JP2007106708A - Device and program for analyzing structural reaction characteristics correlation - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve its analytical accuracy by reducing a burden of a user when analyzing a correlation with reaction characteristics of a molecular structure. <P>SOLUTION: A device has a molecule superimposing part 11 for superimposing at least two molecules on the basis of a predetermined rule, an area dividing part 12 for dividing an area into a plurality of areas with every atom or group of paying attention to the superimposed molecules, a potential energy calculating part 13 for calculating potential energy including at least electrostatic and three-dimensional potential energies in respective lattice points of a mesh by dividing the molecules superimposed by the molecule superimposing part into the meshes, a correlation calculating part 14 for calculating a correlation between averaged potential energy and a reaction characteristic value by calculating the averaged potential energy with respective plural areas and a result presenting part 15 for presenting the reaction characteristics value as a function (a linear expression) with the potential energy as an independent variable on the basis of calculation results. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a structure reaction characteristic correlation analysis apparatus and a structure reaction characteristic correlation analysis program for analyzing a correlation between a reaction characteristic value of a molecule and each potential energy for a predetermined target molecule.

  Drug development is one of the fields that require information on molecular reaction characteristics in research and development. In drug development, the drug activity of various drugs on the living body is examined, and drug molecules having higher activity are designed. The drug activity is evaluated by an activity value (log (1 / C)) which is a logarithmic value of the reciprocal of the concentration C which shows the efficacy against a living body at a certain strength.

  Drug activity is one of the reaction characteristics of a molecule, and is considered to be determined by the interaction potential energy of a drug molecule with respect to a receptor molecule present in a cell membrane of a living body. In other words, it is believed that there is a correlation between interaction potential energy and drug activity. However, the effect of the physicochemical interaction cannot be explained directly because it depends on the receptor, which is a large molecule, and the surrounding environment. Therefore, each molecule of the reference molecule group having a structure similar to the molecule of the specified drug is spatially superimposed, and the correlation coefficient between the three-dimensional interaction potential energy and the activity value of each molecule is obtained. A calculation method called three-dimensional quantitative structure-activity relationship analysis (3D-QSAR), which analyzes the relationship between drug structure and activity value by calculation, has been developed.

As one of the calculation methods, VFA (Voronoi Field Analysis) developed by one of the present inventors is known (for example, see Non-Patent Document 1). VFA is characterized in that a space surrounding a molecule is divided into Voronoi regions surrounding each atom constituting the molecule, and the interaction potential energy is averaged and handled for each Voronoi region. On the other hand, CoMFA (Comparative Molecular Field Analysis: see Patent Document 1), which is a 3D-QSAR calculation method known before that, divides a space by a grid of 1-2 angstrom intervals and interacts at each point. This is a method of handling potential energy values. VFA has a problem that the above-mentioned correlation coefficient tends to increase as the number of grids of CoMFA increases, and the amount of calculation becomes enormous, and that an energy cut-off is necessary in the calculation of steric potential energy described later. Because of this, it is difficult to make a direct comparison with the Hansh-Fujita method (for example, refer to Non-Patent Document 2), which has a problem in which the analysis results are arbitrary, and the analysis examples have been accumulated for over 30 years. It solves problems such as difficulties.
US Pat. No. 5,307,287 H. Chuman et al., Quant. Struct.-Act. Relat., 17, 313-326 (1998) C. Hansch, T. Fujita, J. Am. Chem. Soc., 86, 1616-1626 (1963)

  However, VFA also has many adjustment and optimization operations such as superposition of reference molecule groups and parameter selection when determining correlation coefficients, and these operations are burdens on the user, and the analysis results that VFA should originally provide There is a problem that the accuracy cannot be sufficiently obtained.

  The present invention has been made in order to solve the above-described problems, and reduces the amount of calculation and reduces the burden on the user when analyzing the correlation with the reaction characteristics of the molecular structure, thereby improving the accuracy of the analysis. It is an object of the present invention to provide a structure-reaction characteristic correlation analysis apparatus and a structure-reaction characteristic correlation analysis program that improve performance.

  The invention according to the aspect of the present invention includes a molecule superimposing unit that superimposes at least two molecules based on a predetermined rule, and a plurality of regions for each atom or group of interest with respect to the molecules superimposed in the molecule superimposing unit. And a potential energy calculation unit that divides the molecules superimposed by the molecule superimposing unit into meshes and calculates potential energy including at least electrostatic potential energy and steric potential energy at each lattice point of the mesh. An average potential energy for each of the plurality of regions to which the lattice points belong, a correlation calculation unit for calculating a correlation between the average potential energy and a reaction characteristic value, and a calculation result of the correlation calculation unit The reaction characteristic value is presented as a linear expression with the potential energy as an independent variable. Characterized by comprising a presentation unit. In addition to the invention of an object (apparatus), the present invention is established as an invention of a method realized by an apparatus, and further, an invention of a program for realizing the method.

  According to the present invention, it is possible to reduce the amount of calculation for the correlation calculation, reduce the burden on the user when analyzing the correlation with the reaction characteristic of the molecular structure, and improve the accuracy of the analysis.

Embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a schematic configuration of a structural reaction characteristic correlation analyzer according to an embodiment of the present invention. In the block diagram of FIG. 1, arrows indicate the flow of data. The structural reaction characteristic correlation analysis apparatus according to an embodiment of the present invention may be realized using a general-purpose computer or a dedicated computer, and correlation analysis required by the present invention may be realized. As long as it is possible to execute, the configuration may be such a configuration.
The structural reaction characteristic correlation analysis apparatus according to an embodiment of the present invention includes a structural reaction characteristic correlation analysis unit 10 that performs correlation analysis of reaction characteristics by superimposing at least two molecules, and inputs various data and control data. An input unit 20 for inputting, a data storage unit 30 for storing molecular data, superimposition data, etc., a reaction characteristic value acquiring unit 40 for acquiring reaction characteristics, and a correlation analysis result of the structural reaction characteristic correlation analyzing unit 10 And a display unit 50 for displaying.
The structure-reaction characteristic correlation analysis unit 10 receives a plurality of molecules and superimposes them by a predetermined method, and a region dividing unit 12 divides the superposed molecules around a singular point into predetermined regions. And a potential energy calculation unit 13 for calculating the potential energy at each lattice point by dividing the superposed molecule into a predetermined mesh, and a correlation calculation unit 14 for calculating the correlation between the potential energy and the reaction characteristic in the divided region. And a result presentation unit 15 for presenting the result calculated by the correlation calculation unit 14.

The specific operation of the structural reaction characteristic correlation analyzer configured as described above will be described.
First, the user selects at least two desired mutually similar molecules from the data storage unit 30 via the input unit 20. Here, “similar” refers to molecules having the same principal part of the molecule. For example, a substituent (in this specification, not only a substituent but also a molecule having one function as a “group” The molecule is the same when the ")" is removed. In the following description, the number of molecules may be two or more. However, for convenience of explanation, it is assumed that two molecules are selected. In this case, for example, data downloaded from a communication line or molecules obtained via a storage medium are stored in the data storage unit 30 in advance. In addition, the molecule stored in the data storage unit 30 may be a molecule in which a part thereof is substituted with a substituent by a user, or includes data excluding a substituent or only a substituent. You can leave.

The molecule selected by the input unit 20 is input to the molecular superimposing unit 11 of the structural reaction characteristic correlation analyzing unit 10 and two molecules are superimposed as follows. For the superposition of molecules, the corresponding atoms may be automatically obtained by calculation, or the user should superimpose (or superimpose) two molecules displayed on the display unit 50 as corresponding atoms. You may specify some or all of the possible atoms. Alternatively, the automatically superimposed molecules may be resuperposed by designating a corresponding part (for example, atom) by the user. For superposition, in the two molecules A and B, for the corresponding atoms (or atomic groups), regarding the position vectors thereof, if the internal degrees of freedom of the molecules are not considered, rAi, rBi ( When i = 1, 2, 3,..., n),
F = Σ | rAi−rBi | ²
It is assumed that an optimum overlay is obtained when is minimized. Thus, by rotating the bonding portion between the atoms so that the positions of the atoms of the two molecules coincide as much as possible based on the corresponding atoms (this is referred to as “conformation change”), Superimpose two molecules. This superposition is shown in FIG. First, as shown in FIG. 2A, two different selected molecules A and B are displayed on the display unit 50. About these two molecules A and B, the similar parts are overlapped (that is, atoms that are appropriate for overlapping are matched) and displayed as a superposition result on the display unit 50 (FIG. 2B). reference). In addition, when superimposing, it is preferable to rotate the molecule or rotate the bonding portion in order to appropriately superimpose so that the positions of the atoms of each molecule are more consistent. In addition, when superimposing a substituent on a certain compound, what is necessary is just to superimpose by superimposing the atoms which a substituent couple | bonds. For example, when a methyl group is superimposed on a benzene ring, the carbon of the methyl group is bonded to one carbon of the benzene ring, so a dummy atom is added to the carbon atom of the methyl group, and the atom and benzene It should be superposed on one carbon of the ring.

  The molecules superimposed by the molecular superimposing unit 11 superimposed as described above are output to the region dividing unit 12 and the potential energy calculating unit 13 each time they are superimposed, and are also output to the data storage unit 30 each time. Remembered. In this case, it is preferable that the data storage unit 30 stores the data of each molecule before superposition together with the data after superposition. Here, the space occupied by the molecules needs to be a closed space for calculation. Therefore, this closed space is defined as follows, and calculation is performed in the closed space. This closed space is a space that covers the entire molecule with a space several angstroms larger than the space where the van der Waals radii of each atom are connected.

  The region dividing unit 12 performs the division while paying attention to atoms or groups in the molecule in order to reduce the variables when the correlation calculation unit 14 described later in detail performs the correlation calculation. Specifically, Voronoi division is performed, and in this case, when the region is divided for each atom based on the atoms in the molecule, the range of action of each atom is up to the midpoint of adjacent atoms. Define and divide the region by the plane connecting the midpoints. FIG. 3 is a diagram illustrating the divided areas after the division. As shown in FIG. 3, it can be seen that a Voronoi region is defined as a range in which the action of each atom reaches. Although it is most appropriate to perform splitting centered on atoms as the basis of region splitting, Voronoi splitting may be performed by taking a substituent as one unit and capturing the substituent as one atom.

  First, the potential energy calculation unit 13 divides the entire molecule into a three-dimensional mesh that is a smaller mesh than the region divided by the region division unit 12. The mesh interval is determined by the balance between the ability and accuracy of the computer, and is not particularly defined. However, as described in Non-Patent Document 1, it may be about 0.3 angstrom. Then, the potential energy (electrostatic potential energy and steric potential energy) is obtained for all lattice points of the three-dimensional mesh. Here, electrostatic potential energy is electrostatic potential energy with a molecule when H + (hydrogen nucleus) is placed at a lattice point, and steric potential energy is whether the lattice point exists on the van der Waals surface. It is shown. The potential energy may be used as an index for determining the similarity of molecules. The “similar molecule” is a molecule having a structure as shown in FIG. 4A and having the same structure except for Xn, R1, and R2, except for Xn, R1, and R2. A molecule having a certain structure. For example, in the molecule shown in FIG. 4 (a), as shown in FIG. 4 (b), Xn is H, 2-Cl,..., R1 is H, Me or Et, and R2 is replaced with H or Me. It shows that it is possible.

  The result of division by the region dividing unit 12 and the potential energy calculated by the potential energy calculating unit 13 are output to the correlation calculating unit 14. The correlation calculation unit 14 also outputs the reaction characteristic value acquired by the reaction characteristic value acquisition unit 40. The reaction characteristic value acquired by the reaction characteristic value acquisition unit 40 indicates the activity value of the molecule (that is, the ease of reaction). For example, when the concentration at which a certain drug begins to work is C, This reciprocal can be represented by a logarithmic value (= log (1 / C)). Therefore, in the following description, the reaction characteristic value may be proved as an activity value.

The correlation calculation unit 14 performs correlation calculation according to a flowchart as shown in FIG. The potential energy calculation unit 13 determines to which region of the region division unit the mesh lattice point belongs (step S1). For example, when the molecule is divided into two regions A and B by the region dividing unit 12, it is determined which region each mesh lattice point belongs to. And average potential energy is calculated | required for every area | region (step S2). Here, an arithmetic average is used as the averaged potential energy. As the arithmetic average, a simple average may be obtained, a weighted average in which the weight is closer to the atom may be used, and various other average methods are conceivable. Then, a correlation between the characteristic value and the averaged potential energy is obtained by an approximation method using a support vector machine (SVM) (step S3). The support vector machine is considered to have many applications, and is applied to classification of text categorization, image recognition, character recognition, and the like. In the support vector machine, a boundary that optimally separates two sets arranged in space can be obtained without considering a probability distribution model of points belonging to the set. Where the loss function corresponding to the function approximation is
max {0, | y−f (x) | −ε}
It is known that a convex quadratic programming problem similar to the case of the identification problem results by defining with the ε insensitive function. Using this algorithm, an approximate expression is obtained. In this case, global optimization can be performed by using a support vector machine, and a local optimal solution like other optimization methods is not output as an optimal solution. This result is given, for example, as a linear equation (primary expression) having a coefficient indicating an activity value (activity) for each atom or molecule (substituent) in the target molecule, that is, a reaction characteristic value.

The above result is presented as a first order approximate expression by the result presentation unit 15, for example.
Activity = a ₁ x ₁ + a ₂ x ₂ + a ₃ x ₃ + ...
+ B ₁ y ₁ + b ₂ y ₂ + b ₃ y ₃ +... + C
Is displayed on the display unit 50. Here, for example, x is due to electrostatic potential energy, and y is due to steric potential energy. C is a constant. According to this linear equation, the coefficient a of the atom or molecule is determined as to how large or small the total activity value is when an atom or molecule (substituent) is replaced with another atom or molecule (substituent). Since it is determined by b, it is possible to intuitively obtain an indication of which atom or molecule is substituted to increase the activity value. Therefore, even when a new compound is produced by substituting a part of the molecule with another atom or molecule, the activity value of that compound can be quantitatively predicted. It is effective for.

  Then, based on the presented result, substitution of atoms or molecules in a desired portion (that is, re-superimposition) is performed, and the activity value of the compound can be estimated by repeating the above steps.

  When a compound that is considered to have a desired reaction characteristic value is determined by estimating the activity value as described above, an accurate reaction characteristic value can be obtained by a normal experiment. However, in the embodiment of the present invention, since the correlation calculation is obtained by using a support vector machine, a highly accurate solution can be obtained by changing the kernel function to obtain a nonlinear solution (that is, reaction characteristic value). Can be requested. That is, in the present invention, since an optimal solution is calculated using a support vector machine, a linear equation and a nonlinear solution can be obtained without using different algorithms. Therefore, in one embodiment of the present invention, before the actual reaction characteristic value of a compound is experimentally determined, the correlation calculation unit 14 uses a support vector machine to determine a non-linear and highly accurate solution. Furthermore, since an accurate predicted value of the reaction characteristic value (that is, the activity value) can be obtained in advance, the labor of the experiment can be reduced.

  In addition, in the above, after estimating the response characteristic values of various compounds using the first-order approximation equation obtained by the support vector machine, the support vector machine further calculates nonlinear and more accurate reaction characteristic values. However, the present invention is not limited to this. When a compound is input from the input unit 20 (or selected by the input unit 20), a highly accurate reaction characteristic value is predicted for the compound by a non-linear optimization calculation as it is using a support vector machine. A value may be calculated.

  As described above, according to the present invention, the molecule is divided into a plurality of regions by Voronoi division, and the correlation between the averaged potential energy in the region and the reaction characteristic value is calculated. The number of such variables can be greatly reduced. Accordingly, the calculation speed is increased and the burden on the computer is reduced. In addition, since the correlation is defined by a linear equation with potential energy as an independent variable, a factor (for example, a substituent, a molecule, or an atom that most likely affects the reaction characteristic value with reference to the coefficient of the linear equation). ) Can be estimated easily. Moreover, since the similarity of molecules is presented based on the potential energy, the molecules can be superimposed with high accuracy by the user.

  Furthermore, since the superposition state of the molecules is stored in the data storage unit, the superposition state can be easily reproduced, and a part of the superposition can be easily changed and re-superimposed. This also allows users to easily superimpose large molecules with a large number of set parameters, such as correspondence between atoms and rotation around the bond, by comparing and examining various superposition states of molecules. A state determination can be made. In addition, the potential energy distribution is explicitly displayed by hue and brightness, so the user can check the spatial distribution of the potential energy before calculating the average value for each (Voronoi) segmented area. However, it is possible to more appropriately superimpose molecules.

  Further, by executing correlation calculation using a support vector machine, high-speed calculation can be executed even if the number of variables (parameters) increases. In addition, a global optimum solution can be presented, and the problem of the local minimum solution (Local Minimum) does not occur. Furthermore, since the solution by the support vector machine is more general (accuracy) than the solution obtained by other methods, when preparing new data and predicting the reaction characteristic correlation for the data, other methods are available. More reasonable predictions are possible. In addition, only a support vector machine can be used to change the kernel function for linear expression and non-linear higher-precision prediction calculation, so various calculations according to user requirements can be performed with a single computer. become.

  The present invention is not limited to the above-described embodiments, and various modifications can be made without departing from the scope of the invention at the stage of implementation. Further, the above embodiments include inventions at various stages, and various inventions can be extracted by appropriately combining a plurality of disclosed constituent elements.

  In addition, for example, even if some structural requirements are deleted from all the structural requirements shown in each embodiment, the problem described in the column of the problem to be solved by the invention can be solved, and the effect described in the effect of the invention Can be obtained as an invention.

1 is a block diagram showing a schematic configuration of a structural reaction characteristic correlation analyzer according to an embodiment of the present invention. The figure which shows a mode that two different molecules are piled up. The figure which shows the division area after the division | segmentation by Voronoi division. The figure for demonstrating a similar molecule | numerator. The flowchart for performing a correlation calculation.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... Structural reaction characteristic correlation analysis part 11 ... Molecule superposition part 12 ... Area | region division part 13 ... Potential energy calculation part 14 ... Correlation calculation part 15 ... Result presentation part 20 ... Input part 30 ... Data storage part 40 ... Reaction characteristic value acquisition part 50 ... Display section

Claims (9)

A molecule superimposing unit that superimposes at least two molecules based on a predetermined rule;
A region dividing unit that divides the molecule superimposed by the molecule superimposing unit into a plurality of regions for each atom or group of interest;
A potential energy calculation unit that divides the molecules superimposed in the molecule superimposition unit into meshes and calculates potential energy including at least electrostatic potential energy and steric potential energy at each lattice point of the mesh;
A correlation calculating unit that calculates an averaged potential energy for each of the plurality of regions to which the lattice points belong, and calculates a correlation between the averaged potential energy and a reaction characteristic value;
A structural reaction characteristic correlation analysis apparatus comprising: a result presentation unit that presents a reaction characteristic value as a linear expression having the potential energy as an independent variable based on a calculation result of the correlation calculation unit. 2. The structural reaction characteristic correlation analyzer according to claim 1, further comprising a data storage unit for storing a superposition state of molecules superimposed by the molecular superimposing unit. The structural reaction characteristic correlation analyzer according to claim 1 or 2, wherein the display unit displays a situation before and after superimposition of molecules in the molecular superimposition unit;
An input unit that accepts user input;
The molecular superposition unit re-superimposes the superposed molecule displayed on the display unit based on the operation information input to the input unit. 4. The structural reaction characteristic correlation analyzer according to claim 1, wherein the region dividing unit uses Voronoi division centering on each atom or group of the molecule after superimposition. 5. And dividing the molecule into a plurality of regions. 5. The structural reaction characteristic correlation analysis device according to claim 1, wherein the correlation calculation unit calculates a correlation between the reaction characteristic value and the potential energy using a support vector machine. A structure-reaction characteristic correlation analyzer. 6. The structural reaction characteristic correlation analyzer according to claim 1, wherein the potential energy calculation unit calculates a potential energy of each molecule, and the result presentation unit displays the superimposed molecules. A structure-reaction characteristic correlation analysis apparatus characterized in that the similarity is determined based on the potential energy of each molecule. The structural reaction characteristic correlation analyzer according to any one of claims 1 to 6, wherein the display unit explicitly displays the distribution of the potential energy values by hue or brightness. Structure reaction characteristic correlation analyzer. A molecule superimposing unit that superimposes at least two molecules based on a predetermined rule;
A region dividing unit that divides the molecule superimposed by the molecule superimposing unit into a plurality of regions for each atom or group of interest;
A potential energy calculation unit that divides the molecules superimposed in the molecule superimposition unit into meshes and calculates potential energy including at least electrostatic potential energy and steric potential energy at each lattice point of the mesh;
A correlation calculator that calculates an averaged potential energy for each of the plurality of regions to which the lattice points belong, and calculates a correlation between the averaged potential energy and a reaction characteristic value using a support vector machine;
A structural reaction characteristic correlation analysis device comprising: a result presentation unit for presenting a reaction characteristic value based on a calculation result of the correlation calculation unit. In a computer-executable structural reaction characteristic correlation analysis program for calculating the correlation between the molecular reaction characteristic value and the molecular potential energy,
Molecular superimposing means for superimposing at least two molecules based on a predetermined rule;
Region dividing means for dividing the molecule superimposed by the molecule superimposing means into a plurality of regions for each atom or group of interest;
A potential energy calculation unit that divides the molecule superimposed by the molecule superimposing unit into a mesh and calculates a potential energy including at least electrostatic potential energy and steric potential energy at each lattice point of the mesh;
A correlation calculating means for calculating an average potential energy for each of the plurality of regions to which the lattice points belong, and calculating a correlation between the average potential energy and a reaction characteristic value;
A structure reaction characteristic correlation analysis program comprising: a result presentation unit that presents a reaction characteristic value as a linear expression having the potential energy as an independent variable based on a calculation result of the correlation calculation unit.

Images