Design system of binding structure for polymer molecule

Abstract

A simulation system that increases a polymer molecule binding prediction speed on a parallel and distributed computer system is provided. The simulation system determines a decomposition width that decomposes a search region of polymer molecule by means of translational operation and rotational operation and the number of searches in the decomposed regions into which the search region is decomposed, determines the number of decomposed regions into which the search region is decomposed, determines the number of computing units to which the decomposed regions are to be allocated, allocates the decomposed regions to the respective computing units, determines search points within the decomposed regions, transmits data of the search points to a computing unit that computes the binding energy and the energy gradient vector, performs communication control that receives data associated with the binding energy and the energy gradient vector from the computing unit, and determines the local minimum value of the binding energy in the decomposed region and the minimum value in the search region, thereby making it possible to determine the convergence of the binding energy.

Description

BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWING

FIG. 1 is a conceptual diagram showing a basic structure of a design system of a binding structure for polymer molecule under a parallel and distributed environment according to the present invention;

FIG. 2 is a conceptual diagram showing the configuration of a design system of a binding structure for polymer molecule according to a first embodiment of the present invention, and a diagram for explaining a specific transaction of a step of computing the minimum of a binding energy of protein by the aid of a translational operation of atomic coordinates of compound;

FIG. 3 is a diagram showing an example in which the search region using the translational operation is inputted with the coordinate values of the center position of the search region to display a spatial region which is a search region having a cubic configuration;

FIG. 4 is a diagram showing a change in decomposed transactional regions that are averagely allocated to the respective computing units with respect to the decomposition width, a change in the number of line searches with respect to the decomposition width, a change in the number of line searches of the local minimum value of the binding energy, and an optimum decomposition width with the axis of abscissa indicative of the decomposition width of the translational search region and the axis of ordinate indicative of the number of line searches or the number of decomposed regions;

FIG. 5 is a diagram showing an example of a binding energy in a two-dimensional space of the decomposed translational search region;

FIGS. 6A and 6B are diagrams for explaining an example of determining the number of decomposed translational regions that minimizes the number of line searches on the computing unit and the decomposed translational regions in the case where computing unit groups that are connected on the network have the same transaction performance and the different transaction performance, respectively;

FIG. 7 is a diagram showing an example of the line search in an energy gradient direction in a decomposed search region of the two-dimensional space in the translational search region shown in FIG. 5;

FIG. 8 is a diagram for explaining the transmission of the search point to the computing unit;

FIG. 9 is a diagram showing a change in decomposed transactional regions that are averagely allocated to the respective computing units with respect to the decomposition width, a change in the number of line searches with respect to the decomposition width, a change in the number of line searches of the local minimum value of the binding energy, and an optimum decomposition width with the axis of abscissa indicative of the decomposition width of the translational search region and the axis of ordinate indicative of the number of line searches or the number of decomposed regions;

FIG. 10 is a diagram for explaining an example of determining the number of decomposed translational regions that minimize the number of line searches on the computing units of computing unit groups that are connected to each other on the network, and the decomposed translational regions;

FIG. 11 is a diagram showing the number of line searches that minimize the local minimum value of the binding energy;

FIG. 12 is a conceptual diagram showing the configuration of a design system of a binding structure for polymer molecule according to a second embodiment of the present invention, and a diagram for explaining a specific transaction of a step of computing the minimum of a binding energy of protein by the aid of a rotational operation of compound;

FIGS. 13A, 13B, and 13C are diagrams showing an example of search regions that perform the rotational operation of compound which is indicated by Euler's angles (F, θ, ψ);

FIG. 14 is a diagram showing a change in decomposed rotational regions (hereinafter referred to as “decomposed rotational regions”) that are averagely allocated to the respective computing units with respect to the decomposition angle width, a change in the number of line searches with respect to the decomposition angle width, a change in the number of line searches of the local minimum value of the binding energy with respect to the respective decomposed rotational regions with respect to the decomposed angle width, and an optimum decomposition width with the axis of abscissa indicative of the decomposition angle width of the rotational search region and the axis of ordinate indicative of the number of line searches or the number of decomposed regions;

FIG. 15 is a diagram showing an example in which the decomposed rotational regions are allocated to the computing units of the computing unit groups that are connected to each other on the network;

FIG. 16 is a diagram showing a change in decomposed rotational regions (hereinafter referred to as “decomposed rotational regions”) that are averagely allocated to the respective computing units with respect to the decomposition angle width, a change in the number of line searches with respect to the decomposition angle width, a change in the number of line searches of the local minimum value of the binding energy with respect to the respective decomposed rotational regions with respect to the decomposed angle width, and an optimum decomposition width with the axis of abscissa indicative of the decomposition angle width of the translational search region and the axis of ordinate indicative of the number of line searches or the number of decomposed regions;

FIG. 17 is a diagram showing an example in which the decomposed rotational regions that minimize the number of line searches are allocated to the computing units of the computing unit groups that are connected to each other on the network;

FIG. 18 is a diagram showing the number of line searches that obtains the local minimum value of the binding energy with respect to the decomposition width of the decomposed rotational regions;

FIG. 19 is a diagram showing an example in which a data transfer network is of a grid configuration as one general structural example of the parallel and distributed computer system where drug discovery simulation is performed under a parallel and distributed environment;

FIG. 20 is a diagram showing an example in which the data transfer networks are connected in a ring configuration;

FIG. 21 is a diagram showing an example in which grid machines are connected to each other in a ring configuration on the data transfer network;

FIG. 22 is a diagram showing a concept that pays attention to an example of a transacting function of a related system that performs simulation which predicts a binding structure of polymer molecule on the parallel and distributed computer system; and

FIG. 23 is a diagram showing an example of the minimum search of a binding energy using a Monte Carlo method in a search region of translational operation and rotational operation.

DETAILED DESCRIPTION OF THE INVENTION

Now, a description will be given in more detail of preferred embodiments of the present invention with reference to the accompanying drawings.

In all of figures for explaining the embodiments, parts having the same functions are indicated by identical references, and their duplex description may be omitted.

(Basic Configuration of System)

FIGS. 19 to 21 show an example in which a parallel and distributed environment is structured by parallel computers and PC clusters, and the parallel computers and the PC clusters are controlled by a personal computer 100 as a general structural example of a parallel and distributed computer system. In a design system of a binding structure for polymer molecule according to the present invention, the configuration of a parallel and distributed computer system can be identical with the system configuration described above.

FIG. 1 is a conceptual diagram showing a basic structure of a design system of a binding structure for polymer molecule according to the present invention. Referring to FIG. 1, reference numeral 100 denotes a personal computer for control of the design system of a binding structure for polymer molecule according to the present invention, which corresponds to the personal computer of the parallel and distributed computer system which is described with reference to FIGS. 19 to 21. The design system of a binding structure for polymer molecule includes an input system 101 for inputting data of simulation, a transaction and control unit 102 for executing a transaction for distributing the computation of a binding energy to the respective computing units 104 from inputted data, and for executing a transaction for integrating the results of distributing the computation to the respective computing units and computing the binding energy, and an output system 103 for displaying the operating status of the transaction and control unit 102 or the integrated data. In this example, the personal computer 100 includes a memory system such as a disk or a memory for holding necessary program or data, or a graphical visualization system that generates a display image for an interface with the user. However, for convenience, the memory system and the graphical visualization system is indicated in blocks of the input system 101 and the output system 103. Reference numeral 104 denotes the respective computing units of the parallel and distributed computer system which is described with reference to FIGS. 19 to 21, and executes computation according to information obtained from the personal computer 100 for control, and reports the computation results to the personal computer 100. Also, reference numeral 105 denotes a management unit of the parallel and distributed computer system described with reference to FIGS. 19 to 21. The personal computer 100, the computing units 104, and the management unit 105 of the parallel and distributed computer system are indicated by dashed lines, and connected by heavy lines in the sense that the personal computers 100, the computing units 104, and the management unit 105 are capable of mutually interchanging necessary data with each other. Also, the association between the transaction step of the transaction and control unit 102 and the transaction step of the computing units 104 are connected to each other by thin solid lines.

The user inputs, from the input system 101, search regions with respect to the binding structure of protein which is obtained from genome information and polymer molecule to be subjected to simulation such as DNA or RNA, for example, protein and compound in water molecule, as well as the number of operable computing units in the parallel computers and the PC clusters under the parallel and distributed environment.

In the transaction and control unit 102, the data to be subjected to simulation which is introduced through the input system 101 is inputted, and held in the memory system. Also, data for calculation of the binding energy is distributed to the computing units 104 through the above-described procedures (1) to (6). The computation results of the binding energy which have been performed through the above-described procedures (9) and (10) are integrated through the above procedures (6) to (8) on the basis of the data that has been distributed by the respective computing units 104, and then displayed on the output system 103.

Output data output the minimum binding energy with respect to the binding structure of protein and compound in water molecule, and atomic coordinate data of protein, compound, and water molecule in the binding structure.

Also, one of the computing units 104 acts as the transaction and control unit 102, and is capable of distributing the computation of the binding energy to the respective computing units 104, and integrating the computation results of the binding energy which has been performed by the respective computing units.

Hereinafter, a description will be given of a transaction procedure of distributing the computation of the binding energy to the respective computing units in the transaction and control unit 102 of the personal computer 100, and integrating the computation results. Reference numeral 1021 is a step of determining the decomposition width that decomposes the search region, and the number of searches for computing the binding energy in a search region that expresses a range where the operation of changing the atomic coordinates of compound is performed. Reference numeral 1022 is a step of determining the number of the decomposed search region (hereinafter referred to as “decomposed regions”) and the range of the decomposed regions. Reference numeral 1023 is a step of determining the number of computing units in the parallel and distributed environment where the decomposed regions are allocated, by using the number of decomposed regions. Reference numeral 1024 is a step of determining the number of decomposed regions that are allocated to the respective computing units and the decomposed regions by using the number of decomposed regions, the number of searches, and the number of computing units in the parallel and distributed environment for allocation of the decomposed regions. Reference numeral 1025 is a step of determining search points within the decomposed regions for computation of the binding energy in the decomposed regions that are allocated to the respective computing units. Reference numeral 1026 is a step of communication control, which transmits data of the respective search points that are allocated for computation of the binding energy to the respective computing units which are determined in Steps 1024 and 1025. On the contrary, the binding energies and the energy gradient vectors at the respective search points which have been calculated in the respective computing units are received. Reference numeral 1027 is a step of determining the minimum value from the local minimum values of the binding energies that have been received in Step 1026 and computed by the respective computing units and the local minimum value that has been computed by all of the computing units. Reference numeral 1028 is a step of determining whether the iterative calculation is executed, or not, on the basis of the convergence of the local minimum value of the binding energy within the decomposed region. In the determination, since the number of searches is known in advance as will be understood from embodiments to be described later, the iterative calculation may not be performed when the number of searches exceeds the known value.

In the case of executing the iterative calculation, control is returned to Step 1025. In the case of completing the repetitive calculation, the minimum value of the binding energy of protein and compound in water molecule and the atomic coordinate data with respect to the binding structure are transmitted to the output system 103.

In Step 1029, data of the search points are received from Step 1026, and the operation of changing the atomic coordinates of compound with respect to the designated search points is executed in the respective computing units 104 of the parallel and distributed computer system. The atomic coordinates of protein, the atomic coordinates of water molecule, and the atomic coordinates of compound which have been changed are inputted, and the binding energy of protein and compound and the energy gradient vector, that is, a force that is exerted on compound and a torque that is exerted on compound are computed and outputted. The binding energy, the force that is exerted on compound, and the torque that is exerted on compound which have been outputted are transmitted to the personal computer 100 in Step 1026 of communication control.

First Embodiment

FIG. 2 is a conceptual diagram showing the configuration of a design system of a binding structure for polymer molecule according to a first embodiment of the present invention, and a diagram for explaining a specific transaction of a step of computing the minimum value of a binding energy of protein by the aid of a translational operation of atomic coordinates of compound. The entire configuration is identical with that shown in FIG. 1, but the transaction and control unit 102 and the respective transaction steps 1021 to 1029 shown in FIG. 1 are changed to the transaction and control unit 112 and the respective transaction steps 1121 to 1129 according to the changes of the transaction contents.

The user inputs a search region (hereinafter referred to as “translational search region”) that expresses a range where the translational operation of the atomic coordinates of compound is performed from the input system 101 in order to search the minimum value of the binding energy of protein that is obtained from genome information and polymer molecule to be subjected to simulation such as DNA and RNA, for example, protein and compound in water molecule. As shown in an example of FIG. 3, the coordinate values of the center position 302 of the search region in protein 301, a spatial region width W303 that is a search region having a cubic configuration, and the number of available computing units N_pa of the parallel computers or the PC clusters in the parallel and distributed environment are inputted. The number of available computing units N_pa can be inputted by the user, or data that is supplied from the management unit 105 of the parallel and distributed computer system may be applied. In this example, a small cubic configuration ΔW indicated in the spatial region width W303 schematically shows a cube having a decomposition width ΔW of the translational search region which is determined according to a manner that will be described later.

In Step 1121, the decomposition width that minimizes the number of linearly searching the local minimum value of the binding energy and the number of line searches are determined in the translational search region that performs the translational operation of compound.

FIG. 4 is a diagram showing a change in decomposed transactional regions (hereinafter referred to as “decomposed translational region) that are averagely allocated to the respective computing units with respect to the decomposition width, a change in the number of line searches with respect to the decomposition width, a change in the number of line searches of the local minimum value of the binding energy, and an optimum decomposition width with the axis of abscissa indicative of the decomposition width of the translational search region and the axis of ordinate indicative of the number of line searches or the number of decomposed regions (the number of decomposed search regions). Reference numeral 401 denotes a line that represents a change in the number of decomposed translational regions that are allocated to the respective computing units according to the decomposition width, reference numeral 402 denotes a line that represents a change of the number of line searches according to the decomposition width, and reference numeral 403 denotes a line that represents a change in the number of line searches of the local minimum value of the binding energy on the respective computing units according to the decomposition width. When the decomposition width of the translational search region is ΔW, the number of decomposed translational regions that are averagely allocated to the respective computing units is (W/ΔW)³/N^pa. Also, in the respective decomposed translational regions, the binding energy is calculated by the line search in the energy gradient vector direction, that is, a direction of the force that is exerted on compound, the number of line searches until the local minimum value of the binding energy is found is given by F(X) represented in Expression (3). In this expression, X=ΔW/R_t.

FIG. 5 is a diagram showing an example of the binding energy in a two-dimensional space of the decomposed translational search region. When R_t is within a distance of the radius R_t from a coordinate position 501 at which the minimum value of the binding energy exists in the decomposed translational region, the binding energy is an energy region 502 of a quadratic function with respect to a distance, and when R_t is equal to or higher than the radius R_t, the binding energy is a random energy region 503. The size of the radius R_t can be prepared as the system, but can be inputted by the user.

A line 403 that represents a change in the number of line searches of the local minimum value of the binding energy on the respective computing units with respect to the decomposition width ΔW of the translational search region is represented by the product (W/ΔW)³/N_pa×F (ΔW/R_t). Therefore, when the decomposition width ΔW_opt where (W/ΔW)³/N_pa×F (ΔW/R_t) is the minimum is selected, the decomposition width ΔW_opt of the translational search region W as well as the binding energies on the respective computing units are computed, thereby making it possible to determine N_It=F (ΔW_opt/R_t) where the number of line searches of the local minimum value of the binding energy is minimum.

Hereinafter, a description will be given of how to find the number of line searches F(X) for finding the local minimum value of the binding energy represented by Expression (3) with respect to X of a certain value with the probability P_th.

$\begin{array}{cc}F\left(X\right)=\frac{\log \left(1-P_{\mathrm{th}}\right)}{\log \left\{1-P_1\left(X\right)-P_2\left(X\right)\right\}}& \left(3\right)\end{array}$

where P₁(X) is given by Expression (4).

$\begin{array}{cc}{P}_1\left(X\right)=\frac{1}{2\pi }{\int }_{{\theta }_3}^{\frac{\pi }{2}}\theta {\cos }^{-1}\left(\frac{x_{\alpha 1}x_{\alpha 2}+y_{\alpha 1}y_{\alpha 2}+z_{\alpha 1}z_{\alpha 2}}{\sqrt{x_{\alpha 1}^2+y_{\alpha 1}^2+z_{\alpha 1}^2}\sqrt{x_{\alpha 2}^2+y_{\alpha 2}^2+z_{\alpha 2}^2}}\right)& \left(4\right)\end{array}$

where X_ai, Y_ai, and Z_ai (i=1, 2) are given by Expression (5) and θ₃ is given by Expression (6).

$\begin{array}{cc}\begin{array}{c}\begin{array}{c}{x}_{\alpha i}={\alpha }_i\left(\theta ,X\right)\left(x_3-x_i\right)+x_3\\ y_{\alpha i}={\alpha }_i\left(\theta ,X\right)\left(y_3-y_i\right)+y_3\end{array}\\ z_{\alpha i}={\alpha }_i\left(\theta ,X\right)\left(z_3-z_i\right)+z_3\end{array}\}& \left(5\right)\\ {\theta }_3={\cos }^{-1}\left(\frac{Z_3}{\sqrt{x_3^2+y_3^2+z_3^2}}\right)& \left(6\right)\end{array}$

where a_i(θ,X) is given by Expression (7), Expression (8), Expression (9), and Expression (10).

$\begin{array}{cc}{\alpha }_i\left(\theta ,X\right)=\frac{-b_i\pm \sqrt{b_i^2-a_ic_i}}{a_i}& \left(7\right)\\ a_i=\left\{{\left(x_3-x_i\right)}^2+{\left(y_3-y_i\right)}^2+{\left(z_3-z_i\right)}^2\right\}{\cos }^2\theta -{\left(z_3-z_i\right)}^2& \left(8\right)\\ b_i=\left\{x_3\left(x_3-x_i\right)+\left(y_3-y_i\right)+\left(z_3-z_i\right)\right\}{\cos }^2\theta -z_3\left(z_3-z_i\right)& \left(9\right)\\ c_i=\left\{x_3^2+y_3^2+z_3^2\right\}{\cos }^2\theta -z_3^2& \left(10\right)\end{array}$

Also, (x₁, y₁, z₁), (x₂, y₂, z₂), and (X₃, Y₃, Z₃) are given by Expression (11), Expression (12), and Expression (13).

$\begin{array}{cc}\left(x_1,y_1,z_1\right)=\left(\sqrt{3X^2-2X+1,}0,0\right)& \left(11\right)\\ \left(x_2,y_2,z_2\right)=\left(\frac{3X^2-2X}{\sqrt{3X^2-2X+1}},\sqrt{\frac{6X^2-4X+1}{3X^2-2X+1}},0\right)& \left(12\right)\\ \left(x_3,y_3,z_3\right)=\left(\frac{3X^2-2X}{\sqrt{3X^2-2X+1}},\frac{3X^2-2X}{\sqrt{6X^2-4X+1}\sqrt{3X^2-2X+1}},\sqrt{\frac{9X^2-6X+1}{6X^2-4X+1}}\right)& \left(13\right)\end{array}$

On the other hand, P₂(x) is given by Expression (14).

$\begin{array}{cc}{P}_2\left(X\right)=\frac{3}{2\pi }{\int }_{{\theta }_1}^{{\theta }_2}2{\tan }^{-1}\left(\frac{\sqrt{1-2{B\left(\theta ,X\right)}^2}}{\sqrt{3X^2-4\beta \left(\theta ,X\right)X+2{\beta \left(\theta ,X\right)}^2}}\right)& \left(14\right)\end{array}$

where β(θ, x) is given by Expression (15), Expression (16), Expression (17), and Expression (18).

$\begin{array}{cc}\beta \left(\theta ,X\right)=\frac{-b\pm \sqrt{b^2-ac}}{a}& \left(15\right)\\ a=2-2{\cos }^2\theta & \left(16\right)\\ b=-X\left\{2-2{\cos }^2\theta \right\}& \left(17\right)\\ c=X^2\left\{2-3{\cos }^2\theta \right\}& \left(18\right)\end{array}$

where θ₁ and θ₂ are given by Expression (19) and Expression (20).

$\begin{array}{cc}\cos {\theta }_1=\frac{\sqrt{2X}-\frac{1}{\sqrt{2}}}{\sqrt{3X^2-2X+\frac{1}{2}}}& \left(19\right)\\ \cos {\theta }_2=\frac{\sqrt{2X}-1}{\sqrt{3X^2-2\sqrt{2}X+1}}& \left(20\right)\end{array}$

In Step 1122, the number of decomposition of the translational search region and a range of the decomposed translational region are determined by the aid of the decomposition width ΔW_opt of the translational search region W. The number of translational regions by which the translational search region W is decomposed is N_W=(W/ΔW_opt)³. When the translational coordinates within the decomposed translational region are (x, y, z), 0=x=ΔW_opt, 0=y=ΔW_opt, and 0=z=ΔW_opt are set as the decomposed translational region W₁, 0=x=ΔW_opt, 0=y=ΔW_opt, and W_opt=z=2ΔW_opt are set as the decomposed translational region W₂, . . . , and (W−ΔW_opt)=x=W, (W−ΔW_opt)=y=W, and (W−ΔW_opt)=z=W are set as the decomposed translational region W_Nw.

In general, the decomposed translational region W_n is represented by (n_x−1)ΔW_opt=x=n_xΔW_opt, (n_y−1)ΔW_opt=y=n_yΔW_opt, and (n_z−1)ΔW_opt=z=n_zΔW_opt, where n=1, 2, . . . , N_w, n_x, n_y, n_z=1, 2, . . . , W/ΔW_opt.

As described above, the number of decomposed translational regions N_w, as well as the range of the decomposed translational regions W₁, W₂, . . . , W_n, . . . , W_Nw can be determined.

In Step 1123, the number of computing units in the parallel and distributed environment for allocation of the decomposed translational regions by using the number of decomposed translational regions N_w. In the case where the number of decomposed translational regions N_w is larger than the number of available computing units N_pa, the number of computing units N_p to which the decomposed translational regions are allocated is N_pa. On the other hand, in the case where the number of decomposed translational regions N_w is smaller than the number of available computing units N_pa, the number of computing units N_p to which the decomposed translational regions are allocated is N_w. In this way, the number of computing units N_p in the parallel and distributed environment for allocation of the decomposed translational regions can be determined.

In Step 1124, the number of decomposed translational regions that are allocated to the respective computing units and the decomposed translational regions are determined by using the number of decomposed translational regions N_w the number of line searches N_It as well as the number of computing units N_p in the parallel and distributed environment for allocation of the decomposed translational regions. Since the number of line searches with respect to the respective decomposed translational regions is N_It, the number of decomposed translational regions that are allocated to the respective computing units can be substantially equalized, and the number N_di of allocated decomposed translational regions is [N_w/N_p] or [N_w/N_p]−1. In this expression, a value obtained in the form of [**] is rounded out to the whole number, and N_dt decomposed translational regions that are arbitrarily selected from the decomposed translational regions W₁, W₂, . . . , W_Nw are allocated to each of the N_p computing units under the parallel and distributed environment.

FIG. 6A shows an appearance in which the decomposed translational regions W₁, W₂, . . . , W_Ndt are allocated to the computing unit P₁ of the computing units P₁, P₂, . . . , P_NP in the computer unit group 203 that is connected on the network 204, the decomposed translational regions W_Nd1+1, W_Ndt+2, . . . , W_2Ndt are allocated to the computing unit P₂, . . . , and the decomposed translational regions W_{Np(Ndt−1)+1}, W_{Np(Ndt−1)+2}, . . . , W_NpNdt (=W_Nw) are allocated to the computing unit P_Np. In general, the range of the decomposed translational region W_n that is allocated to the computing unit P_m is (n_x−1)ΔW_opt=x=n_xΔW_opt, (n_y−1)ΔW_opt=y=n_yΔW_opt, and (n_z−1)ΔW_opt=z=n_zΔW_opt, where m=1, 2, . . . , N_p, n=1, 2, . . . , and N_w, n_x, n_y, n_z=1, 2, . . . , W/ΔW_opt.

Also, a description will be given of an example in which the number of decomposed translating regions that are allocated to the respective computing units, and the decomposed translational regions are determined in the case where the transaction performance of the computing units in the parallel and distributed computer system is different. In the parallel computers and the PC clusters under the parallel and distributed environment, the transaction performances TP₁, TP₂, . . . , TP_N1 of the available computing units, and the number of computing units N_pa,1, N_pa,2, . . . , N_pa,N1 are inputted from the input system 101, or applied with data that is obtained from the management unit 105 of the parallel and distributed computer system. In this example, N₁ is the type of available computing units.

In Step 1121 where the decomposition width that minimizes the number of line searches of the local minimum value of the binding energy, and the number of line searches are determined, SN_pa,i×{T_pi/TP_min} (i=1, 2, . . . , N₁) is applied instead of the above-mentioned number of available computing units N_pa. In this case, a value obtained in the form of [**] is rounded out to the whole number, and TP_min is the minimum transaction performance in the computing unit as a standard for conversion to the same transaction performance.

As described above, the number of decomposed translational regions N_w is determined in Step 1122 with respect to the decomposition width ΔW_opi and the number of line searches N_It which are determined by converting the number N_pa of available computing units to the equivalent number of computing units. Then, the number of computing units N_p to which the decomposed translational regions are to be allocated is determined in Step 1123. In Step 1124, since the number of line searches with respect to the respective decomposed translational regions is N_It, it is possible that the number of decomposed translational regions that are allocated to the respective computing units is substantially equalized.

The number of allocated decomposed translational regions N_dt,_i=[N_w/N_p]×N_pa,i×{TP_i/TP_min}(i=1, 2, . . . , N₁) is obtained by using the transaction performance of the respective computing units. N_di,i decomposed translational regions that are arbitrarily selected from the decomposed translational regions W₁, W₂, . . . , W_Nw are allocated to each of the N_i computing units in the parallel and distributed environment.

FIG. 6B shows an appearance in which the decomposed translational regions W₁, W₂, . . . , W_Ndt,1 are allocated to the computing unit P₁ of the computing units P₁, P₂, . . . , P_NP in the computing unit group 203 that is connected on the network 204, the decomposed translational regions W_(Ndt,1)+1, W_(Ndt,1)+2, . . . , W_{(Ndt,1)+Ndt,2} are allocated to the computing unit P₂, . . . , and the decomposed translational regions W_{(Ndt,1+, . . . ,+Ndt,N1−1)+1}, W_{(Ndt,1+, . . . +Ndt, N1−1)+2}, . . . , W_{(Ndt,1+, . . . , +Ndt, N1−1)+Ndt,N1} (=W_Nw) are allocated to the computing unit P_N1.

In Step 1125, the search point within the decomposed translational region where the binding energy is to be computed is determined in the decomposed translational regions W_n that are allocated to the computing units.

FIG. 7 is a diagram showing an example of line search in the energy gradient direction in the decomposed search regions of the two-dimensional space of the translational search regions shown in FIG. 5. It is assumed that one of search points within the decomposed translational region W_n is (x_n, y_n, z_n). At the search point 701, it is assumed that a force norm 702 that is the gradient vector of the binding energy with respect to the translational coordinates is F=(F_x, F_y, F_z). The search point within the decomposed translational region for computation of the binding energy is given by (x_n±kF_xdw, y_n±kF_ydw, z_n±kF_zdw). In this expression, k is an integer that satisfies (n_x−1)ΔW_opt=x_n±kF_xdw =n_xΔW_opt, (n_y−1)ΔW_opt=y_n±kF_ydw=n_yΔW_opt, and (n_z−1)ΔW_opt=Z_n±kF_zdw=n_zΔW_opt. Also, dw is the translational width in the line search. Further, the translational width dw can be prepared at the system side as default, or can be set through the input system 101 by the user.

As described above, when it is assumed that the number of search points is N_It,n, N_It,n search points (x_n±kF_xdw, Y_n±kF_ydw, Z_n±kF_zdw) within the decomposed translational regions where the binding energy is to be computed can be determined with respect to the decomposed translational regions W_n that have been allocated to the respective computing units.

Step 1126 is a communication control in which, in order to compute the binding energy, data of the search points that are allocated to the respective computing units which are determined in Step 1124 and Step 1125 is transmitted. When it is assumed that the decomposed translational regions W_n are allocated to the computing units P_m in Step 1124, and N_It,n search points (x_n±kF_xdw, y_n±kF_ydw, z_n±kF_zdw) which exist within the decomposed translational region W_n are given in Step 1125, N_It,n search points (x_n±kF_xdw, y_n±kF_ydw, and z_n±kF_zdw) are transmitted to the computing unit P_m.

As shown in FIG. 8, data of N_It,n search points (x_n±kF_xdw, y_n±kF_ydw, and z_n±kF_zdw) that exist within the decomposed translational regions W_n which are allocated to the respective computing units are transmitted to all of the computing units P₁, P₂, . . . , P_Np of the computing unit group 203. On the contrary, the binding energies with respect to the search points which have been computed by the respective computing units and the gradient vectors of the binding energy with respect to the translational coordinate, that is, a force that is exerted on compound is received. The binding energy of compound that has been subjected to translational operation and protein in water molecule, and a force that is exerted on compound are received with respect to N_It,n translational coordinates (x_n±kF_xdw, y_n±kF_ydw, and z_n±kF_zdw) which exist within the decomposed translational regions W_n which have been transmitted to the computing units P_m.

In Step 1127, the minimum value is determined from the local minimum values of the binding energies that have been received in Step 1126 and computed by the respective computing units and the local minimum value that has been computed by all of the computing units. The search point that gives the smallest binding energy in N_It,n binding energies is selected in correspondence with the N_It,n search points (x_n±kF_xdw, y_n±kF_ydw, and z_n±kF_zdw) which exist within the decomposed translational regions W_n which have been computed by the computing units P_m. The line search repeats the iterative calculation that is N_Ii times or more in the number of line searches.

The selected search point is a search point at which the gradient vector of the binding energy is obtained with respect to the translational coordinates described in Step 1125. The binding energy that has been obtained as the result of the iterative calculation is the local minimum value of the binding energies within the decomposed translational regions W_n. The smallest energy among the local minimum values of the binding energies that have been obtained in correspondence with all of the computing units P₁, P₂, . . . , P_Nw is the minimum value of the binding energy.

In Step 1128, it is determined whether the iterative calculation of the line search is repeated, or not, on the basis of the convergence of the local minimum values of the binding energy within the decomposed translational region. When the number of line searches of the decomposed translational region W_n which have been allocated to the computing units P_n is N_It or lower, control is returned to Step 1125, and the iterative calculation of the line search is executed. When the number of line searches of the decomposed translational region W_n which have been allocated to the computing units P_n is equal to or higher than N_It, or when a difference between the local minimum value of the binding energy and the local minimum value of the binding energy which has been found in the previous iterative calculation is converged to a threshold energy or lower, the line search is completed. The minimum value of the binding energy of protein and compound in water molecule, as well as the atomic coordinate data with respect to the binding structure of protein and compound in water molecule is transmitted to the output system 103.

In Step 1129, the binding energy corresponding to the search point in the respective computing units, and the gradient vector of the binding energy with respect to the translational coordinates are calculated. The respective computing units P_m receive the data of the search points, that is, N_It,n translational coordinates (x_n±kF_xdw, y_n±kF_ydw, and z_n±kF_zdw) within the decomposed translational regions W_n from Step 1126. The respective computing units P_m execute the translational operation with respect to the atomic coordinates of compound by using the search points (x_n±kF_xdw, y_n±kF_ydw, and z_n±kF_zdw). The respective computing units Pm computes N_It,n binding energies and a force that is exerted on compound from compound that has been subjected to translational operation, and the atomic coordinates of protein in water molecule. The binding energies corresponding to N_It,n search points (x_n±kF_xdw, y_n±kF_ydw, and z_n±kF_zdw) which exist within the decomposed translational regions W_n, and the force that is exerted on compound are transmitted to Step 1126.

A description will be given of a specific example in which the cubic configuration of the spatial region width W=6 Å in the translational search region is inputted from the input system 101 as the search region with respect to the binding structure of protein and compound in water molecule, and the number of available computing units N_pa=60 in the parallel and distributed computer system is inputted with reference to FIG. 9. In Step 1121, the decomposition width of the translational search region and the number of searches are determined.

FIG. 9 shows a change 401 in the number of decomposed translational regions (hereinafter, referred to as “the number of decomposed translational regions) which are averagely allocated to the respective computing units with respect to the decomposition width, a change in the number of line searches with respect to the decomposition width when R_t is 0.5 Å, the number of line searches 403 that obtains the local minimum value of the binding energy on the respective computing units, and the optimum decomposition width with the axis of abscissa indicative of the decomposition width of the translational search region and the axis of ordinate indicative of the number of line searches or the number of decomposed regions as in FIG. 4. The decomposition width ΔW_opt of the translational search region where the number of line searches is the minimum on the respective computing units becomes 1 Å, and the number of line searches N_It can obtain 105.

In Step 1122, the number of decomposed translational regions N_w=(W/ΔW_opt)³ is 216 by using the optimum decomposition width ΔW_opt=1 Å. In Step 1123, since the number of decomposition regions N_w=216 is larger the number of available computing units N_pa=60, the number of computing units N_p to which the decomposed translational regions are to be allocated is 60. In Step 1124, since the number of line searches N_It with respect to the respective decomposed translational regions is 105, the number of decomposed translational regions that are allocated to the respective computing units can be substantially equalized. For example, as shown in FIG. 10, the number of decomposed translational regions that are allocated to 36 computing units is distributed 4 by 4 with respect to the number of computing units N_p=60 of the computing unit group 203 that is connected on the network 204, and the number of decomposed translational regions N_dt-1 that are allocated to 24 computing units is distributed 3 by 3.

In Step 1125, the search points of line searches within the decomposed translational regions which have been allocated to the respective computing units are determined. Data of the search points of the line searches is transmitted to the respective computing units in Step 1126, and the binding energy corresponding to the search points in the respective computing units, and the gradient vectors of the binding energies with respect to the translational coordinates are computed and transmitted to Step 1126, in Step 1129. In Step 1127, the local minimum values of the binding energies that have been computed by the respective computing units are obtained. In Step 1128, the iterative calculation that is 105 in the number of line searches N_It is executed.

FIG. 11 is a diagram showing the number of line searches that obtain the local minimum values of the binding energy. A solid line 402 represents a predicted value that is obtained in Step 1121, and a dotted line 1101 is the computation results of the simulation according to the present invention. In the simulation, the average number of line searches that obtain the local minimum value of the binding energy in the decomposed translational region is 134, which relatively excellently coincides with the predicted value. The number of line searches on the respective computing units is a product of the number of line searches of the respective decomposed translational regions and the number of decomposed regions on the respective computing units, and therefore 420.

On the other hand, the number of line searches for obtaining the local minimum value of the binding energy when only one of the computing units is used is predicted as 4740 by the aid of Expressions (3) to (20). Therefore, the number of line searches that obtain the local minimum values of the binding energy is reduced down to 4740/420= 1/11 times by means of the design system of a binding structure for polymer molecule using the parallel and distributed computer system, thereby making it possible to compute the binding structure that minimizes the binding energy in a high speed. In other words, the user is capable of performing high-speed computation by merely setting the search region that represents a range where the translational operation of the atomic coordinates of polymer molecule to be subjected to simulation, and the number of available computing units N_pa in the parallel computers or the PC cluster under the parallel and distributed environment.

Second Embodiment

FIG. 12 is a conceptual diagram showing the configuration of a design system of a binding structure for polymer molecule according to a second embodiment of the present invention, and a diagram for explaining a specific transaction of a step of computing the minimum value of a binding energy of protein by the aid of a rotational operation of compound whereas the first embodiment describes the transaction of computing the minimum value of the binding energy of protein by the aid of the translational operation of the atomic coordinates of compound. In this example, the entire configuration is identical with the configuration shown in FIG. 1 as in the first embodiment. However, the transaction and control unit 102 and the respective transaction steps 1021 to 1029 in FIG. 1 are changed to a transaction and control unit 122 and the respective transaction steps 1221 to 1229 according to the change of the processing contents.

The user inputs an angle region width x from the input system 101 as a search region (hereinafter referred to as “rotational search region”) that expresses a range where the rotational operation of compound is performed in order to search the minimum value of the binding energy of protein that is obtained from genome information and polymer molecule to be subjected to simulation such as DNA and RNA, for example, protein and compound in water molecule. As shown in FIGS. 13A to 13C, the search region where the rotational operation of compound 1301 is performed is represented by 0=F=x, 0=θ=x/2, and 0=ψ=x when being expressed by Euler's angles (F, θ, ψ), where F is the rotational angle about a z-axis, θ is the rotational angle about an x-axis subsequent to the rotation of the z-axis, and ψ is the rotational angle about the z-axis subsequent to the rotations of the z-axis and the x-axis, and x is 0=x=2p. Also, the user inputs the number of computing units N_pa in the parallel computers or the PC cluster in the parallel and distributed environment.

Hereinafter, a description will be given of a transaction flow of distributing the binding computation to the respective computing units and integrating the results in the transaction and control unit 122.

In Step 1221, the decomposition width that minimizes the number of linearly searching the local minimum value of the binding energy, and the number of line searches are determined in the rotational search region that performs the rotational operation of compound. FIG. 14 is a diagram showing a change 1401 in decomposed rotational regions (hereinafter referred to as “decomposed rotational regions”) that are averagely allocated to the respective computing units with respect to the decomposition angle width, a change 1402 in the number of line searches with respect to the decomposition angle width, a change 1403 in the number of line searches of the local minimum value of the binding energy with respect to the respective decomposed rotational regions with respect to the decomposed angle width, and an optimum decomposition width with the axis of abscissa indicative of the decomposition angle width of the rotational search region and the axis of ordinate indicative of the number of line searches or the number of decomposed regions. When the decomposition width of the rotational search region is Δx, the number of decomposition rotational regions that are averagely allocated to the respective computing units is (x/Δx)³/2N_pa. Also, in the respective decomposed rotational regions, the binding energy is calculated by the line search in the energy gradient vector direction, that is, a rotational direction of the torque that is exerted on compound, and the number of line searches until the local minimum value of the binding energy is found is given by F(X) represented in Expression (3).

In this expression, X=ΔW/R_r. When R_r is within a distance of the radius R_r from a coordinate position at which the minimum value of the binding energy exists in the decomposed rotational region, the binding energy is an energy region of a quadratic function with respect to a distance, and when R_r is equal to or higher than the radius R_r, the binding energy is a random energy region. The number of line search 1403 of the local minimum value of the binding energy on the respective computing units is represented by their product (x/Δx)³/2N_pa×F(Δx/R_r). Accordingly, when the decomposition width Δ_xopt where (x/Δx)³/2N_pa×F(Δx/R_r) becomes the minimum is selected, the decomposition width Δx_opt of the rotational search region x, as well as the binding energy on the respective computing units are calculated, thereby making it possible to determine N_Ir=F (Δx_opt/R_r) where the number of line searches of the local minimum value of the binding energy becomes the minimum.

In Step 1222, the number of decomposition of the rotational search region and the range of the decomposed rotational regions are determined by the aid of the decomposition width Δx_opt of the rotational search region x. The number of rotational regions that decompose the rotational search region x is N_x=(x/Δx_opt)³/2. When the rotational coordinates within the decomposed rotational region are (F, θ, ψ), 0=F=Δx_opt, 0=θ=Δx_opt, and 0=ψ=Δx_opt are set as the decomposed rotary region x₁, 0=F=Δx_opt, 0=θ=Δx_opt, and x_opt=ψ=2x_opt are set as the decomposed rotary region x₂, . . . , and (x−Δx_opt)=F=x, (x/2−Δx_opt)=θ=x/2, and (x−Δx_opt)=ψ=x are set as the decomposed rotary region X_Nx.

In general, the decomposed rotational region x_n is represented by (n_F−1)Δx_opt=F=n_FΔx_opt, (n_θ−1)Δx_opt=θ=n_θΔxopt, and (n_θ−1)Δx_opt=ψ=n_ψΔx_opt, where n=1, 2, . . . , N_x, n_F, n_ψ=1, 2, . . . , x/Δx_opt, and n_θ=1, 2, . . . , x/2Δx_opt. As described above, it is possible to determine the number of decomposed rotational regions N_x, as well as the range of the decomposed rotational regions x₁, x₂, . . . , x_n, . . . , and x_Nx can be determined.

In Step 1223, the number of computing units in the parallel and distributed environment where the decomposed rotational regions are allocated is determined by using the number of decomposed rotational regions N_x. In the case where the number of decomposed rotational regions N_x is larger than the number of available computing units N_pa, the number of computing units N_p to which the decomposed rotational regions are allocated is N_pa. On the other hand, in the case where the number of decomposed rotational regions N_x is smaller than the number of available computing units N_pa, the number of computing units N_p to which the decomposed translational regions are allocated is N_x. In this way, the number of computing units N_p in the parallel and distributed environment for allocation of the decomposed translational regions can be determined.

In Step 1224, the number of decomposed rotational regions that are allocated to the respective computing units, and the decomposed rotational region are determined by the aid of the number of decomposed rotational regions N_x, the number of line searches N_Ir as well as the number of computing units N_p in the parallel and distributed environment for allocation of the decomposed rotational regions. Since the number of line searches with respect to the respective decomposed rotational regions is N_Ir, the number of decomposed rotational regions that are allocated to the respective computing units can be substantially equalized, and the number N_dr of allocated decomposed translational regions is [N_x/N_p] or [N_x/N_p]−1. In this expression, a value obtained in the form of [**] is rounded out to the whole number, and N_dr decomposed rotational regions that are arbitrarily selected from the decomposed translational regions x₁, x₂, . . . , x_Nx are allocated to each of the N_p computing units under the parallel and distributed environment.

In Step 1225, the search point within the decomposed rotational region where the binding energy is to be computed is determined in the decomposed rotational regions x_n that are allocated to the computing units. It is assumed that one of search points within the decomposed translational region x_n is (F_n, θ_n, ψ_n). At the search point, it is assumed that the norm of a gradient vector with respect to the rotational coordinates of the binding energy is F=(F_F, F_θ, F_ψ). Then, F=(F_F, F_θ, F_ψ) can be obtained from R(F_n+F_F), θ_n+F_θ, ψ_n+F_θ)=T (T_x, T_y, T_z) R (F_n, θ_n, ψ_n) by the aid of the torque vector T=(T_x, T_y, T_z). In this example, R is an Euler's matrix, T is the rotational matrix of the torque. The search point within the decomposed rotational region for computing the binding energy is given by (F_n±kF_Fdx, θ_n±kF_θdx, ψ_n±kF_ψdx). In this expression, k is an integer that satisfies (n_F−1)Δx_opt=F_n±kF_Fdx=n_FΔx_opt, (n_θ−1)Δx_opt=θ_n±kF_θdx=n_θΔx_opt, and (n_ψ−1)Δx_opt=ψ_nkF_ψdx=n_θΔxopt. Also, dx is the rotational width in the line search. Further, the rotational width dx can be prepared at the system side as default, or can be set through the input system 101 by the user.

As described above, when it is assumed that the number of search points is N_Ir,n, N_Ir,n search points (F_n±kF_Fdx, θ_n±kF_θdx, ψ_n±kF_ψdx) within the decomposed rotational regions where the binding energy is to be computed can be determined with respect to the decomposed rotational regions x_n that have been allocated to the respective computing units.

In the communication control of Step 1226, data of the search points that are allocated in order to compute the binding energy by the respective computing units which are determined in Step 1224 and Step 1225 is transmitted. When it is assumed that the decomposed rotational regions x_n are allocated to the computing units P_m in Step 1224, and N_Ir,n search points (F_n±kF_Fdx, θ_n±kF_θdx, ψ_n±kF_ψdx) which exist within the decomposed rotational region x_n are given in Step 1225, N_Ir,n search points (F_n±kF_Fdx, θ_n±kF_θdx and ψ_n±kF_ψdx) are transmitted to the computing unit P_m. In this way, data of N_Ir,n search points (F_n±kF_Fdx, θ_n±kF_θdx, and ψ_n±kF_ψdx) that exist within the decomposed rotational regions x_n which are allocated to the respective computing units are transmitted to all of the computing units P₁, P₂, . . . , P_Np. On the contrary, the binding energies at the search points which have been computed by the respective computing units and the gradient vectors of the binding energy with respect to the rotational coordinate, that is, a force that is exerted on compound is received. The binding energy of compound that has been subjected to rotational operation and protein in water molecule, and a force that is exerted on compound are received with respect to N_Ir,n rotational coordinates (F_n±kF_Fdx, θ_n±kF_θdx, and ψ_n±kF_ψdx) which exist within the decomposed rotational regions x_n which have been transmitted to the computing units P_m.

In Step 1227, the minimum value is determined from the local minimum values of the binding energies that have been received in Step 1226 and computed by the respective computing units and the local minimum value that has been computed by all of the computing units. The search point that gives the smallest binding energy in N_Ir,n binding energies is selected in correspondence with the N_Ir,n search points (F_n±kF_Fdx, θ_n±kF_θdx, and ψ_n±kF_ψdx) which exist within the decomposed rotational regions x_n which have been computed by the computing units P_m. The line search repeats the iterative calculation that is N_Ir times or more in the number of line searches. The selected search point is a search point at which the gradient vector of the binding energy is obtained with respect to the rotational coordinates described in Step 1225. The binding energy that has been obtained as the result of the iterative calculation is the local minimum value of the binding energies within the decomposed rotational regions x_n. Then, the smallest energy among the local minimum values of the binding energies that have been obtained in correspondence with all of the computing units P₁, P₂, . . . , P_Nx is the minimum value of the binding energy.

In Step 1228, it is determined whether the iterative calculation of the line search is repeated, or not, on the basis of the convergence of the local minimum values of the binding energy within the decomposed rotational region. When the number of line searches of the decomposed translational region x_n which have been allocated to the computing units P_n is N_Ir or lower, control is returned to Step 1225, and the iterative calculation of the line search is executed. When the number of line searches of the decomposed rotational region x_n which have been allocated to the computing units P_n is equal to or higher than N_Ir, or when a difference between the local minimum value of the binding energy and the local minimum value of the binding energy which has been found in the previous iterative calculation is converged to a threshold energy or lower, the line search is completed. The minimum value of the binding energy of protein and compound in water molecule, as well as the atomic coordinate data with respect to the binding structure of protein and compound in water molecule is transmitted to the output system 103.

In Step 1229, data at the search points from Step 1226 is received in the computing unit of the parallel and distributed computer system. The respective computing units P_m receive N_Ir,n rotational coordinates (F_n±kF_Fdx, θ_n±kF_θdx, and ψ_n±kF_ψdx) within the decomposed rotational regions x_n. The respective computing units P_m execute the rotational operation with respect to the atomic coordinates of compound by using the search points (F_n±kF_Fdx, θ_n±kF_ψdx, and ψ_n±kF_ψdx). The respective computing units P_m computes N_Ir,n binding energies and a torque that is exerted on compound on the basis of the atomic coordinates of compound and protein that have been subjected to rotational operation. The binding energies corresponding to N_Ir,n search points (F_n±kF_Fdx, θ_n±kF_θdx, and ψ_n±kF_ψdx) which exist within the decomposed rotational regions X_n, and the torque that is exerted on compound are transmitted to Step 1226.

FIG. 15 is a diagram showing an appearance in which the decomposed rotational regions x₁, x₂, . . . , x_Ndr are allocated to the computing unit P₁ in the computing unit group 203 that is connected on the network 204, the decomposed rotational regions x_Ndr+1, x_Ndr+2, . . . , x_2Ndr are allocated to the computing unit P₂, . . . , and the decomposed rotational regions x_{Np(Ndr−1)+1}, X_{Np(Ndr−1)+2,} . . . , x_NpNdr(=x_Nx) are allocated to the computing unit P_Np. The range of the decomposed translational region x_n that is allocated to the computing unit P_m is (n_F−1)Δx_opt=F=n_FΔx_opt, (n_θ−1)Δx_opt=θ=n_θΔx_opt, and (n_ψ−1)Δx_opt=ψ=n_ψΔx_opt, where m=1, 2, . . . , N_p, n=1, 2, . . . , and N_x, n_F, n_ψ=1, 2, . . . , x/Δx_opt, and n_θ=1, 2, . . . , x/2Δx_opt.

Referring to FIG. 16, a description will be given of a specific example in which the angle region width x=360° of the rotational search region is inputted from the input system 101, and the number of available computing units N_pa=250 in the parallel and distributed computer system is inputted as the search region with respect to the binding structure of protein and compound in water molecule. In Step 1221, the number of decomposed rotational regions on the respective computing units with respect to the decomposition width of the rotational search region, the number of line searches in order to obtain the local minimum value of the binding energy in the decomposed rotational region, as well as the number of line searches that obtain the local minimum value of the binding energy on the respective computing units are determined. FIG. 16 is a diagram showing a change 1401 in the decomposed transactional regions (hereinafter referred to as “decomposed translational region”) that are averagely allocated to the respective computing units with respect to the decomposition width, a change 1402 in the number of line searches of the local minimum value of the binding energy with respect to the decomposition width in the respective decomposed translational regions when R_r is 15°, as well as a change 1403 in the number of line searches that obtain the local minimum value of the binding energy on the respective computing units, and an optimum decomposition angle width with the axis of abscissa indicative of the decomposition angle width (degree) of the rotational search region and the axis of ordinate indicative of the number of line searches or the number of decomposed regions. It can be obtained that the decomposition width Δx_opt of the rotational search region which minimizes the number of line searches on the respective computing units is 36°, and the number of line searches N_Ir is 160.

In Step 1222, the number of decomposed rotational regions is 500 because of N_x=(x/Δx_opt)³/2 by the aid of the optimum decomposition width Δx_opt=36°. In Step 1223, since the number of decomposed regions N_x=500 is larger than the number of available computing units N_pa=250, the number of computing units N_p to which the decomposed rotational regions are to be allocated is 250. In Step 1224, since the number of line searches N_Ir with respect to the respective decomposed rotational regions is 160, the number of decomposed rotational regions that are allocated to the respective computing units can be substantially equalized. For example, as shown in FIG. 17, the number of decomposed rotational regions N_dr that are allocated to the respective computing units is distributed as 2 with respect to the number of computing units N_p=250. In Step 1225, the search points of the line searches within the decomposed rotational region which are allocated to the respective computing units are determined. In Step 1226, data at the search points of the line searches is transmitted to the respective computing units, and in Step 1229, the binding energy corresponding to the search point in the respective computing units, and the gradient vector of the binding energy with respect to the rotational coordinates are computed, and transmitted to Step 1226. In Step 1227, the local minimum value of the binding energy which has been computed by the respective computing units is obtained. In Step 1228, the iterative calculation that is 160 or more in the number of line searches N_Ir is executed.

FIG. 18 is a diagram showing the number of line searches that obtain the local minimum values of the binding energy. A solid line 402 represents a predicted value that is obtained in Step 1221, and a dotted line 1801 is the computation results of the simulation. In the simulation, the average number of line searches that obtain the local minimum value of the binding energy in the decomposed rotational region is 214, which relatively excellently coincides with the predicted value. The number of line searches on the respective computing units is a product of the number of line searches of the respective decomposed rotational regions and the number of decomposed regions on the respective computing units, and therefore 320.

On the other hand, the number of line searches for obtaining the local minimum value of the binding energy when only one of the computing units is used is predicted as 9612 by the aid of Expressions (3) to (20). Therefore, the number of line searches that obtain the local minimum values of the binding energy is reduced down to 9612/320= 1/30 times by means of the design system of a binding structure for polymer molecule using the parallel and distributed computer system, thereby making it possible to compute the binding structure that minimizes the binding energy in a high speed. In other words, the user is capable of performing high-speed computation by merely setting the angle region width that represents a range of the rotational search region of the atomic coordinates of polymer molecule to be subjected to simulation, and the number of available computing units N_pa in the parallel computers or the PC cluster under the parallel and distributed environment.

Claims

1. A design system of a binding structure for polymer molecule which is executed by a parallel and distributed computer system, the parallel and distributed computer system comprising: a plurality of computing units including a CPU that executes computing, a memory system that holds necessary program and data, an input/output interface, and a bus that connects the CPU, the memory system, and the input/output interface;a personal computer including an input system that inputs necessary data by a user, an output system that outputs data for the user, a memory system that holds necessary program and data, an input/output interface, and a bus that connects the input system, the output system, the memory system, and the input/output interface, and executing a transaction for distributing computation to the plurality of computing units and integrating the computation result of the plurality of computing units; anda network that executes data transfer between the personal computer and the computing units through the input/output interface,a transaction procedure of distributing computation to the respective computing units by the personal computer and integrating the computation results of the respective computing units with respect to the number of operable computing units that are designated by the user, and the search region of polymer molecule, comprising the steps of:(1) determining a decomposition width that decomposes a search region and the number of searches that compute a binding energy;(2) determining the number of decomposed search regions (decomposed regions) and a range of the decomposed regions by using the decomposition width of the search region;(3) determining the number of computing units in the parallel and distributed environment for allocation of the decomposed regions within the number of operable computing units by using the number of decomposed regions;(4) determining the number of decomposed regions that are allocated to the respective computing units and the decomposed regions by using the number of decomposed regions, the number of searches, and the number of computing units to which the decomposed regions are allocated;(5) determining the search points within the decomposed regions where the binding energy is computed in the decomposed regions that are allocated to the respective computing units;(6) transmitting data at the respective search points which are allocated in order to compute the binding energy to the respective computing units that are determined in the steps (4) and (5), and receiving the binding energies and the energy gradient vector at the respective search points which are computed by the respective computing units for communication control;(7) determining the minimum value in the local minimum values of the binding energies that are computed by the respective computing units which are received in the step (6), and the local minimum value that is computed by all of the computing units;(8) determining whether iterative calculation is executed, or not, on the basis of the convergence of the local minimum value of the binding energy within the decomposed region, or denies the iterative calculation when the number of searches exceeds a given value; and(9) returning the control to the step (5) in the case where the iterative calculation is executed, and outputting data of the atomic coordinates with respect to the minimum value of the binding energy of polymer molecule and the binding structure of polymer molecule to the output system in the case where the iterative calculation is completed,wherein the transaction in the respective computing units includes a step of receiving data at the search points from the step (6), and computing the binding energy of polymer molecule and the energy gradient vector at the designated search point.

2. The design system of a binding structure for polymer molecule according to claim 1, wherein the transaction procedures of distributing the computation to the respective computing units and integrating the computation results of the respective computing units among the functions of the personal computers are shared to one of the computing units.

3. The design system of a binding structure for polymer molecule according to claim 1, wherein the personal computer and the computing units are formed of a single computer.

4. The design system of a binding structure for polymer molecule according to claim 1, wherein the network includes a management unit of a parallel and distributed computer system that collects the number of operable computing units among the computing units, the transaction performance of the operable computing units, and the information on the network speed, and supplies the information to the personal computer.

5. The design system of a binding structure for polymer molecule according to claim 1, wherein the step of determining the number of searches which computes the decomposition width that decomposes the search region and the binding energy selects a decomposition width ΔWopt that minimizes a product of the number of decomposed translational regions that are averagely allocated to the respective computing units according to the decomposition width that decomposes the search region, and the number of line searches of the local minimum values of the binding energy in the decomposed translational region as the decomposition width, and selects the number of searches in the decomposition width as the number of searches minimizes a product of the number of decomposed translational or rotational regions that are averagely allocated to the respective computing units according to the decomposition width that decomposes the search region, and the number of line searches of the local minimum values of the binding energy in the decomposed translational or rotational region as the decomposition width, and selects the number of searches in the decomposition width as the number of searches.

6. The design system of a binding structure for polymer molecule according to claim 5, wherein a prediction of the number of searches in the decomposed region using the decomposition width of the search region has, as a predicted value of the number of searches, the number of line searches until the binding energy is computed at the search points on a line in a direction of the energy gradient vector, the computation of the energy gradient vector is repeated at the search points on the line which minimize the binding energy, and the local minimum value of the binding energy is obtained.

7. The design system of a binding structure for polymer molecule according to claim 6, wherein the number of line searches determines the number of searches of decomposed regions which is predicted by using the decomposition width of the search region, and a distance of the energy region which is a quadratic function with respect to a distance from the local minimum position of the binding energy.

8. The design system of a binding structure for polymer molecule according to claim 7, wherein the size of a radius Rt is prepared at the system side as default, or is designated by the user, in the energy region where binding energy from a coordinate position at which the minimum value of the binding energy exits is a quadratic function with respect to a distance.

9. The design system of a binding structure for polymer molecule according to claim 5, wherein when the number of decomposed regions is smaller than the number of available computing units which is inputted from the input system by the user, the number of decomposed regions is set as the number of computing units to which the decomposed regions are to be allocated, and when the number of decomposed regions is larger than the number of available computing units which is inputted from the input system by the user, the number of computing units to which the decomposed regions are to be allocated is set to the number of operable computing units.

10. The design system of a binding structure for polymer molecule according to claim 9, wherein the number of operable computing units is data that is given from the management unit of the parallel and distributed computer system which is disposed in the system, or a value that is determined by the user with reference to the data given from the management unit.

11. The design system of a binding structure for polymer molecule according to claim 9, wherein the number of decomposed regions that are allocated to the respective computing units and a range of the decomposed region are determined according to the number of decomposed regions that is set according to the number of operable computing units, the number of searches of the decomposed regions, and the number of computing units to which the decomposed regions are to be allocated.

12. The design system of a binding structure for polymer molecule according to claim 4, wherein the number of computing units having the transaction performance converted to the identical transaction performance is deviated from the number of operable computing units and the transaction performance of the operable computing units which are supplied from the management unit of the parallel and distributed computer system to determine the number of decomposed regions that are allocated to the respective computing units, and a range of the decomposed regions.

13. The design system of a binding structure for polymer molecule according to claim 1, wherein, in the decomposed regions that are allocated to the respective computing units, a gradient vector with respect to the binding energy is computed at one point of search points that exist within the decomposed regions, and a search point that exists on a line in a direction of the gradient vector and at which the binding energy that exists within the decomposed region is to be computed is set.

14. The design system of a binding structure for polymer molecule according to claim 1, wherein data related to the decomposed regions that are allocated to the respective computing units and the search points that exist within the decomposed regions and compute the binding energy is transmitted to the computing unit that computes the binding energy and the energy gradient vector on the parallel and distributed computer system, whereas the binding energy and the energy gradient vector at the search points that are computed by the respective computing units are received.

15. The design system of a binding structure for polymer molecule according to claim 1, wherein the respective computing units determine the minimum value from the local minimum value of the binding energy that is computed in correspondence with the search points within the decomposed regions, and the local minimum values that are computed by all of the computing units.

16. The design system of a binding structure for polymer molecule according to claim 1, wherein the determination of the search points within the decomposed regions which are allocated to the respective computing units to the determination of the local minimum values and the minimum value of the binding energy are executed by the iterative calculations having the predicted number of searches of the decomposed regions, or more, and the iterative calculation is executed until a difference between the local minimum values of the binding energy and the minimum values of the binding energy which is obtained in a previous iterative calculation is converged to a threshold energy or lower.

17. The design system of a binding structure for polymer molecule according to claim 1, wherein each of the computing units receives the search points within the decomposed regions which are allocated to the respective computing units through the communication control step of the transaction and control unit of the personal computer, performs the translational operation and the rotational operation with respect to the position of polymer molecule, computes the binding energy and the energy gradient vector with respect to the atomic coordinates of polymer molecule that performs the translational operation and the rotational operation, and transmits the computation results of the binding energy and the energy gradient vector to the transaction and control unit of the personal computer through the communication control step.

18. The design system of a binding structure for polymer molecule according to claim 1, wherein the user inputs the search region of the translational operation with respect to compound and the rotational operation with respect to compound for a binding structure of protein and compound in water molecule as the binding structure of polymer molecule through the input system of the personal computer.

19. The design system of a binding structure for polymer molecule according to claim 18, wherein a force that is exerted on polymer molecule at the search point is computed with respect to the translational operation, and a torque that is exerted on polymer molecule at the search point is computed with respect to the rotational operation, as the energy gradient vector.

20. The design system of a binding structure for polymer molecule according to claim 18, wherein the translational width dw of the translational operation and the decomposed angle width of the rotational operation in the computation of the energy gradient vector are prepared at the system side as default, or can be designated by the user through the input system.