC representation according to the PSC contact surface score function. Similarly, the PSC converter 47 converts the grid data of each ligand into a PSC expression according to the PSC contact surface score function. The PSC contact surface score function will be described later. The converted PSC data of each receptor is referred to as receptor PSC data, and the converted PSC data of each ligand is referred to as ligand PSC data. The receptor PSC data and the ligand PSC data are input to the docking simulation unit 49.

  The docking simulation unit 49 executes docking simulation calculation using the receptor PSC data and the ligand PSC data. In the docking simulation, the three-dimensional surface irregularities of two proteins are combined to calculate a shape complementarity evaluation value representing how much the two three-dimensional surface irregularities match. In the present embodiment, the contact surface score value Spsc is obtained as the shape complementarity evaluation value. Details of the docking simulation will be described later.

  The teacher docking evaluation unit 31 processes the above-described data, that is, data acquisition from the receptor storage unit 3 and the ligand storage unit 5, and processing in the grid conversion units 41 and 43, the PSC conversion units 45 and 47, and the docking simulation unit 49. Repeat several times. Thereby, a docking simulation is performed for each of a large number of combinations made up of a plurality of receptors and a plurality of ligands, and a shape complementarity evaluation value is obtained.

  The above-mentioned multiple times of processing is realized by calculation of inner and outer double loops. First, the inner loop is a loop for all the ligand data stored in the ligand storage unit 5. This loop focuses on one receptor and repeats the calculation by sequentially replacing the ligand. For example, when the number of data stored in the ligand storage unit 5 is 200, the inner loop is repeated 200 times. The resulting Spsc is also 200. Therefore, in this example, the configuration of one teacher docking data includes a receptor name and 200 Spsc.

  Next, the outer loop is a loop for all receptor data stored in the receptor storage unit 3. When the number of data stored in the receptor storage unit 3 is 100, the outer loop is repeated 100 times. Therefore, considering the outer and inner loops, the number of teacher docking data obtained is 200 × 100 = 20000.

  Note that the conversion to grid data and the conversion to PSC data may be performed only once for each receptor and each ligand. In this case, a plurality of receptor PSC data and a plurality of ligand PSC data are retained. Then, the docking simulation is performed on all combinations of the plurality of receptor PSC data and the plurality of ligand PSC data.

  FIG. 4 shows data obtained by the teacher docking evaluation unit 31. As shown, the Spsc of all combinations of multiple receptors and multiple ligands is obtained. In the illustrated example, the receptor and the ligand are numbered. In the figure, S (i, j) represents the Spsc of receptor i and ligand j. FIG. 4 corresponds to the above-described example, and 20000 Spsc is obtained from 100 receptors and 200 ligands.

  Next, generation of charge information by the teacher residue charge evaluation unit 35 will be described. The teacher residue charge evaluation unit 35 acquires the three-dimensional structure data of a plurality of receptors from the receptor storage unit 3. The following processing is performed on the three-dimensional structure data of each receptor to calculate charge information.

  The teacher residue charge evaluation unit 35 includes a molecular surface calculation unit 51, an amino acid residue counting unit 53, and a charge calculation unit 55. The molecular surface calculation unit 51 calculates the solvent exposed area and detects amino acid residues (surface exposed residues) exposed on the protein surface. Here, the surface exposed residue is detected by rolling a probe sphere having a size of about a water molecule along the protein surface and detecting an amino acid residue contacting the probe sphere. The area of the part (rolled part) in contact with the probe ball at the time of detection is obtained as the solvent exposure area. The amino acid residue counting unit 53 counts the number of positively charged residues, the number of negatively charged residues, and the number of histidine residues. Only the surface exposed residues determined in the solvent exposed area calculation are counted, and the number of charged residues in all the vacuum states is counted. A histidine residue is a residue whose polarity changes depending on pH. Here, the histidine residue is treated as neutral. From the receptor data and the processing results of the molecular surface calculation unit 51 and the amino acid residue counting unit 53, the charge calculation unit 55 calculates the total charge of the entire protein (receptor), the total charge of the protein surface, and the difference between these two charges. Ask. In this manner, the teacher residue charge evaluation unit 35 obtains the following charge information (E1) to (E10) in the vacuum state as shown in FIG. The obtained charge information is stored in the teacher residue charge data storage unit 37.

(E1) Solvent exposed area of protein (receptor)
(E2) Number of positively charged residues in the entire protein
(E3) Number of negatively charged residues in the entire protein
(E4) Number of histidine residues in the whole protein
(E5) Total charge of whole protein (positive charge + negative charge)
Calculate the number of positive and negative charges and the total charge in vacuum without taking into account the acid dissociation constant pKa. As mentioned above, histidine is assumed to be neutral and is not included in the charge calculation.
(E6) Number of positively charged residues on the protein surface
(E7) Number of negatively charged residues on the protein surface
(E8) Number of histidine residues on the protein surface
(E9) Total charge on the protein surface (positive charge + negative charge)
(E10) Difference between the total charge of the whole protein and the total charge on the protein surface (= E5-E9)

  As described above, in the teacher data generation unit 21, the teacher docking evaluation unit 31 performs docking simulation of the receptor and the ligand, and stores Spsc (shape complementarity evaluation value) of the simulation result in the teacher docking data storage unit 33. In addition, the teacher residue charge evaluation unit 35 calculates the charge information of each of the plurality of receptors and stores it in the teacher residue charge evaluation unit 35. These data are stored in the teacher data storage unit 23 as teacher data. Also, information on known functions of each of the plurality of receptors is read from the receptor storage unit 3 and stored in the teacher data storage unit 23 as part of the teacher data. In the example of the present embodiment, there are three types of functions of the receptor, such as an antibody, an enzyme, and the like. The teacher data storage unit 23 stores information on what type of function each receptor has.

  The teacher data unit 7 has been described above. Next, the identification input data unit 11 will be described. The identification input data 11 is configured to generate identification input data to be applied to the identification process based on the teacher data from the three-dimensional structure data of the function unknown protein. The teacher data section 7 and the identification input data section 11 have substantially the same configuration, but the data to be processed is different. That is, the teacher data unit 7 processes the three-dimensional structure data of a receptor having a known function obtained from the receptor storage unit 3, whereas the identification input data unit 11 has a function unknown protein input from the unknown protein input unit 9. 3D structure data is processed. In the following description, description of matters common to the teacher data unit 7 will be appropriately omitted.

  As shown in FIG. 1, the identification input data unit 11 includes an identification input data generation unit 25 and an identification input data storage unit 27. The identification input data generation unit 25 receives the three-dimensional structure data of the unknown function protein from the identification input data unit 11. This three-dimensional structure data is data in the PDB format, similar to the data in the receptor storage unit 3 and the ligand storage unit 5. The identification input data generation unit 25 generates identification input data from the three-dimensional structure of the unknown protein and the three-dimensional structures of the plurality of ligands in the ligand storage unit 5, and the generated identification input data is stored in the identification input data storage unit 27. The

  FIG. 3 shows a configuration of the identification input data generation unit 25. As shown in FIG. 3, the identification input data generation unit 25 includes an input docking evaluation unit 61, an input docking data storage unit 63, an input residue charge evaluation unit 65, and an input residue charge data storage unit 67. The input docking evaluation unit 61 calculates a shape complementarity evaluation value by performing a docking simulation on each combination made from a protein with unknown function and a plurality of ligands, and the calculated shape complementarity evaluation value is input to the docking data storage unit 63. To store. The input residue charge evaluation unit 65 calculates the charge information of the protein whose function is unknown, and stores the calculated charge information in the input residue charge data storage unit 67.

  The input docking evaluation unit 61 includes grid conversion units 71 and 73, PSC conversion units 75 and 77, and a docking simulation unit 79. The grid conversion unit 71 converts the three-dimensional structure data of the function unknown protein in PDB format into grid data. Again, Voxel representation data is generated. The grid data of this unknown protein is converted into PSC data by the PSC converter 75. In addition, the grid conversion unit 73 and the PSC conversion unit 77 convert the three-dimensional structure data of each ligand into grid data and further convert into PSC data, as in the generation of the teacher data.

  The docking simulation unit 79 performs docking simulation on the protein with unknown function and the ligand. The docking simulation process may be the same as the docking simulation process in generating teacher data. However, instead of receptor PSC data, PSC data of a protein with unknown function is docked with ligand PSC data. Then, the Spsc of the protein with unknown function and the ligand is obtained as the shape complementarity evaluation value.

  The input docking evaluation unit 61 obtains the Spsc between the protein with unknown function and each of the plurality of ligands by repeating the above process a plurality of times. A single loop calculation is preferably performed for this multiple simulations. This loop is a loop for all the ligands stored in the ligand storage unit 5, and is the same as the inner loop described in the teacher data generation process. When the number of ligand data stored in the ligand storage unit 5 is 200, the loop is repeated 200 times to obtain 200 Spsc. Then, 200 Spscs are stored in the input docking data storage unit 63 along with the names of the proteins with unknown functions.

  Next, generation of charge information by the input residue charge evaluation unit 65 will be described. The input residue charge evaluation unit 65 has the same configuration as the teacher residue charge evaluation unit 35. However, the teacher residue charge evaluation unit 35 processes the three-dimensional structure data of the receptor, whereas the input residue charge evaluation unit 65 processes the three-dimensional structure data of the function-unknown protein. Then, the input residue charge evaluation unit 65 calculates the data (E1) to (E10) described above as the charge information about the protein with unknown function (FIG. 5). The charge information is stored in the input residue charge data storage unit 67.

  The input residue charge evaluation unit 65 includes a molecular surface calculation unit 81, an amino acid residue counting unit 83, and a charge calculation unit 85, which include the molecular surface calculation unit 51 of the teacher residue charge evaluation unit 35, amino acids It corresponds to the residue counting unit 53 and the charge calculation unit 55, respectively.

  As described above, in the identification input data generation unit 25, the input docking evaluation unit 61 performs docking simulation between the unknown protein and the ligand, and stores the simulation result Spsc (shape complementation evaluation value) in the input docking data storage unit 63. Further, the input residue charge evaluation unit 65 calculates the charge information of the protein whose function is unknown and stores it in the input residue charge data storage unit 67. These data are stored in the identification input data storage unit 27 as identification input data.

  Next, the learning identification unit 13 will be described. The learning identifying unit 13 is configured to learn the teacher data and identify the function of the function unknown protein.

  FIG. 6 shows the concept of learning identification processing. The teacher data was the function of each of the plurality of receptors, the Spsc in each of all combinations of the plurality of receptors and the plurality of ligands, and the charge information of each of the plurality of receptors. In FIG. 6, the teacher data is organized as data for each receptor. Receptors and ligands are numbered. S (i, j) in the figure is Spsc of receptor i and ligand j. E1 (i) is the charge information E1 of the receptor i (the same applies to the charge information E2 to E10). As shown, each of the receptors 1, 2,... Has docking Spsc with all the ligands 1 to 200, and charge information E1 to E10. In addition, receptors are classified into a plurality of functional categories 1, 2, and 3 according to function (antibody, enzyme, etc.).

  On the other hand, the identification input data is data on a protein whose function is unknown as shown in the figure. In the figure, S (t, j) represents Spsc of the protein with unknown function and ligand j, and E1 (t) represents charge information E1 of the protein with unknown function. It is the same as the receptor data in that it has a plurality of Spsc and charge information.

  Here, the present invention pays attention to the fact that teacher data (that is, shape complementarity evaluation values (200 Spsc) and charge information (E1 to E10)) are similar in a plurality of receptors having a common function. And the learning identification part 13 identifies the function of a function unknown protein by calculating | requiring the function of the receptor in which several shape complementarity evaluation values and electric charge information are similar to a function unknown protein. That is, the learning identification unit 13 determines which group of the three functional categories has the most similar information on the unknown protein.

  The learning identification unit 13 classifies the charge information and the shape complementarity evaluation value included in the teacher data into a plurality of functional categories by the receptor function to realize the above processing, and the charge of the unknown protein included in the identification input data Processing is performed to determine to which functional category the information and shape complementarity evaluation values belong. The learning identification unit 13 is configured to perform so-called machine learning. Specifically, the processing of the learning identification unit 13 is realized using a support vector machine (hereinafter referred to as “SVM”), and a Radius Basis Function is used as a kernel at that time.

  FIG. 7 shows the concept of SVM. In SVM, a plurality of vectors are divided into a plurality of groups. And the separation surface between several groups is calculated | required. The separation plane is a plane that separates the vector space and is called a hyperplane.

  As shown in FIG. 7, in this embodiment, a plurality of receptor vectors respectively corresponding to a plurality of receptors are obtained from the teacher data. As illustrated, the receptor vector includes a receptor name, a plurality of Spsc, charge information, and a function. A plurality of Spsc corresponds to a plurality of ligands, respectively. The charge information is all information (E1) to (E10) obtained by the teacher residue charge evaluation unit 35 described above. The function is one of antibodies, enzymes, and other three types. In FIG. 7, as in FIG. 6, the receptor and the ligand are numbered, Spsc (i, j) is Spsc of the receptor i and the ligand j, and E1 (i) is the charge information E1 (i) of the receptor i. ). The example of FIG. 7 corresponds to the example of FIG. 6, where the number of receptors is 100 and the number of ligands is 200.

  As shown in FIG. 7, similar vector data can be obtained from the identification input data. This vector data is called identification input vector data. The identification input vector data includes a plurality of Spsc and charge information. However, since the function is unknown, the identification input vector data has no function information.

  First, a plurality of receptor vector data corresponding to a plurality of receptors are input to the SVM. The SVM detects separation surfaces that divide the plurality of receptor vector data into a plurality of functional categories (antibodies, enzymes, etc.). Here, in order to show the concept of SVM, “separation surface of antibody and enzyme” is shown. In practice, “antibody and other separation surfaces” and “enzyme and other separation surfaces” are also detected. These separation surfaces may be held in the storage device of the protein function identification device 1.

  After the separation plane is obtained as described above, the identification input vector data is input to the SVM. The SVM determines the function category to which the identification input vector belongs from the positional relationship between the separation plane and the identification input vector. A function corresponding to the discriminated function category is obtained as a function of the function unknown protein. Information on the function obtained by the learning identification unit 13 is output from the output unit 15 as an identification result.

  The configuration and processing of each part of the protein function identification device 1 have been described above. Next, grid data conversion, contact surface score, and docking simulation will be described in more detail.

"Grid data conversion (Voxel expression)"
As described above, in the present embodiment, the three-dimensional structure data of proteins (receptors, ligands and functionally unknown proteins) is converted into grid data before docking simulation. At this time, grid data is generated in a Voxel expression format as described below.

  The Voxel expression is performed by performing a conversion process called Voxel conversion on the protein three-dimensional structure in the PDB format. Voxelization means that a target protein is discretized into a three-dimensional l × m × n grid, and a SURFACE that is a structural attribute value of the protein for each grid point according to the geometric characteristics resulting from the shape of the protein. (S), INNERE (I), CAVITY (C), or OUTER (O) is assigned. The values of l, m and n must be large enough that the space created by the grid can cover the size of all proteins to be processed. In the example of the present embodiment, l = m = n = 64 is set. The grid interval is set to 1.2 angstroms.

  8 to 13 show the flow of Voxelization processing. The Voxel model Voxel (V) has two main features. One is that the radius of the probe sphere used to determine the Connolly surface is adaptively defined according to the atomic species in order to more accurately model the unevenness of the protein surface. In addition, by adding a bonus radius according to the size of the van der Waals radius of the atomic species, a more concavo-convex surface is determined than a protein surface with a constant value radius such as a water molecule that is widely used in general. can do. Another feature is that, in order to calculate the shape complementarity of two proteins more accurately, the surface thickness was changed at the final stage of voxelization to define the protein surface and the interior more precisely. It is. Determination of the size of the bonus radius and the thickness of the voxelized surface is preferably adjusted so that the final discrimination performance is maximized.

A Voxel model Voxel (V) is defined in Equation (1). Here, INNER (I) indicates a region where a protein molecule exists, and SURFACE (S) indicates a region on the protein surface. CAVITY (C) indicates the region inside the protein and where no protein molecule is present. OUTER (O) indicates a region where no protein molecule exists and is not CAVITY (C). However, DEFAULT (D) is an attribute value indicating the initial area of Voxel (V), and indicates an empty set that does not take any attribute value of I, O, S, and C. TMP-SURFACE (Stmp) indicates a region temporarily defined as the protein surface.

The Voxelization flow is formulated into Equation (2). 8 to 13 show details of the Voxelization flow for each step.

Procedure 1 (DEFAULT determination): An initial attribute value DEFAULT is assigned to all grid points of Voxel.

Procedure 2 (Determination of INNER): INNER is assigned to grid points included in a sphere having a radius Ri (= Rvi + Δri) centered on the space center of gravity of the atom. Here, Rvi is the van der Waals radius of each atom, and Δri is an increment of the radius defined for each atomic species.

Procedure 3 (Determination of OUTER): OUTER is assigned to grid points included in the area surrounded by the border of Voxel and INNER.

Procedure 4 (determination of TMP-SURFACE): TMP-SURFACE is temporarily assigned to INNER adjacent to OUTER. This TMP-SURFACE is a boundary layer between INNER and OUTER, which is a reference for determining the final SURFACE in the procedure 5.

Procedure 5 (Determination of SURFACE): SURFACE is allocated to TMP-SURFACE adjacent to OUTER that is equal to or greater than the threshold. Assign INNER to the remaining TMP-SURFACE. By changing this threshold, the thickness of the voxelized protein surface can be adjusted.

Procedure 6 (decision of CAVITY): CAVITY is allocated to the remaining DEFAULT.

"PSC contact surface score function"
Next, the PSC contact surface score function is used to convert Voxel-represented grid data into PSC data for simulation.

  The PSC contact surface score function is a function defined to evaluate shape complementarity between proteins, and defines a score scoring method for the above-mentioned Voxelized proteins (INNER, OUTER, SURFACE, CAVITY). Using this function, the contact surface score Spsc is calculated in a docking simulation described later. In the present embodiment, Pairwise Shape Complementarity (PSC) defined by ZDOCK (see Non-Patent Document 3) is employed as the contact surface score function. Hereinafter, the PSC contact surface score function will be further described. The PSC contact surface score function according to the present embodiment follows Non-Patent Document 3.

FIG. 14 shows an overview of the PSC contact surface score function, where the receptor and ligand grid points are scored by the contact surface score function. Equation (3) shows the PSC contact surface score function that gives the score of FIG. Formula (3) is characterized in that the scoring method differs between the receptor and the ligand. In addition, a strategy is adopted in which a high score is arranged around the receptor pocket so as to induce a ligand into the receptor pocket structure.

  The contact surface score function of Formula (3) will be described in detail. Rpsc and Lpsc in the equation are the contact surface score functions of the receptor and the ligand, respectively. Re [] and Im [] indicate the real part and imaginary part of the complex function, respectively. Equation 3 defines the real and imaginary parts of the receptor contact surface score function Rpsc and the imaginary part of the ligand contact surface score function Lpsc.

  In Equation (3), regarding the real part of the contact surface score function Rpsc of the receptor, a parameter called bonusR is introduced in order to effectively induce the convex surface of the ligand with respect to the pocket of the receptor as described above. This bonusR is used to determine the score of OUTER. Specifically, the number of SURFACE grid points within the distance range obtained by adding bonusR to the van der Waals radius is given as a score to the OUTER grid points. In this process, a score of 1 or more is given to the outer near SURFACE, but the score of the outer far from SURFACE is zero. Therefore, a score of 1 or more is given to OUTER near the SURFACE (outer peripheral layer near the surface). Further, since there are many SURFACEs near the OUTER grid points in the pocket, the score is larger in the pocket. Thereby, the convex surface of the ligand can be effectively guided to the pocket by the following docking simulation. In the example of FIG. 14, the score at the bottom of the pocket is 5.

  In the expression (3), 0 is given to the part other than the OUTER as a real part of the contact surface score function Rpsc of the receptor. Further, when looking at the real part of the contact surface score function Lpsc of the ligand, 1 is given to SURFACE and INNER, and 0 is given to the other parts. The imaginary part is common to the receptor and the ligand, 3i is given to SURFACE (surface), 9i is given to INNER and CAVITY (inside), and OUTER (outside) has no imaginary part.

  In the protein function identification device 1, when grid data (Voxel data) is converted into PSC data, scores are assigned to all grid points according to the contact surface score function of Equation (3). That is, at each grid point of the receptor ligand, the real part and the imaginary part are determined according to Equation (3).

"Docking simulation"
Next, docking simulation using the contact surface score function will be described. As described above, scores are assigned to all grid points of grid data (Voxel data) by the contact surface score function, and converted to PSC data. In the docking simulation, Spsc (contact surface score value) is calculated from the PSC data of the receptor and the PSC data of the ligand. In the calculation of Spsc, the receptor and the PSC data of the receptor are overlapped in space, and the scores of the overlapping grid points are multiplied together. The sum of the product of scores for all pairs of grid points is determined. Spsc is an evaluation value indicating how much the surface irregularities match when the receptor and the ligand are docked, and is a kind of shape complementarity evaluation value of the present invention.

  In the docking simulation, the maximum value of Spsc when the positional relationship between the receptor and the ligand is changed little by little is searched for. Thereby, the positional relationship and Spsc when the concave and convex portions of the receptor and the ligand are best matched are obtained. The search algorithm comprehensively calculates the entire space of rotation and translation of the ligand relative to the receptor. When the increment of the rotation angle is determined, the rotation space is determined. For example, the increment of the rotation angle is set to 15 °. In this case, the number of rotating bodies of the ligand is 3600. As the search algorithm, FFT is preferably used for high-speed search of the translation space between the receptor and the rotating body of the ligand.

Equation (4) shows Spsc. Once scores have been assigned to all grid points in the Voxel according to equation (3) above, Spsc can be calculated using equation (4).

  Here, o, p, and q in Equation (4) are the number of grid points at which the ligand moves relative to the receptor. For example, when the receptor and the ligand do not contact, the contact surface score Spsc becomes zero. In the case of contact between SURFACE and SURFACE, the receptor OUTER and the ligand SURFACE overlap, and Spsc is 1 (= 1 × 1) (imaginary number is ignored). As shown in FIG. 14, in the pocket or the like, the score of OUTER of the receptor is larger than 1, thereby causing Spsc to be larger than 1. Further, in the case of contact between INNER and INNER, −81 (= 9i × 9i) is given as a penalty with respect to the collision between the physically prohibited insides. Even when SURFACE and SURFACE overlap, -9 (= 3i × 3i) is given as a penalty.

Spsc can be calculated finally using the correlation function of Rpsc and Lpsc expressed by Equation (5). Executing this calculation is the execution of the docking simulation, and Spsc is the output value.

"Application example of this embodiment"
Next, a specific example of the identification process performed by the protein function identification device 1 of the present embodiment will be described.

About the sample
Figures 15 and 16 show samples of receptor and ligand, respectively. In this example, 52 receptors and 50 (including two duplicates) ligands were used as teaching data and input data as samples. All samples were extracted from Protein-Protein Docking Benchmark 2.0 of Weng et al., Boston University. Not only this benchmark data set, but also a data set of a protein pair consisting of a receptor and a ligand, there are two data classifications “bound” and “unbound” depending on the collection state of the three-dimensional structure by experiment. “bound” refers to the three-dimensional structure data obtained in a state where the protein pair is bound, and “unbound” refers to the three-dimensional structure data obtained individually in a state where the protein pair is not bound. From this classification point of view, most data registered in PDB (Protein Data Bank) is classified as unbound. In this example, from the above benchmark data set, the biochemical functions (antibody-antigen, enzyme-inhibitor, enzyme-substrate, etc.) are less biased in consideration of the difficulty defined by Weng et al. A protein pair was used as a sample (FIGS. 15 and 16). The breakdown of this protein pair is as follows. 52 receptors consist of 12 “antibodies”, 19 “enzymes” and 21 “others”. 50 of the ligands consist of 11 (including 2 overlaps) “antigens”, 19 “substrates or inhibitors” and 20 “others”.

“SVM conditions”
The SVM conditions are shown in FIG. LIBSVM was used for the implementation, and Default was used for all kernels and parameters.

"Identification result"
18 and 19 show the evaluation results of Spsc in the sample. FIG. 18 is the Spsc for one receptor. That is, the Spsc of one receptor and 50 ligands is shown. FIG. 19 is the Spsc for all 52 receptors. Each line in FIG. 19 shows the Spsc of one receptor and 50 ligands. 18 and 19 show the normalized Spsc values for the receptor ligands. Normalized Spsc is Spsc when the maximum value of the Spsc group (Spsc group of the number of ligands) obtained for each receptor is normalized to 1. 18 and 19, the horizontal axis represents Ligand, and the vertical axis represents normalized Spsc. 18 and 19 show only Spsc. In the protein function identification apparatus 1, in addition to Spsc in FIGS. 18 and 19, charge information is input to the SVM to perform function identification.

  FIG. 20 shows the identification results in the form of a table. In this example, when the prevalence is 33%, the protein function identification apparatus 1 has an accuracy of diagnosis = 92%, sensitivity = 88%, specificity = 94%, recall = 88%, precision. = 88% of identification results could be obtained.

  According to the above identification result, it is possible to identify protein functions in a large classification (antibody, enzyme, etc.). This large classification is a biological function. On the other hand, each function of each protein belonging to a large classification can be said to be a biochemical function. The above identification result can provide an experimental strategy for identifying a protein with unknown function to an experimental researcher in the field of biochemistry. Furthermore, the above identification results can greatly reduce the search space for biochemical function identification. For example, in technologies such as Non-Patent Documents 1 and 2, since a large classification of protein functions is unknown, a minute feature amount is searched for in a huge search space. On the other hand, if a large classification is known, the search space can be greatly reduced, thereby improving search efficiency and accuracy.

  Regarding the function classification, the following can also be said. In the present embodiment, as described above, protein functions in a large classification are identified. In each of these large classifications, the application of the present invention may identify functions in the smaller classifications. Within each of the small classifications, three-dimensional structures of a plurality of proteins with known functions are prepared, and the above-described present invention is applied. Further, the function may be identified by a fine classification. In this way, more detailed function identification is possible by performing the function identification of the present invention in a plurality of stages. Thus, it is possible to apply the present invention not only to the above-mentioned large-scale biological functions but also to the identification of smaller-scale biochemical functions.

  In the present embodiment, the following measures are taken for the processing of the shape complementarity evaluation value and the charge information, and an advantageous effect can be obtained.

  First, the processing of the present invention is organized in comparison with the prior art. Both shape complementarity and charge information play important roles in protein function. Conventionally, one integrated score using both shape complementarity and charge information as parameters is calculated. In other words, this integrated score was a function of shape complementarity and charge information.

  On the other hand, the docking simulation of the present invention obtains only Spsc that is a shape complementarity evaluation value. The charge information is processed separately from the docking simulation result. The charge information is input to the SVM together with the shape complementarity evaluation value of the docking simulation result. However, until the end, the charge information is not a parameter for the docking simulation.

  In this regard, ZDOCK is in common with the present invention in the principle of shape complementarity evaluation. However, ZDOCK also obtains an integrated score by a function of shape complementarity and charge information, and this point is different from the present invention.

  The configuration of the present invention described above is advantageous in the following points. The prior art seeks an integrated score that is a function of both shape complementarity and charge information. This integrated score is based on the assumption that the data of docking proteins (receptors and ligands) is complete, that is, there are no missing parts. If a part of the protein is missing due to the problem of experimental accuracy, the integrated score is greatly affected and the accuracy is greatly reduced.

  On the other hand, in the present invention, the shape complementarity evaluation value and the charge information are handled separately. In this case, the effect of missing remains in the shape complementarity evaluation value. The charge information is processed separately and is not easily affected by the loss. Thereby, the influence of the missing part on the final identification result is reduced, and higher identification accuracy is obtained.

  In the present invention, the charge information E1 to E10 is used. Among these, the charge information of E10 is important. E10 is the difference between the total charge of the whole protein and the total charge on the protein surface (= E5-E9). By inputting E10 as a vector element in addition to E1 to E9 to the SVM, the discrimination ability is improved.

  The preferred embodiments of the present invention have been described above. As described above, according to the present invention, 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 includes, for each receptor, a plurality of shape complementarity evaluation values when each receptor is docked with a plurality of ligands, and charge information of each receptor. In addition, identification input data is obtained from the three-dimensional structure of the unknown protein and the three-dimensional structure of a plurality of ligands. The identification input data includes a plurality of shape complementarity evaluation values when the unknown protein is docked with a plurality of ligands, respectively. And charge information of proteins of unknown function. And this invention learns said teacher data, and identifies the function of a function unknown protein. In this way, by using the shape complementarity evaluation value and the charge information, it is possible to perform function identification considering protein-protein interaction, and it is possible to perform function identification with high discrimination ability.

  The preferred embodiments of the present invention have been described above. However, the present invention is not limited to the above-described embodiments, and it goes without saying that those skilled in the art can modify the above-described embodiments within the scope of the present invention.

  As described above, the protein function identification device according to the present invention is useful as a technique for identifying protein functions based on the three-dimensional structure of the protein by bioinformatics.

It is a figure which shows the protein function identification device which concerns on embodiment of this invention. It is a figure which shows the structure of a teacher data production | generation part. It is a figure which shows the structure of an identification input data generation part. It is a figure which shows the data of the shape complementarity evaluation value obtained in a teacher docking evaluation part. It is a figure which shows charge information. It is a figure which shows learning identification processing notionally. It is a figure which shows the learning identification process using a support vector machine. It is a figure which shows the flow of a Voxelization process, Comprising: It is a figure which shows the allocation process of DEFULT. It is a figure which shows the flow of a Voxelization process, Comprising: It is a figure which shows the allocation process of INNER. It is a figure which shows the flow of a Voxelization process, Comprising: It is a figure which shows the assigning process of OUTER. It is a figure which shows the flow of a Voxelization process, Comprising: It is a figure which shows the allocation process of TMP-SURFACE. It is a figure which shows the flow of a Voxelization process, Comprising: It is a figure which shows the allocation process of SURFACE. It is a figure which shows the flow of a Voxelization process, Comprising: It is a figure which shows the allocation process of CAVITY. It is a figure for demonstrating a PSC contact surface score function. It is a figure which shows the sample of the receptor in a specific example. It is a figure which shows the sample of the ligand in a specific example. It is a figure which shows the processing conditions of the support vector machine in a specific example. It is a figure which shows the evaluation result of Spsc in a sample. It is a figure which shows the evaluation result of Spsc in a sample. It is a figure which shows the example of the identification result by this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Protein function identification device 3 Receptor memory | storage part 5 Ligand memory | storage part 7 Teacher data part 9 Unknown protein input part 11 Identification input data part 13 Learning identification part 15 Output part 21 Teacher data generation part 23 Teacher data storage part 25 Identification input data generation part 27 Identification Input Data Storage Unit 31 Teacher Docking Evaluation Unit 33 Teacher Docking Data Storage Unit 35 Teacher Residue Charge Evaluation Unit 37 Teacher Residue Charge Data Storage Unit 41, 43, 71, 73 Grid Conversion Units 45, 47, 75, 77 PSC Conversion unit 49, 79 Docking simulation unit 51, 81 Molecular surface calculation unit 53, 83 Amino acid residue counting unit 55, 85 Charge calculation unit 61 Input docking evaluation unit 63 Input docking data storage unit 65 Input residue charge evaluation unit 67 Input residue Base charge data storage

Claims (9)

A protein function identification device for identifying the function of a protein with unknown function based on the three-dimensional structure of a protein with known function and the three-dimensional structure of a protein with unknown function,
A receptor storage unit that stores the three-dimensional structure of a plurality of receptors as proteins of known function;
A ligand storage unit that stores a three-dimensional structure of a plurality of ligands;
When each receptor docks with the plurality of ligands for each of the plurality of receptors whose functions are known as teacher data for function identification based on the three-dimensional structures of the plurality of receptors and the three-dimensional structures of the plurality of ligands. A teacher data generation unit for calculating a plurality of shape complementarity evaluation values and calculating charge information of each receptor;
An unknown protein input section for inputting the three-dimensional structure of the function unknown protein;
Based on the three-dimensional structure of the unknown protein and the three-dimensional structure of the plurality of ligands, a plurality of shape complementarity evaluation values when the unknown protein is docked with the plurality of ligands as identification input data for function identification And an identification input data generation unit for calculating the charge information of the protein with unknown function,
A learning discriminating unit that learns the teacher data and identifies the function of the function-unknown protein, and the learning discriminating unit includes the plurality of shapes of docking with the plurality of ligands in a plurality of receptors having a common function. Based on the complementarity evaluation value and the similarity of the charge information, the function of the function unknown protein is determined by obtaining the function of the receptor whose shape complementarity evaluation value and the charge information are similar to the function unknown protein. Identify and
The teacher data generation unit performs a docking simulation to match the unevenness of the surface of the compatible body with respect to the combination of the three-dimensional structure of each receptor and the three-dimensional structure of each ligand, and expresses how much the unevenness of the surface of the compatible body matches. A shape complementarity evaluation value is calculated, and the identification input data generation unit calculates the shape complementarity evaluation value by performing the docking simulation on a combination of the three-dimensional structure of the unknown protein and the three-dimensional structure of each ligand. ,
The teacher data generation unit and the identification input data generation unit perform a process of specifying the protein surface from the three-dimensional structure of the protein before the docking simulation, and the thickness of the protein surface can be adjusted.
The charge information calculated by the teacher data generation unit and the identification input data generation unit includes a difference between the total charge of the protein that is each receptor or the protein of unknown function and the total charge of the surface. protein function identification device to. A protein function identification device for identifying the function of a protein with unknown function based on the three-dimensional structure of a protein with known function and the three-dimensional structure of a protein with unknown function,
A receptor storage unit that stores the three-dimensional structure of a plurality of receptors as proteins of known function;
A ligand storage unit that stores a three-dimensional structure of a plurality of ligands;
When each receptor docks with the plurality of ligands for each of the plurality of receptors whose functions are known as teacher data for function identification based on the three-dimensional structures of the plurality of receptors and the three-dimensional structures of the plurality of ligands. A teacher data generation unit for calculating a plurality of shape complementarity evaluation values and calculating charge information of each receptor;
An unknown protein input section for inputting the three-dimensional structure of the function unknown protein;
Based on the three-dimensional structure of the unknown protein and the three-dimensional structure of the plurality of ligands, a plurality of shape complementarity evaluation values when the unknown protein is docked with the plurality of ligands as identification input data for function identification And an identification input data generation unit for calculating the charge information of the protein with unknown function,
A learning discriminating unit that learns the teacher data and identifies the function of the function-unknown protein, and the learning discriminating unit includes the plurality of shapes of docking with the plurality of ligands in a plurality of receptors having a common function. Based on the complementarity evaluation value and the similarity of the charge information, the function of the function unknown protein is determined by obtaining the function of the receptor whose shape complementarity evaluation value and the charge information are similar to the function unknown protein. Identify and
The teacher data generation unit performs a docking simulation to match the unevenness of the surface of the compatible body with respect to the combination of the three-dimensional structure of each receptor and the three-dimensional structure of each ligand, and expresses how much the unevenness of the surface of the compatible body matches. A shape complementarity evaluation value is calculated, and the identification input data generation unit calculates the shape complementarity evaluation value by performing the docking simulation on a combination of the three-dimensional structure of the unknown protein and the three-dimensional structure of each ligand. ,
The teacher data generation unit and the identification input data generation unit perform a process of specifying the protein surface from the three-dimensional structure of the protein before the docking simulation, and the thickness of the protein surface can be adjusted.
The charge information calculated by the teacher data generation unit and the identification input data generation unit is the solvent exposure area of the protein that is each receptor or the function-unknown protein, the number of positive charge residues of the whole protein, The number of negatively charged residues of the whole protein, the number of histidine residues of the whole protein, the total charge of the whole protein, the number of positively charged residues of the protein surface, and the negative charge residue of the protein surface. A protein function identification apparatus comprising: the number of groups, the number of histidine residues on the protein surface, the total charge on the protein surface, and the difference between the total charge on the entire protein and the total charge on the surface . The learning identification unit classifies the charge information and the shape complementarity evaluation value included in the teacher data into a plurality of functional categories by a receptor function, and the charge information of the unknown protein included in the identification input data and the The protein function identification apparatus according to claim 1 or 2 , wherein a functional category to which the shape complementarity evaluation value belongs is determined. The plurality of shape complementarity evaluation values of each of the receptors and the plurality of ligands and the charge information of each of the receptors constitute a receptor vector of each of the receptors,
The plurality of shape complementarity evaluation values of the function unknown protein and the plurality of ligands and the charge information of the function unknown protein constitute an identification input vector of the function unknown protein,
The learning identification unit detects a separation plane that divides a plurality of receptor vectors corresponding to the plurality of receptors into the plurality of function categories, and a function to which the identification input vector belongs from a positional relationship between the separation plane and the identification input vector The protein function identification device according to claim 3 , wherein a category is discriminated. The protein function identification device according to claim 3 or 4 , wherein the plurality of function categories are three categories of antibodies, enzymes, and other functions. A protein function identification method for identifying the function of a protein with unknown function based on the three-dimensional structure of a protein with known function and the three-dimensional structure of a protein with unknown function,
Read the three-dimensional structure of multiple receptors as proteins of known function from the receptor storage unit,
Read the three-dimensional structure of multiple ligands from the ligand storage unit,
Processing to identify the protein surface from the three-dimensional structure of the protein, the thickness of the protein surface can be adjusted,
When each receptor docks with the plurality of ligands for each of the plurality of receptors whose functions are known as teacher data for function identification based on the three-dimensional structures of the plurality of receptors and the three-dimensional structures of the plurality of ligands. A plurality of shape complementarity evaluation values are calculated by performing a docking simulation to match the unevenness of the surface of the compatible body, and calculating charge information including a difference between the total charge of each receptor and the total charge of the surface ,
Enter the three-dimensional structure of the unknown protein
Processing to identify the protein surface from the three-dimensional structure of the protein, the thickness of the protein surface can be adjusted,
Based on the three-dimensional structure of the unknown protein and the three-dimensional structure of the plurality of ligands, a plurality of shape complementarity evaluation values when the unknown protein is docked with the plurality of ligands as identification input data for function identification And calculating the charge information including the difference between the total charge of the entire protein of unknown function and the total charge of the surface, by performing the docking simulation ,
The learning identification is performed by learning the teacher data to identify the function of the function-unknown protein, and the learning identification is performed by the plurality of shape complementarity evaluations of docking with the plurality of ligands in a plurality of receptors having a common function. Identifying the function of the function-unknown protein by determining the function of a receptor whose plurality of shape complementarity evaluation values and the charge information are similar to the function-unknown protein based on the similarity of the value and the charge information A protein function identification method characterized by the above. A protein function identification method for identifying the function of a protein with unknown function based on the three-dimensional structure of a protein with known function and the three-dimensional structure of a protein with unknown function,
Read the three-dimensional structure of multiple receptors as proteins of known function from the receptor storage unit,
Read the three-dimensional structure of multiple ligands from the ligand storage unit,
Processing to identify the protein surface from the three-dimensional structure of the protein, the thickness of the protein surface can be adjusted,
When each receptor docks with the plurality of ligands for each of the plurality of receptors whose functions are known as teacher data for function identification based on the three-dimensional structures of the plurality of receptors and the three-dimensional structures of the plurality of ligands. A plurality of shape complementarity evaluation values are calculated by performing a docking simulation that matches the unevenness of the surface of the compatible body, and the solvent exposed area of each receptor , the total number of positively charged residues, and the total negatively charged residues The total number of histidine residues, the total total charge, the number of surface positively charged residues, the number of negatively charged residues on the surface, the number of histidine residues on the surface, the surface Calculate charge information including the total charge and the difference between the total charge and the total charge on the surface ,
Enter the three-dimensional structure of the unknown protein
Processing to identify the protein surface from the three-dimensional structure of the protein, the thickness of the protein surface can be adjusted,
Based on the three-dimensional structure of the unknown protein and the three-dimensional structure of the plurality of ligands, a plurality of shape complementarity evaluation values when the unknown protein is docked with the plurality of ligands as identification input data for function identification Is calculated by performing the docking simulation and the solvent exposure area of the protein of unknown function , the total number of positively charged residues, the total number of negatively charged residues, the total number of histidine residues, Total total charge, number of surface positive charge residues, number of surface negative charge residues, number of surface histidine residues, total surface charge, total total charge and total surface charge Charge information including the difference between
The learning identification is performed by learning the teacher data to identify the function of the function-unknown protein, and the learning identification is performed by the plurality of shape complementarity evaluations of docking with the plurality of ligands in a plurality of receptors having a common function. Identifying the function of the function-unknown protein by determining the function of a receptor whose plurality of shape complementarity evaluation values and the charge information are similar to the function-unknown protein based on the similarity of the value and the charge information A protein function identification method characterized by the above. A protein function identification program for causing a computer to execute a process of identifying a function of a protein with unknown function based on the three-dimensional structure of a protein with known function and the three-dimensional structure of a protein with unknown function,
Read the three-dimensional structure of multiple receptors as proteins of known function from the receptor storage unit,
Read the three-dimensional structure of multiple ligands from the ligand storage unit,
Processing to identify the protein surface from the three-dimensional structure of the protein, the thickness of the protein surface can be adjusted,
When each receptor docks with the plurality of ligands for each of the plurality of receptors whose functions are known as teacher data for function identification based on the three-dimensional structures of the plurality of receptors and the three-dimensional structures of the plurality of ligands. A plurality of shape complementarity evaluation values are calculated by performing a docking simulation to match the unevenness of the surface of the compatible body, and calculating charge information including a difference between the total charge of each receptor and the total charge of the surface ,
Enter the three-dimensional structure of the unknown protein
Processing to identify the protein surface from the three-dimensional structure of the protein, the thickness of the protein surface can be adjusted,
Based on the three-dimensional structure of the unknown protein and the three-dimensional structure of the plurality of ligands, a plurality of shape complementarity evaluation values when the unknown protein is docked with the plurality of ligands as identification input data for function identification And calculating the charge information including the difference between the total charge of the entire protein of unknown function and the total charge of the surface, by performing the docking simulation ,
The computer executes a process for learning to identify the function of the unknown protein by learning the teacher data, and the learning identification process includes docking with the plurality of ligands in a plurality of receptors having a common function. Based on the similarity of the plurality of shape complementarity evaluation values and the charge information, the function of the receptor having the plurality of shape complementarity evaluation values and the charge information similar to the function unknown protein is obtained. A protein function identification program characterized by identifying the function of an unknown protein. A protein function identification program for causing a computer to execute a process of identifying a function of a protein with unknown function based on the three-dimensional structure of a protein with known function and the three-dimensional structure of a protein with unknown function,
Read the three-dimensional structure of multiple receptors as proteins of known function from the receptor storage unit,
Read the three-dimensional structure of multiple ligands from the ligand storage unit,
Processing to identify the protein surface from the three-dimensional structure of the protein, the thickness of the protein surface can be adjusted,
When each receptor docks with the plurality of ligands for each of the plurality of receptors whose functions are known as teacher data for function identification based on the three-dimensional structures of the plurality of receptors and the three-dimensional structures of the plurality of ligands. A plurality of shape complementarity evaluation values are calculated by performing a docking simulation that matches the unevenness of the surface of the compatible body, and the solvent exposed area of each receptor , the total number of positively charged residues, and the total negatively charged residues The total number of histidine residues, the total total charge, the number of surface positively charged residues, the number of negatively charged residues on the surface, the number of histidine residues on the surface, the surface Calculate charge information including the total charge and the difference between the total charge and the total charge on the surface ,
Enter the three-dimensional structure of the unknown protein
Processing to identify the protein surface from the three-dimensional structure of the protein, the thickness of the protein surface can be adjusted,
Based on the three-dimensional structure of the unknown protein and the three-dimensional structure of the plurality of ligands, a plurality of shape complementarity evaluation values when the unknown protein is docked with the plurality of ligands as identification input data for function identification Is calculated by performing the docking simulation and the solvent exposure area of the protein of unknown function , the total number of positively charged residues, the total number of negatively charged residues, the total number of histidine residues, Total total charge, number of surface positive charge residues, number of surface negative charge residues, number of surface histidine residues, total surface charge, total total charge and total surface charge Charge information including the difference between
The computer executes a process for learning to identify the function of the unknown protein by learning the teacher data, and the learning identification process includes docking with the plurality of ligands in a plurality of receptors having a common function. Based on the similarity of the plurality of shape complementarity evaluation values and the charge information, the function of the receptor having the plurality of shape complementarity evaluation values and the charge information similar to the function unknown protein is obtained. A protein function identification program characterized by identifying the function of an unknown protein.