JPH1083389A - Computing system for analyzing molecular orbital and molecular orbital computing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable the material prediction and pharmacological action prediction of after-mentioned material by enabling executing the computation of an abinitio molecular orbital by a realistic time and a computer resource. SOLUTION: A matrix element is computed by a computer cluster 1 consisting of plural mutually connected computers and the obtained matrix element is transferred to a vector computer 2 through a transmission line for connecting the vector computer 31 to execute the linear arithmetic of a matrix on the vector computer. At the time of computing the abinitio molecular orbital, 1-electronic integration, 2-electronic integration and Fock matrix element consisting of 1-electronic integration and 2-electronic integration with respect to a prescribed material are generated on the computer cluster 2 to transmit electronic integration data and Fock matrix element to the computer 2 through the computer cluster and the transmission line 31 to obtain the fixed value and the fixed vector of Fock matrix on the vector computer to output the molecular orbital energy and the wave function of the prescribed material.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a calculation system for performing high-speed scientific and technical calculations centering on the generation of large-scale matrices and the linear operation of large-scale matrices.

In particular, the present invention relates to a calculation system for molecular orbital analysis which processes and executes molecular orbital analysis frequently used for predicting material properties and pharmacological action in the chemical and pharmaceutical industries. Electron integration, two electron integration,
In addition, the generation of Fock matrix elements is executed in a distributed manner by a computer cluster including a plurality of computers, the generated data is transferred to a vector computer having a pipeline function, and the eigenvalues and eigenvectors of the Fock matrix are solved on the vector computer. The present invention relates to a molecular orbital analysis calculation system and a molecular orbital calculation method capable of solving a molecular orbital energy and a wave function of a given material at high speed.

[0003]

2. Description of the Related Art In scientific and technical calculations, such as the molecular orbital method and the finite element method, matrix generation and matrix linear operation are central to processing. A vector computer capable of pipeline processing has been devised in order to speed up such scientific and technical calculations, and the speed of linear operation can be increased by 10 to tens of times on a vector computer as compared with a conventional general-purpose computer. However, generation of a matrix (calculation of a matrix element) may not be accelerated so much by a vector computer. In particular, in the ab initio molecular orbital calculation, the calculation time of the Fock matrix element occupying about 80% of the calculation time is reduced to about 70% of that of the general-purpose computer, but the effect of the vector computer is hardly obtained.

On the other hand, it has been clarified that the calculation of matrix elements is accelerated according to the number of constituent processors according to a parallel computer or a computer cluster. For example, when parallelizing the ab initio molecular orbital calculation and executing it on a computer cluster, the calculation of the Fock matrix element is accelerated to some extent in proportion to the number of computers constituting the cluster.
Journal of Computational Chemistry, Vol. 14, No. 11 (1993), pp. 1347 to 1363 (Journal of Computational Chemistry,
Vol. 14, No. 11 (1993) PP.1347-1363). That is, if it is desired to complete the calculation of the matrix elements within a desired calculation time, the number of computers constituting the computer cluster may be increased accordingly. In that case, the solution of the eigenvalues and eigenvectors of the Fock matrix is not excellent in dense symmetric matrix diagonalization algorithms for parallel computers,
This has been an obstacle to speeding up large-scale molecular orbital calculations using computer clusters (parallel computers).

[0005] As an algorithm for performing matrix diagonalization at high speed on a computer cluster, Japanese Patent Application No. 1-295417 reports an algorithm of Householder transformation. However, the Householder transform can be used to solve eigenvalues and eigenvectors of matrices up to several thousand dimensions.However, in the calculation of matrices exceeding tens of thousands of dimensions, the communication overhead between the computers constituting the computer cluster increases, and it is practical. Disappears. In the chemical and pharmaceutical industries, the analysis target system for which material property prediction and pharmacological action prediction are desired is thousands of atoms, and the size of the Fock matrix in molecular orbital calculation is tens of thousands. It is generally known that eigenvalues and eigenvectors of matrices of tens of thousands or more are preferably obtained by a simultaneous inverse iteration method on a vector computer (Takeo Murata, Riki Oguni, Koichi Karaki, supercomputer, science and technology) Application to Calculation, Maruzen (1986), pp. 154 to 171
page).

[0006] Regarding the relationship between scientific and technological calculations and computer systems, for example, Japanese Patent Application No. 3-167672 discloses the following two speed-up methods in the case of numerical operations comprising explicit and implicit methods. (1) The explicit part is performed by a vector computer, and the implicit part is processed by a plurality of parallel processors.
(2) Both the explicit part and the implicit part are processed by a plurality of parallel processors. In the above-mentioned known example, a specific connection method between the vector computer and the parallel processor (parallel computer) is not described with respect to (1), and an expensive vector computer is required. We concluded that data processing needs to be redistributed and communication processing overhead could hinder speedups.
Only the speedup method according to (2) is described.

[0007] As an example of scientific and technical calculations, molecular orbital analysis, which is widely used for predicting material properties, will be described in detail below. In molecular orbital analysis, molecular orbitals and associated physical property values are generally calculated by a calculation process flow shown in FIG. These specific calculation methods are described in, for example, Abu Initio Molecular-Obital Seoli [HEHRE et al, AB
INITIO MOLECULAR ORBITAL THEORY; Wiley-Interscienc
e, published in 1986], and other textbooks related to the present invention. The molecular orbital calculation starts with the input of the molecular shape, specifically, the atomic species and atomic coordinates in the molecule. The one-particle wave function | Y> called the molecular orbital that constitutes the wave function of the whole molecule is the linear sum of each atomic orbital | Xj> (1)
It is expanded according to the formula. This corresponds to the step labeled "designate basis function" in FIG.

| Y> = ΣCj | Xj> (j = 1, 2, 3,..., N) (1) where n is the total number of atomic orbitals used in the calculation. The procedure for calculating the expansion coefficient Cj in a self-adhesive manner is the essence of ab initio molecular orbital calculation. The processing required in this case is
As shown in the flowchart of FIG. 1, an electron integration calculation step,
It comprises an initial vector calculation step and an SCF equation calculation step. In the last step, the physical properties such as chemical bond and charge distribution are calculated, and the molecular orbital calculation is completed.

In the conventional molecular orbital calculation, a series of steps shown in FIG. 1 is generally processed on a single computer.
As the electron integration calculation step, both one electron integration calculation and two electron integration calculation are calculated and recorded and stored in the high-speed memory. First, the one-electron integral is a core integral Huv and an overlap integral Suv represented by the equations (2) and (3).

Hv = <Xu | Hcore | Xv> (2) Suv = <Xu | Xv> (3) However, Hcore is a Hamiltonian when an electron moves in a molecule without being influenced by other electrons at all. The bracket notation <|> is the same as that usually used in quantum mechanical calculations.

Next, the two-electron integral is given by the following equation (4).
The electron repulsion integral (uv | ds).

(Uv | ds) = <Xu (1) Xd (2) | 1 / R12 | Xv (1) Xs (2)> (4) where R12 indicates the distance between electron 1 and electron 2.

The number of one-electron integrals represented by the equations (2) and (3) is about the square of the total number of atomic orbitals, whereas (4)
The number of two-electron repulsion integrals in the equation increases in proportion to the fourth power of the total number of atomic orbitals. For example, set the total number of atomic orbitals to 61
In the SiH2C2H4 molecule, the number of two-electron integration is about 1,800,000, and a capacity of about 15 MB is required to store the result in the high-speed memory by the double-precision integration calculation. The CPU time required for the integral calculation requires about 10 seconds when a general mainframe computer, for example, a Hitachi M880 computer is used. In general, molecules handled in the chemical and pharmaceutical industries are not simple molecules such as SiH2C2H4, but have a size several to several tens of times larger. For example, if it is 10 times, a large capacity of 150 GB is required, and it is impossible to store the data by using a magnetic disk memory as well as a semiconductor memory. In addition, since the CPU time required for the integral calculation also increases in proportion to the fourth power of the total number of atomic orbitals, about 30 hours of CPU time is required. Considering the I / O time required for inputting and outputting the integral calculation value, It takes 10 times or more the processing time, and it takes 10 days for a single molecular orbital calculation. As can be easily understood from the above description, the conventional molecular orbital calculation that uses a single computer to store and calculate integral calculation values in a high-speed memory involves molecular orbital analysis of molecules that are commonly used in the chemical and pharmaceutical industries. There was a problem that was impossible.

Now, in the actual molecular orbital analysis, after the electron integral calculation, a calculation process for solving the SCF (Self Consistent Field) equation and calculating the expansion coefficient Cj shown in the equation (1) to determine the molecular orbital is necessary. is there. This step is performed as follows.

First, the values of the one-electron integral and the two-electron integral calculated and stored in the electron integral calculation step are used (5).
The Fock matrix element Fuv represented by the equation is calculated.

Fuv = Huv + ΣΣPds [(uv | ds) −0.25 (ud | vs) −0.25 (us | vd)] (5) where Pds represents an element of a square matrix P having a dimension of the total number of atomic orbitals. Since it corresponds to the electron density of each atomic orbital,
Is called the density matrix. Specifically, it is given as in equation (6).

Pds = 2ΣCdi * Csi (i = 1, 2,..., N) (6) However, the sum regarding i is taken up to the occupied molecular orbital N. At the beginning of the calculation, the Fock matrix is obtained by the equations (6) and (5), assuming appropriate initial values for the molecular orbitals, ie, Cdi and the like. As the initial value, an expansion coefficient based on an empirical molecular orbital method such as the Huckel method or a semi-empirical molecular orbital method called the MINDO method is often used. The step shown as initial vector calculation in FIG. 1 corresponds to this.

Next, a matrix equation (7) consisting of the Fock matrix F and the overlap integral S obtained above is solved.

Σv (Fuv−eiSuv) Cvi = 0 u = 1, 2, 3,..., N (7) Thus, a total of n molecular orbitals including N occupied molecular orbitals | Yi> Ei and Cvi are obtained as candidates for the orbital energy and the expansion coefficient of. Here, not only the N occupied molecular orbitals but also the (nN) empty molecular orbitals are obtained because the expansion coefficient of the empty molecular orbital is calculated to accelerate the solution of the SCF equation or to calculate the physical properties of the molecule. Because it is necessary. Using the expansion coefficients of the N occupied molecular orbitals, the density matrix and the orbital energy are repeatedly calculated by the equations (6) and (7), and when the change becomes smaller than a specified value, the calculation is considered to have converged and the calculation is terminated. .

This step corresponds to the step shown as SCF equation calculation in FIG. The molecular orbital | Yi> corresponding to the orbital energy ei is as shown in the equation (8), and the molecular orbital calculation is completed by using this molecular orbital to determine the binding property, charge distribution and other physical properties of the material.

| Yi> = Σ | Xj> Cji (8) As described above, in the step of obtaining the molecular orbital from the Fock matrix, it is necessary to repeatedly solve the linear equation described by the equation (7) until the density matrix converges many times. is there. In ordinary molecular orbital calculations, the density matrix does not converge unless this calculation is repeated several tens to several hundred times. Since the time to solve the linear equation increases approximately in proportion to the cube of the total number of atomic orbitals, for the molecules handled in the chemical and pharmaceutical industries, the above-mentioned electronic integration calculation, that is, the calculation time of the Fock matrix element In addition to the increase, the time required for solving the linear equation also increases.

As in the above-mentioned ab initio molecular orbital calculation,
The amount of data and calculation time required for matrix element calculation and the calculation time for linear operation of the matrix are 3 times the size of the system to be solved.
The phenomenon of a sharp increase in proportion to the power or the fourth power also occurs in other material simulations such as the finite element method or in scientific and technical calculations.

[0023]

As described above, since the amount of data and the calculation time for molecular orbital calculations and other scientific and technical calculations rapidly increase with the size of the system to be calculated, they are actually used. Computer simulation of large-scale systems such as materials and chemicals has not been possible until now. However, in recent years, the recognition that the shortening of new product development time is a necessary condition for the market to become dominant has been increasing, and large-scale scientific and technical Has become a major issue.

A first object of the present invention is to execute a large-scale scientific and technical calculation with a realistic calculation time and computer resources,
It is an object of the present invention to provide a computer system capable of shortening a development period by applying a computer simulation to a development process of a new product, a new material or a medicine. In particular, a molecular orbital analysis system that performs ab initio molecular orbital calculations with realistic time and computer resources for molecules handled in the chemical and pharmaceutical industries, and enables prediction of material properties and pharmacological effects of molecules. To provide.

A second object of the present invention is to reduce computation time and reduce computer resources by simultaneously and concurrently performing computation processing in an ab initio molecular orbital calculation using both a computer cluster and a vector computer. It is an object of the present invention to provide a calculation algorithm which enables efficient use of a program.

[0026]

A first object of the present invention is to calculate a matrix element in a computer cluster (parallel computer), transfer the calculation result to a vector computer, and perform a linear operation of the matrix on the vector computer. This is achieved by: Depending on the calculation method, the communication overhead of data transfer between the computer cluster and the vector computer becomes large, but even in that case, the total calculation time is obtained by directly connecting the computer cluster and the vector computer via a channel without passing through a normal network. Need not be increased. In particular, the first object of the ab initio molecular orbital calculation is to provide a one-electron integral, a two-electron integral and a Fock matrix element composed of the one-electron integral and the two-electron integral for a given material. A computer cluster consisting of a plurality of computers connected to the computer cluster, a part of the one-electron integral generated by the computer cluster, and communication means for transferring a Fock matrix factor to a vector computer having a pipeline processing function;
A Fock matrix is generated from a part of the one-electron integral transferred via the communication means and a Fock matrix factor, eigenvalues and eigenvectors of the Fock matrix are determined, and a molecular orbital energy and a wave function of a given material are output. This is achieved by using a computer system for molecular orbital analysis composed of a vector computer.

The first object is that the communication means for transferring the computer cluster generation data to the vector computer is an input / output channel device directly connected to the main storage of the vector computer. Achieved more effectively.

Further, the first object can be more effectively achieved by configuring the computer cluster by an arbitrary combination of a personal computer and a work station.

The second object is to create a density matrix at the time when all the eigenvectors corresponding to the occupied molecular orbitals are obtained in the process of solving the eigenvalues and eigenvectors of the Fock matrix on the vector computer, transfer the density matrix to the computer cluster, and then transfer the vector to the computer cluster. This is achieved by continuing calculation of eigenvectors corresponding to empty molecular orbitals while continuing to calculate new Fock matrix elements using the density matrix transferred on the computer cluster. FIG. 2 shows the flow of this processing. The processes (6), (7), and (8) in FIG. 2 are performed on a vector computer. If it is determined in the process (8) that the density matrix has not converged, the density matrix is transferred to the computer cluster, and the processes (4) and (5) are started on the computer cluster. In the meantime, the processing (9) is performed on the vector computer, and the determined empty molecular orbital is used for the convergence acceleration in the processing (5). These processes are repeated until the density matrix converges (until Yes is determined in process (8)).

According to the present invention, the calculation of matrix elements in scientific and technical calculations is accelerated by parallel processing on a computer cluster, and the linear operation of a matrix is accelerated by pipeline processing on a vector computer. Calculation time is greatly reduced. For example, in a molecular orbital calculation, a plurality of interconnected one-electron integrals and two-electron integrals for a given material to be generated and a Fock matrix element composed of the one-electron integrals and the two-electron integrals are generated. Distributed execution by a computer cluster consisting of computers, generated electron integration,
In addition, the Fock matrix elements are transferred to the vector computer, and the linear calculation required for solving the eigenvalues and eigenvectors of the Fock matrix is performed on the vector computer, thereby realizing a realistic time for molecules handled in the chemical and pharmaceutical industries. And computer resources to execute the ab initio molecular orbital calculation, and thus the prediction of material properties and pharmacological action of these materials.

According to the present invention, in the ab initio molecular orbital calculation, the SCF calculation repeated at least several tens of times until the convergence of the density matrix, that is, the calculation processing of the Fock matrix elements and the calculation processing of the eigenvalues and eigenvectors of the Fock matrix are performed. Since the units can be executed in parallel by the computer cluster and the vector computer, the calculation time can be further reduced.
This effect increases as the quality of the basis set used for expanding the molecular orbital increases. This is because the number of empty molecular orbitals increases as the number of basis functions increases.

[0032]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 3 is an overall configuration diagram of an embodiment of a calculation system for molecular orbital analysis according to the present invention. Computers 11, 1
A computer cluster 1 composed of 2, 13, 14, 15, and 16 is connected to the vector computer 2 via a computer cluster / vector computer connection transmission line 31. The computer cluster 1 includes six computers 11, 12,
13, 14, 15, and 16, and are interconnected by a computer cluster connection line 30. In the present embodiment, the computers 11, 12, 13, 14, 15, 16
A personal computer in which an Intel 486DX4 processor was operated at a clock frequency of 100 MHz was used. UNIX was set as the OS of the personal computer. The vector computer 2 used was a Hitachi S810 model 20. This computer is a vector computer in which four 143 MFLOPS pipeline processors are arranged in parallel. Next, for the connection line 30 in the computer cluster and the transmission line 31 for the vector computer connection, Ethernet which is frequently used as a communication line for LAN at present is used, and T is used as a communication protocol.
Data communication was performed between personal computers and between computer clusters and vector computers using CP / IP. In this embodiment, a computer cluster 1 is constituted by six computers.
However, as long as there is no problem such as packet collision in communication between computers, there is no particular limitation on the number of computers.

Next, specific calculation of molecular orbitals according to the present embodiment will be described. GAMES for molecular orbital calculations
S, using distribution software such as Gaussian (for details, see, for example, Abu Inishio Molecular-Obitarseori [HEHRE et al, AB INITIO MO
LECULAR ORBITAL THEORY; Wiley-Interscience, published in 1986] and other textbooks on quantum chemistry). The programs constituting the computer cluster 1 and the vector computer 2 are installed in advance in such a manner that they can be executed.

Since the specific procedure of the molecular orbital calculation is as shown in FIG. 1, a part related to the present invention will be described in some detail.

First, the structure of a given material, for example, in the case of a molecule, the input of its molecular geometry, molecular symmetry, basis set, etc. are input. In many cases, the relevant data is input in an input format specific to the molecular orbital calculation program to be used. This is done by using one of the personal computers constituting the computer cluster 1. May be input using an editor such as Vi. The basis function input to the molecular orbital calculation system is an atomic orbital function having a center on each atom constituting a given material. For this atomic orbital function, it is necessary to calculate the one-electron integral and the two-electron repulsion integral shown in the equations (2), (3), and (4). This is the repulsive integral (uv | ds). In this molecular orbital analysis computer system, this integral calculation is distributed on each personal computer constituting the computer cluster. The distribution is adjusted so that the load on each personal computer is even. Specifically, the two-electron integral index (ijkl) for all atomic orbitals of a given material is evenly distributed to each personal computer, and the corresponding two-electron integral (ij | kl) is separately stored on each personal computer. Is calculated. Of course (ij | k
(kl | ij), which has exactly the same value as l), is naturally calculated only once, thus avoiding duplication. This distributed processing can be easily achieved by distributing a two-electron integral index (ijkl) using one of the computers constituting the computer cluster as a master. The two-electron integral generated in this way is transferred to the vector calculator 2 via the transmission line 31, and the Fock matrix element Fuv represented by the equation (5) is calculated. Next, the solution of the linear eigenvalue equation represented by the equation (7) is repeated on the vector computer, and the molecular orbital and the orbital energy are obtained. This process is different from the electron integration calculation described earlier,
Since the eigenvectors and eigenvalues are sequentially obtained by using the density matrix of the previous stage as input, if distributed processing is performed by multiple computers, it becomes extremely inefficient due to communication lag, etc.
Processing is performed on a single computer. The processing for solving the eigenvalue problem of the Fock matrix can of course be performed on a personal computer constituting a computer cluster. However, in general, a personal computer can perform only numerical operations by sequential processing, and is a pipe for so-called matrix operation. Since line processing cannot be performed, a great deal of time is required for the solution. On the other hand, according to the vector computer 2, the matrix eigenvalue can be obtained at high speed by pipeline processing operation, and the molecular orbital and energy can be obtained efficiently. For example, in a high-speed operation by a vector computer, a "supercomputer (edited by the Physical Society of Japan, published by Baifukan, 198
5) Since it is described in detail in a textbook explaining the numerical calculation method such as "", it will not be described again here.

Next, the results of calculations performed using the molecular orbital analysis calculation system described in this embodiment will be described. GAMESS with direct SCF
Was used. Take a large basis function, more specifically 6
Calculation of the SiH2C2H4 molecule using a basis function called -31G * (for example, see Abinitio Molecular-Obital Seoli, supra) (the total number of basis functions is 61 basis) gave the following results. Total CPU time is 53
Seconds, of which CPU time required for integration calculation including two-electron integration was 50 seconds, and CPU time required for iterative processing for obtaining eigenvectors and eigenvalues was 3 seconds. On the other hand, an iterative process for finding an eigenvector and an eigenvalue is performed by a vector computer 2
Was performed using one of the personal computers constituting the computer cluster 1 without using
It took 9 seconds of CPU time. When the integral calculation including the two-electron integral was performed using only the vector computer 2 without using the computer cluster 1, the CP required for this was calculated.
The U time was 96 seconds. From these results, it was found that when the calculation system for molecular orbital analysis shown in the present example was used, the calculation time was greatly reduced in both electron integral calculation and eigenvalue problem solving. In this example, the results were shown for relatively small molecules such as SiH2C2H4 molecules. However, in the case of large molecules used in the chemical and pharmaceutical industries, the effect of shortening the calculation time becomes more remarkable.

Next, another embodiment of the calculation system for molecular orbital analysis according to the present invention will be described with reference to FIG. In this embodiment, unlike the previous embodiment, the transmission line for connecting the computer cluster 1 and the vector computer 2 is replaced with an Ethernet, and a channel input / output device 21 for directly transferring data to the vector computer main memory 22 is provided. The configuration uses a transmission line 32 for connecting a channel input / output device. In this configuration, since integration data and Fock matrix elements can be directly transferred to the main memory 22, an increase in TAT (turn around time) time due to transmission can be suppressed. The result of the molecular orbital calculation using this embodiment will be described. The test was performed on the SiH2C2H4 molecule calculated in the previous example. In the configuration shown in FIG. 3, the total CPU time is 53 seconds and the TAT time is 195 seconds. However, according to the configuration shown in this embodiment, the total CPU time is 53 seconds.
The TAT time could be reduced to 80 seconds, about one third, which was the same as the second.

As described in the above two embodiments, according to the present invention, both the CPU time and the TAT time can be reduced as compared with the conventional molecular orbital calculation using a single computer.

Next, an embodiment of a material analysis / design system using the molecular orbital analysis calculation system according to the present invention is shown in FIG.
This will be described with reference to FIG. The present embodiment is a material analysis / design system in which a material analysis / design input / output unit 4 is provided in addition to a molecular orbital analysis calculation system including a computer cluster 1 and a vector computer 2. The material analysis / design input / output unit 4 includes a display display 41, an input keyboard 42, a mouse 43,
The input / output unit computer 45 for material analysis / design is constituted. The material analysis / design input / output unit computer 45 has a function of generating and correcting a molecular structure and a crystal structure and generating display image data based on the design information input by the input keyboard 42 and the mouse 43. It goes without saying that the input / output unit computer 45 for material analysis / design can be replaced by a computer constituting the computer cluster 1.

Next, an enzyme reaction analysis performed using the material analysis / design system will be described. First, each step of a general enzyme reaction analysis procedure will be described with reference to FIG. The first step is to specify an enzyme molecule to be analyzed using the input keyboard 42 or the mouse 43. When specifying using a keyboard, it is possible to specify the specification number of the enzyme molecule, for example, the PDB (protein data bank) collection number compiled by the Brookhaven National Laboratory in the United States. For example, in the case of tryptophan synthase which catalyzes the reaction of synthesizing tryptophan from indole and serine, 1 W
Just type SY. PDB (Protein Data Bank)
The collection number is a protein code number used worldwide, and the correspondence with each protein is unique. In the material analysis / design system according to the present invention, PDB information is stored on a hard disk in the input / output unit computer 45 for material analysis / design. Specific PDB entry information corresponding to each code number includes a sequence of amino acid residues constituting an enzyme molecule, a heteroatom coordinate, an SS binding site, a helical structure, and higher-order structural information such as a sheet structure. The computer 45 generates three-dimensional information and binding information of the enzyme molecule from these information and displays them on the display display 41. When specifying an enzyme molecule with a mouse, an enzyme name such as “Tryptophan synthase” can be specified from a pull-down menu shown on the display 41.

The next step is to specify the substrate molecule. Regarding the substrate molecule, there is no one corresponding to the PDB code number, and the molecule is formed on the display 41 using a builder. In recent years, various distribution software has been developed and used for the builder function and the molecule generation processing, and a description of the contents thereof is omitted here.

The next step is to move the enzyme molecule and the substrate molecule displayed on the display 41 into an appropriate positional relationship, that is, move the substrate molecule into the reaction pocket of the enzyme molecule (dotting). In this system, the substrate molecule is dragged by the mouse 43 on the display 41,
This is done by moving it to the reaction pocket of the enzyme molecule. As described above, the functions of the molecular orbital analysis calculation system including the computer cluster 1 and the vector computer 2 are not used at all in the steps from the designation and display of the enzyme molecule, the designation and display of the substrate molecule, and the docking operation.

The next step is to analyze the reaction coordinates of the (enzyme-substrate) complex by the molecular orbital method. In this embodiment, the eigenvalue tracking method (Eigenvalue Following)
The reaction coordinates are determined by uphill in the direction of the maximum gradient of the complex potential surface by the method (enzyme-substrate). Regarding the procedure of eigenvalue tracking, refer to research papers on chemical physics (eg, Journal of Chemical Physics, 8
0, page 2464, published by the American Physical Society, 1984), and further description is omitted here. The uphill change in the direction of the maximum gradient of the potential surface can be interactively tracked as a change in the (enzyme-substrate) complex structure on the display 41 by the high-speed analysis processing function of the molecular orbital analysis calculation system. At each stage of tracking, normal vibration analysis of the (enzyme-substrate) complex is performed, and the entire vibration mode is also tracked and displayed. By sequentially performing this step, an active complex of the (enzyme-substrate) complex, a so-called transition structure can be obtained. The transition structure is defined as a primary saddle point on the complex potential surface, and can be easily determined from the fact that the frequency of the reference vibration in the reaction coordinate direction is imaginary. At the stage when the transition structure is determined, the energy is compared with the initial state, and the activation energy is obtained. Activation energy is an index that can determine the degree of difficulty of the progress of the reaction, and it is one of the very important functions for a material analysis / design system that it is required to be used at high speed and interactively. Further, a reaction product can be obtained by conducting a reaction coordinate analysis of the (enzyme-substrate) complex. This step is the final step of the enzyme reaction analysis.

Next, the results of applying the above-described general enzyme reaction analysis procedure to specific enzyme reactions are summarized below. In the present example, a peptide bond cleavage reaction by an alpha-chymotrypsin (a-chymotrypsin) enzyme was analyzed and traced using the material analysis / design system according to the present invention. Oligopeptide RB if the number of residues is 20 in the molecule to be the reaction partner of alpha chymotrypsin, that is, the substrate peptide molecule
NHC1 (O) RA was chosen and the peptide bond cleavage reaction between RA-RB was followed. The alpha chymotrypsin enzyme is an enzyme having 245 residues and a total of 3543 atoms (excluding hydrogen atoms), and has a PDB registration number of 4CHA. PDB registration number
After inputting 4CHA from the keyboard 42, the structure of the alpha chymotrypsin molecule displayed on the display 41 is shown in FIG. In this figure, a ribbon diagram is shown to clearly show the higher-order structure of the enzyme. The ribbon diagram is a display method in which a protein structure such as a helix and a sheet structure is displayed by a tube model or a strip model, and is a display method in which a higher-order structure of the protein can be clearly recognized. As shown in FIG. 7, alpha-chymotrypsin has a partial structure (called a domain) which is divided into upper and lower portions on the screen, and a pocket space sandwiched between the two is the enzyme active region 50. The enzyme active region 50 is a central region that reacts with the substrate peptide molecule. Oligopeptide following setting and display of alpha chymotrypsin molecule
RBNHC1 (O) RA was generated and displayed on the display 41. First, the names of the residues constituting the oligopeptide RBNHC1 (O) RA are sequentially selected from the pull-down menu using the mouse 43, designated, and the amide bond NHC connecting the residues is used.
(O) is added and the structure is assembled.At the final stage, the structure is optimized by the molecular force field method and the oligopeptide RBNHC1 (O) RA
Generated. Next, the present oligopeptide additionally displayed on the display 41 was dragged to the position of the enzyme active region 50, and docking (enzyme-substrate) was advanced. Upon docking, the oligopeptide RBNHC1 (O) R was located at about 195 residues of serine, 57 residues of histidine, and 102 residues of aspartic acid constituting the enzyme active region 50 of alpha chymotrypsin.
The A bond cleavage site was positioned so as to make contact. The reaction mode of alpha chymotrypsin which has been roughly known so far will be described with reference to FIG. This is, for example, "Enzyme Struc
ture and Mechanism (by A. Fersht, FREEMAN Bookstore, 19
1985, New York), pp. 405 or less, and other textbooks on biochemistry. The alpha chymotrypsin molecule is an enzyme active region 5 in a single state.
The OH portion of serine 195 residue within 0 forms a hydrogen bond with 57 residues of histidine, and is in a stabilized state [State A]. Next, when the peptide approaches the enzyme active region 50, serine 195 is detected.
Hydrogen in residue OH moves to histidine residue, serine 195
Residue O forms an active complex bound to the amide bond carbonyl carbon at the cleavable bond of the substrate molecule peptide [State
B]. In this state, the hydrogen atom previously transferred to histidine attacks the amino terminus of the substrate peptide, releases the cleaved peptide and generates an enzyme intermediate, ending the first step of the reaction. The alpha chymotrypsin molecule is sometimes called a serine protease because 195 serine residues are located in the active center. In this embodiment, [state A], [state B], [state C]
Was tracked by molecular orbital calculations to see if the changes could actually occur. When the uphill change (reaction coordinate analysis step in FIG. 6) in the direction of the maximum gradient of the potential surface is advanced, the numerator using the parallel processing between the computer cluster and the vector computer of the present invention to track the change of one step. The orbit analysis calculation system required 12 hours. Without parallel processing between the computer cluster and the vector computer, the same calculation required 18 hours. As described above, the alpha chymotrypsin molecule is an enzyme having 245 residues and 3543 total atoms (excluding hydrogen atoms). Therefore, the reaction analysis for all atoms is a heavy load even though the molecular orbital analysis calculation system according to the present invention is fast, and it is not easy to proceed interactively with the calculation process up to the generation and display of the reaction product. It turned out. Therefore, enzyme active region 5
The enzymatic skeleton constituting 0 is cut out and the oligopeptide RBNHC1 is added to a model alpha chymotrypsin with a reduced total number of atoms.
(O) We decided to apply RA. Upon excision, 195 residues of serine, histidine 57
Residues, an enzyme skeleton A consisting of 102 residues of aspartic acid,
Enzyme skeleton B and enzyme skeleton C were divided into three skeletons, which were arranged so as to maintain the spatial configuration of the active region 50. That is, each enzyme skeleton is connected by a structural linker consisting of dummy atoms,
The end of the skeleton was not moved. Since the dummy atoms are excluded from the calculation at the stage of the molecular orbital calculation, they do not hinder the reaction analysis. The total number of atoms of the model alpha chymotrypsin was about 200 (excluding hydrogen atoms), and the number of atoms of the oligopeptide RBNCH1 (O) RA was about 150 (excluding hydrogen atoms). In fact, in this case, one step of uphill change in the direction of the maximum gradient of the potential surface is greatly reduced to 19 seconds by the molecular orbital analysis calculation system using parallel processing between the computer cluster and the vector computer according to the present invention.
It is now possible to interactively track structural changes on the display 41. When parallel processing between the computer cluster and the vector computer is not used, 40
Seconds needed. After each uphill change step, a vibration analysis was performed, the frequency was checked, and the transition structure was confirmed. FIG. 9 shows a schematic diagram of a transition structure obtained through about 70 uphill change steps. The transition structure is a serine-C-CH2-O group extending from the enzyme skeleton A bonded to the central amide bond C1 of the oligopeptide RBNCH1 (O) RA, forming a tetrahedral structure centered on the C1 atom, and further comprising the enzyme skeleton B A hydrogen bond to has occurred. It was also found that the oligopeptide RB chain formed a sheet structure with the enzyme skeleton C and contributed to stabilization. As described above, by using the material analysis / design system according to the present invention, it has become possible to interactively trace an enzyme reaction mechanism which has not been out of the conventional estimation range.

In the present embodiment, the effectiveness of the material analysis / design system according to the present invention was shown by taking an enzymatic reaction as an example. It goes without saying that it is just as effective. Also in other computer simulations such as the finite element method, the present invention is effective in increasing the processing speed as in the molecular orbital method.

[0047]

According to the present invention, it is possible to greatly increase the speed of scientific and technical calculations, and to shorten the development period by applying computer simulation to product development. In particular, it is possible to execute ab initio molecular orbital calculations with realistic time and computer resources for molecules handled in the chemical industry and the pharmaceutical industry, and to predict material properties and pharmacological effects of these materials.

[Brief description of the drawings]

FIG. 1 shows schematic steps of ab initio molecular orbital calculation.

FIG. 2 shows a method for calculating molecular orbitals according to the present invention.

FIG. 3 is a schematic diagram showing the configuration of an embodiment of a computer system according to the present invention.

FIG. 4 is a schematic diagram showing the configuration of another embodiment of the computer system according to the present invention.

FIG. 5 is a schematic diagram showing one embodiment of a material analysis / design system using a molecular orbital analysis calculation system according to the present invention.

FIG. 6 is a schematic step of an enzyme reaction analysis procedure.

FIG. 7 is a schematic diagram of an enzyme higher-order structure displayed on a display.

FIG. 8 is a schematic diagram of a reaction mode of an enzyme.

FIG. 9 is a schematic diagram of a transition structure of an enzyme reaction.

[Explanation of symbols]

1 ... computer cluster, 2 ... vector computer, 11,
12, 13, 14, 15, 16 ... interconnected computers forming a computer cluster, 21 ... channel input / output device, 22 ... main storage device, 23 ... vector register, 24 ... vector arithmetic processor , 25 scalar register, 26 scalar operation processor, 30 connection line in a computer cluster, 31 transmission line for connecting a computer cluster, vector computer, 32 transmission line for connecting a channel input / output device, 4 ... … Input / output unit for material analysis / design, 41
…… Display display, 42 …… Input keyboard, 43
...... Mouse, 45 ... Material analysis / design input / output unit computer.

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Jurgen Schulti 1-280 Higashi Koigakubo, Kokubunji-shi, Tokyo Inside Central Research Laboratory, Hitachi, Ltd.

Claims (4)

[Claims] 1. A plurality of interconnected calculation means having a function of generating a one-electron integral, a two-electron integral, and a Fock matrix element composed of the one-electron integral and the two-electron integral for a given material. A communication means for transmitting a part of the one-electron integral generated by the calculation means cluster and the Fock matrix element to a vector calculation means having a pipeline processing function; Generating a Fock matrix from a part of the one-electron integral transmitted through the Fock matrix element,
A calculation system for molecular orbital analysis, comprising: the vector calculating means for solving eigenvalues and eigenvectors of the Fock matrix and outputting molecular orbital energies and wave functions of the given material. 2. The apparatus according to claim 1, wherein said communication means for transmitting said calculation means cluster generation data to said vector calculation means is an input / output channel device directly connected to a main storage device of said vector calculation means. 2. The calculation system for molecular orbital analysis according to 1. 3. The molecular orbital according to claim 1, wherein each of the calculation means constituting the calculation means cluster comprises a personal computer, a workstation, or any combination thereof. Analysis calculation system. 4. A computing means cluster (parallel computing means) comprising a plurality of computing means interconnected to process matrix elements at a high speed, for performing a linear operation on a matrix obtained by said computing. In a calculation system including a vector calculation unit having a pipeline processing function, and a data transfer communication unit between the calculation unit cluster and the vector calculation unit, the vector calculation unit calculates a solution of an eigenvalue of a Fock matrix and an eigenvector. Calculating the density matrix necessary to create a new Fock matrix at the time when all the eigenvectors corresponding to the occupied molecular orbitals are obtained, transferring the density matrix to the calculation means cluster; Start generation of matrix elements and continue to hit empty molecular orbitals by the vector calculation means A molecular orbital calculation method, wherein eigenvectors are solved, and a new Fock matrix element is generated in the calculation means cluster during the solution.