JP2005234699A - Apparatus and method for estimation of spatial structure of protein by cellular automaton - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a computerized technique for a spatial structure analysis of protein that helps efficient drug discovery by shortening a time span bottlenecking existing techniques while ensuring practical accuracy. <P>SOLUTION: A proposed model based on a cellular automaton method is a discrete three-dimensional model with layered hexagonal lattice planes, and unit cells have information on a contained amino acid and its accompanying charge and mass. The transition of the unit cells depends on a vector sum from the surrounding cells. The mobility of the unit cells also varies with temperature. The framework represents the structure of protein. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a technique for predicting the three-dimensional structure of a protein.

  Recently, human genome analysis has been completed, and drug discovery using the genetic information has attracted a great deal of attention among those concerned. Because it uses fundamental information, such as genetic information, instead of something that is per se, it is expected that it will reduce drug development costs and create innovative drugs that have never existed before. Because it is done. Pharmaceutical industry officials often refer to drug discovery as gambling. This is because even after a long R & D period of more than 10 years, the probability that a new compound will be approved for a new drug is extremely low. In general, when a new drug is marketed, 5 out of 100 seeds will reach clinical trials, and 1 of 5 will eventually be on the market. According to a survey by the Japan Pharmaceutical Manufacturers Association (Pharmaceutical Association), a voluntary organization of pharmaceutical companies, it was possible to obtain approval from among 432,312 new drug candidate compounds synthesized by 18 Japanese pharmaceutical companies from 1997 to 2001. There are 36 compounds. The probability was really 1/2004. This low probability does not seem to improve at all compared to the previous one. Each drug costs 40 billion yen on average and takes 12 years of research and development, 60% of which is said to be the cost of clinical trials. Therefore, it is essential for pharmaceutical companies to create better seeds, reduce the cost of clinical trials, and shorten research and development periods. The key to this is the protein, gene, and genome information industry, the so-called bioinformatics (Bioinformatics) industry. This makes it possible to search for new seeds and shorten the research and development period. Here, in order to confirm at which stage drug discovery using bioinformatics is being performed, the flow of drug discovery using genome information is summarized below.

・ Gene discovery ・ Functional analysis of genes ・ Identification of drug discovery and treatment targets ・ Identification and optimization of lead compounds ・ Drug metabolism and safety studies ・ Clinical trials ・ Approval ・ Commercialization As mentioned above, research has started in the past. Since then, a period of 10 to 15 years has been required for drugs to enter the market. However, by utilizing IT infrastructure and bioinformatics, the development of (1) to (4) is efficiently performed, and it is said that the period can be shortened by almost 10 years. We will look specifically at each stage below.
The discovery of genes in (1) aims to identify genes and proteins that are likely to be targets for drug discovery and treatment. It is said that about 3 to 40,000 genes are scattered in about 3 billion human genome sequences, and bioinformatics is used to discover the genes. This estimates the position and sequence of the gene. An experimental technique called cDNA analysis is then used for the estimated gene. Thereby, accurate gene identification is possible. The data analyzed in this step is stored in public or private data banks.
In the gene function analysis in (2), in order to narrow down genes targeted for drug discovery and treatment from genes discovered on the genome, the correlation between genes and specific diseases, the role in the body, The purpose is to clarify the function. As a method for analyzing this function, a DNA microarray, SNP data, or the like is used.
In (3) drug discovery and treatment target identification, it is very important to determine the three-dimensional structure of the protein. As the method, NMR and X-ray crystal structure analysis are experimentally used, and bioinformatics is used computationally.
In the identification and optimization of the lead compound in (4), it is objectively proved whether or not the target gene and candidate substance produced in the processes (1) to (3) are correct. At this stage, studies on genes from various directions including genome information databases, medical literature databases, bioinformatics using IT infrastructure, and experiments using gene chips will be conducted. Derivatives are synthesized by applying various chemical modifications to lead compounds that have been confirmed to be effective through these processes.
In the drug metabolism and safety test of (5), the effects and safety of the drug are confirmed by conducting animal experiments.
In the clinical trial (6), drugs are administered to humans to confirm the efficacy and safety.
In the approval of (7), the experimental data obtained in (6) is submitted to a national institution, and if it is confirmed to be effective, it is allowed to be marketed.
In the market of (8), when a drug is actually put on the market and an unexpected side effect occurs, the drug may be collected from the market.

Proteins are macromolecules in which amino acids are linked in a chain, have various functions in the body, some form the body itself, and some are enzymes that are responsible for the smooth progression of physiological reactions. . The functions of these diverse and sophisticated proteins that enable the maintenance of complex life activities are actually closely related to the three-dimensional structure, and the three-dimensional structure analysis of biopolymers inevitably plays a particularly important role. It is. This is particularly important in step (3). Accurate understanding of the three-dimensional structure of proteins in drug discovery facilitates narrowing down of drug candidate substances, and the development period is expected to be greatly shortened. This three-dimensional structure is determined by physicochemical methods such as NMR and X-ray crystal structure analysis, but it takes time, manpower and cost. Therefore, there has been a demand for finding a promising protein by a computer and performing only its three-dimensional structure by an experimental method. Therefore, existing methods for protein tertiary structure prediction have begun to be used, but time and accuracy have become a bottleneck.
Susumu Okazaki, Yuki Okamoto, “Computer simulation of biological systems-how far can we predict the structure of proteins?” Kagaku Doujin, 2002, p. 12-49 (Chapter 1, Chapter 2), p. 98-109 (Chapter 8) Kato Yasuyoshi, Mitsunari Tomotaka, Tsukiyama Hiroshi, "Cellular automaton method-Self-organization and massively parallel processing of complex systems" Morikita Publishing, p. 17-27

  An object of the present invention is to reduce the time that has become a bottleneck in existing methods while maintaining practical accuracy, and to assist in efficient drug discovery.

The three-dimensional structure estimation method according to the present invention includes:
This is performed on the structure of a substance composed of a plurality of connected elements using a computer, and the space where the target structure exists is divided, the state is embedded in the divided space, and the structure is made using mathematical formulas. The whole process is estimated while controlling the formation process.

  The element is an amino acid.

  The substance is a protein.

  The space is divided by a lattice.

  The lattice is a hexagonal lattice.

  The lattice is overlapped in layers.

  As the state, information including presence / absence of an element, a charge when the element is present, and a position of an adjacent element is stored.

  The mathematical formula is a metropolis standard of annealing method.

  The mathematical expression is a function of a vector (F) and a control parameter (T).

  The vector (F) is a Coulomb force.

  The state is changed so that an element is moved when the mathematical formula exceeds a threshold value.

  The threshold value is calculated by a function.

  The function generates a random number.

  The random number is a uniform random number.

であることを特徴とする。 The formula is

It is characterized by being.

  The state is based on a local rule.

  The state depends on the state in the vicinity.

  It is characterized in that the Coulomb force is calculated in the nearest neighborhood or a range of a constant multiple thereof.

The three-dimensional structure estimation apparatus according to the present invention is
This is performed on the structure of a substance composed of a plurality of connected elements using a computer, and the space where the target structure exists is divided, the state is embedded in the divided space, and the structure is made using mathematical formulas. The whole process is estimated while controlling the formation process.

  While maintaining practical accuracy, the time that has become a bottleneck with existing methods can be shortened, and this can be an aid for efficient drug discovery.

Embodiment 1 FIG.
The experimental method will be described. An X-ray crystal structure analysis method is widely used as an experimental method for the three-dimensional structure analysis of proteins. However, the problem is that a large amount of pure purification of the target protein is required, and the crystal structure is observed instead of the structure in the original solvent (mainly water). In recent years, NMR methods and direct observations using an electron microscope have been performed.

  The computer method will be described. In recent years, with the remarkable development of computational capabilities, structural predictions using computational methods have been attempted. They are roughly classified into those that do not use structural information of proteins with known structures and those that do not use structural information. The former representative is the homology method. Representative of the latter are the Monte Carlo method and the molecular dynamics method, which are methods for searching for the most stable three-dimensional structure using only amino acid sequence information from a random initial three-dimensional structure. The former cannot be used without similarity to known structural information, but the latter does not have such restrictions.

本手法のモデルは、ＣＡ法からヒントを得ている。以下では、ＣＡ法との対応関係を考慮にいれながら、提案モデルについて順に述べる。
１．解析空間の設定
はじめに設定しなければならないのは状態空間をどのように定義するかである。本手法では、六方格子面を層状に重ねた離散的３次元モデルを用いている。図１に解析空間の分割図を示す。
２．単位セルの保持する情報
次に、上記のように分割された単位セルが保持している情報について述べていく。単位セルには以下のような情報が含まれている。
（１）どのアミノ酸が含まれているか（もしくは、アミノ酸が存在しないか）。
（２）アミノ酸の種類に付随した電荷および質量はいくらか。
３．近傍の定義
近傍１の定義は、同一平面上（六方格子面上、ｘ−ｙ平面）では１つ隣のセル（図２）、高さ方向（層、Ｚ軸）では上下セルおよびこれらのセルの１つ隣のセルである（図２）。近傍の大きさはシミュレーション実験ごとに設定できる。このモデルを用いたのは、これによりＸ−Ｙ平面で６０ｎ_１度、Ｚ軸方向で９０ｎ_２度、斜め上（もしくは下）方向で４５ｎ_３度が表現でき、これらの組み合わせによりアミノ酸の結合角度をある程度表せると考えたからである。ただし、ｎ_１＝１，・・・，５、ｎ_２＝１，・・・，３、ｎ_３＝１，・・・，７、である。これによって、タンパク質の構造で重要な２次構造のαへリックス、βシートが表現できる。
４．遷移規則
任意のセルに対して適用される遷移規則を以下で定義する。
（１）近傍として設定された範囲にあるアミノ酸から受けるベクトル和によって、セル情報を移動させる方向の候補を決定する。アミノ酸の電荷Ｚは、実際に特定されている値を使っている。ただし、電荷Ｚが０のアミノ酸に対しては、後述の揺らぎ定数で表現している。任意のアミノ酸が受けるベクトル和は、式（５）

の第１項を式（６）

に代入する事により得られる。すなわち、 Next, the cellular automaton method will be described. In the cellular automaton method, the analysis target is divided into divided regions called cells, and the discrete state quantities defined on each cell are defined by local neighborhood rules provided between neighboring cells. This is a phenomenon modeling method that changes state quantities. This local neighborhood rule makes it possible to represent the phenomenon of the entire system. This local neighborhood rule can be set arbitrarily, and if it represents a physical phenomenon, it can also be treated as a neighborhood rule. This technique is used for fluid analysis and the like.

The model of this method is inspired by the CA method. Below, the proposed model will be described in order, taking into account the correspondence with the CA method.
1. Setting the analysis space The first thing you need to set is how to define the state space. In this method, a discrete three-dimensional model in which hexagonal lattice planes are layered is used. FIG. 1 shows a division diagram of the analysis space.
2. Information held by the unit cell Next, information held by the unit cell divided as described above will be described. The unit cell includes the following information.
(1) Which amino acids are contained (or which amino acids are not present).
(2) What is the charge and mass associated with the type of amino acid?
3. Definition of neighborhood The definition of neighborhood 1 is as follows: one cell on the same plane (on the hexagonal lattice plane, xy plane) (FIG. 2), upper and lower cells in the height direction (layer, Z axis), and these cells Is a cell next to (Fig. 2). The size of the neighborhood can be set for each simulation experiment. Was of the use of this model, thereby 60n ₁ time in the X-Y plane, 90n ₂ degrees Z-axis direction, in an obliquely upward (or downward) can be expressed 45n ₃ degrees, bond angles amino acid a combination thereof This is because I thought that I could express to some extent. _{However, n 1 = 1, ···,} 5, n 2 = 1, ···, 3, n 3 = 1, ···, 7, a. As a result, α-helix and β-sheet of secondary structure important in protein structure can be expressed.
4). Transition rules The transition rules that apply to any cell are defined below.
(1) A candidate for a direction in which cell information is moved is determined based on a vector sum received from amino acids in a range set as a neighborhood. The actually specified value is used as the charge Z of the amino acid. However, an amino acid having a charge Z of 0 is expressed by a fluctuation constant described later. The vector sum that any amino acid receives is given by equation (5)

The first term of Eq. (6)

It is obtained by assigning to. That is,

となる。
（２）周囲の状況を調べ、（１）で決定された移動候補先に実際に移動できるかどうかを判断する。例えば、移動候補先に他のアミノ酸がある場合はそこには移動する事ができない。
（３）ベクトル和の大きさＦとパラメータＴ（以下で、便宜上温度と呼ぶ）、パラメータＴｈ（以下で閾値と呼ぶ）が、以下の不等式を満たしているかどうかを調べる。

It becomes.
(2) The surrounding situation is examined, and it is determined whether or not the movement candidate destination determined in (1) can actually be moved. For example, if there is another amino acid at the transfer candidate destination, it cannot move there.
(3) It is examined whether the magnitude F of the vector sum, the parameter T (hereinafter referred to as temperature for convenience), and the parameter Th (hereinafter referred to as threshold) satisfy the following inequality.

ただし、閾値は０から１の間に正規化された一様乱数の値である。
（４）上記条件を全て満たしていれば、移動させる。

However, the threshold value is a uniform random number value normalized between 0 and 1.
(4) If all of the above conditions are met, move it.

  Equation (8) is based on the Metropolis standard used in the SA method. If the vector sum is large or the temperature is high in Equation (8), the left side becomes a large value, and the probability of satisfying Equation (8) increases.

  Here, to examine the properties of this function, consider the following two conditions separately. The first is when the temperature is constant. As the vector sum is larger, the information held in the cell moves, that is, the possibility of transition increases. In other words, the greater the force received from the surroundings, the easier the amino acid sequence is deformed. The second is when the vector sum is constant. The higher the temperature, the higher the possibility of transition. In other words, even if the force received from the surroundings is not so strong, the amino acid sequence is easily deformed at high temperatures. Here, the uniform random number is used as the threshold value, because the conditional expression is not satisfied even at high temperatures. In other words, even in an environment where the protein is easy to move, the conditional expression is satisfied even at low temperatures. That is, to create a state where the protein can move even in an environment where the protein is difficult to move.

  From the above, I think that the following environment has been established.

The higher the temperature or the greater the vector sum received from the surroundings, the more active the protein is deformed. Conversely, the lower the temperature or the smaller the vector sum, the slower the movement. However, even if the temperature is low, it is possible to move around if the vector sum or threshold value received from the surroundings is high. In other words, it is actively folded at first, but converges to a certain structure as the temperature decreases. These are set in consideration of actual protein folding.
Figure 3 shows the flow of the proposed method based on the above.

(1) A three-dimensional cell state space (FIG. 1) is generated.
(2) A linear amino acid sequence is arranged in an arbitrary state space (FIG. 4). In the figure, A represents an amino acid, and the subsequent numbers are serial numbers assigned to the amino acids.
(3) In order to add a modification to an arbitrary initial individual, a part of the amino acid sequence is selected as a modification candidate by a random number. This means that the funnel theory shows that "in the folding process, the transition state is not a single trajectory (path) but various trajectories". Because I thought it was good.
(4) The transformation candidate is allowed to undergo transformation (transition) only when the conditions (1) to (4) of the transition rule are satisfied. Then, (3) and (4) are performed a specified number of times for all initial individuals. This corresponds to the number of trials that appear later.
(5) The temperature T, which is an index of ease of movement, is changed after the transition is performed a specified number of times. This time, a change is made by applying a proportional constant of less than 1 to the previous temperature. This corresponds to a decreasing constant that appears later.
An image of the protein that can be expressed using the proposed method is shown in FIG. 0 represents a state in which no amino acid information is held, A represents an amino acid, and the subsequent number is a serial number assigned to the amino acid. In the figure, α-helix is represented by A1 to A15, and β sheet is represented by A16 to A33.

  From the above, using the author's proposed method, we can obtain results faster than existing methods that do not use structural information, and we can meet the needs indicated in the problem.

  However, since various approximations including space approximation are introduced, the accuracy may be inferior to that of existing methods. Next, we conducted a simulation experiment using this method to prove that the author's assertion was correct.

  “Experimental results and discussion” will be explained.

First, the experimental conditions are described, and then the experimental results for each condition are shown.
1. Setting of Experimental Conditions A simulation experiment using the proposed method was performed for the following four types of proteins. The following names are protein code names.
(1) 1 d x g A
(2) 2 er l
(3) 1 g p t
(4) 1 i x a
Tables 1 to 4 show experimental conditions for the proteins (1) to (4). The definitions of the words used in the table are as follows.
Code name: A code name uniquely given to a protein.
-Number of residues: The number of amino acids contained in the protein.
-Initial temperature: The value of the temperature parameter at the start of simulation.
-End temperature: Temperature parameter value at the end of the simulation.
• Decreasing constant: The next temperature at a certain time will multiply this temperature by the constant at that time. This is a proportionality constant less than 1.
-Number of trials: Number of trials at the same temperature. It is defined as the number of residues.
-Number of individuals: Number of state spaces to be generated.
• Neighborhood radius: the radius of the neighborhood that is taken into account to calculate the vector sum.
-Number of random transitions: Indicates the number of random state transitions when the temperature is sufficiently high. It is defined as the number of residues.
-Fluctuation constant: Expresses the charge fluctuation of amino acids with zero charge. Within this constant size range centered around 0, plus and minus charge fluctuations are given by random numbers. However, the charge of electrons is 1.
Lattice constant: The length of one side of the hexagonal lattice plane. The unit is Å.

２．結果の画像
各実験条件に対する結果を以下に示す。

（１）１ ｄ ｘ ｇ Ａに関する実験結果

最初に乱数依存性を調べるために乱数の種を変えて実験を行った。この実験結果を図６〜図８に示す。これは、Ｒａｓｍｏｌと呼ばれるタンパク質の立体構造を観察するために作成されたＶｉｅｗｅｒから得た画像である。実験結果を評価するために、このＲａｓｍｏｌを用いて各タンパク質の立体構造を観察した。

2. Result image The result for each experimental condition is shown below.

(1) Experimental results for 1 d x g A

First, in order to investigate the dependence on random numbers, we experimented by changing the seeds of random numbers. The experimental results are shown in FIGS. This is an image obtained from a viewer created to observe the three-dimensional structure of a protein called Rasmol. In order to evaluate the experimental results, the three-dimensional structure of each protein was observed using this Rasmol.

  First, in order to compare the predicted structure with a naturally occurring protein, numbers are assigned to FIG. In FIG. (B), parts corresponding to those in FIG. (A) are given the same numbers as in FIG. Also, numbers are similarly assigned to FIGS. In FIG. 6, the numbers are assigned from 1 to 5.

  First, the overall shape is a little smaller than (a) because (b) extends entirely behind the image. Next, in the individual parts, the parts from 1 to 4 are almost the same. In both figures, the curve extends from 1 to 2, and the direction changes in the opposite direction while drawing a counterclockwise curve. This is exactly the same for 2-4. However, in 4 to 5, (a) extends to the lower left after drawing a curve to the right, whereas (b) extends to the upper left after drawing a curve to the right. However, since both ends are not so important in the structure of the protein, this experimental result was considered good.

Researchers specializing in protein structure for this experimental result
“The overall structure is similar, but some parts are similar and some are not.”
Met. From this, it can be concluded that the experimental results shown in FIG. 6 are relatively good.

Next, FIG. 7 will be described. In the figure, numbers are assigned from 1 to 5.
First, in the overall form, (b) in the figure is not as small as (b) in FIG. Next, in the individual parts, the parts from 1 to 4 are almost the same. In both figures, the curve extends from 1 to 2, and the direction changes in the opposite direction while drawing a counterclockwise curve at 2. This is exactly the same for 2-4. However, from 4 to 5, (a) extends to the lower left after drawing a curve to the right, whereas (b) extends to the lower right after drawing a curve to the left. It is thought that there is not much problem because this is also different at both ends. Therefore, this experimental result was considered good.

Similar to FIG. 6, the evaluation of researchers specializing in protein structure for this experimental result was the same as FIG. From this, it can be concluded that the experimental results shown in FIG. 7 are good.

Next, FIG. 8 will be described. In the figure, numbers are assigned from 1 to 5.
First, in the overall form, (b) in the figure is not as small as (b) in FIG. Next, in each part, the parts 1 to 2 are missing. In 4 to 5, (a) extends to the lower left after drawing a curve to the right, whereas (b) curves to the left. After drawing, it extends to the lower left but is cut off. This is also different at both ends, but the lack of one or two parts is unacceptable. Therefore, this experimental result was considered somewhat good.

  Similar to FIG. 6, the evaluation of researchers specializing in protein structure for this experimental result was the same as FIG. However, since the parts 1 and 2 were missing, the evaluation was lower than the other results.

In (1), the experiment was performed by changing the seed of random numbers. As long as these three results are seen, all the results are good to some extent, and it is considered that the dependence of the present method on the random number is not so strong. However, this result is for one protein, and the same is not true for other proteins. Therefore, in (2), an experiment was performed by changing the random number for confirmation.

(2) Experimental results on 2 er l

In (2) following (1), an experiment was conducted by changing the seed of random numbers in order to investigate the random number dependency. The results are shown in FIGS. Experimental result 1 for 2 er l
Next, FIG. 9 will be described. The figure is numbered from 1 to 6.
First, the overall shape is almost the same ((b) in the figure, one end forms a spiral toward the back of the image, and then returns to the front while forming a spiral). Next, in each part, (b) in the figure is different from (a) in 5 to 6, but (b) in the figure forms an α helix in the same way as (a) in 2 to 3, The spiral cycle number is three. Furthermore, it can be seen that the spiral in the second cycle is slightly broken even at 3 to 4, but an α helix is formed. Therefore, this experimental result was considered good.

Similar to FIG. 6, the evaluation of researchers specializing in protein structure for this experimental result was the same as FIG. Furthermore, it was said that the characteristics were better than the experimental result of (1) in each part. From this, it can be concluded that the experimental results shown in FIG. 9 are good.
Next, FIG. 10 will be described. In the figure, numbers 1 to 6 are assigned.
First, in the overall form, (b) in the figure is different from 1 and 5, unlike (a). Next, in each part, from 2 to 3, (b) in the figure forms an α helix as in (a), and the number of cycles of the helix is three. Even in 4 to 5, the second cycle helix is slightly broken but forms an α helix ((b) forms a helix slightly from the lower right to the upper left). And the number of cycles of the helix is three. Therefore, this experimental result was considered good.

Similar to FIG. 6, the evaluation of researchers specializing in protein structure for the experimental results was the same as FIG. From this, it can be concluded that the experimental results shown in FIG. 10 are good.

  Next, FIG. 11 will be described. The figure is the best result among individuals, but it is clearly not good.

  Similar to FIG. 6, the evaluation of researchers specializing in protein structure for this experimental result was not similar.

In (2), the experiment was performed by changing the seed of random numbers as in (1). As far as seeing these three results, the two results were good, but the other one was not good. From the results of (1) and (2), the dependence on the random number of this method cannot be denied, but it can be considered that there is no problem in practical use.

(3) Experimental results on 1 g pt

In (3), an experiment was conducted on a protein having a more complex structure than (1) and (2). The experimental results are shown in FIG.

First, FIG. 12 will be described. In the figure, numbers are assigned from 1 to 9.
First, assuming that the overall shape is (b) in the figure and 1 can be moved to the right from 3 as a base point, and 9 can be moved to the right from 8 as a base point. ) Has almost the same shape as (a). Next, in each part, from 4 to 5, (b) in the figure forms an α helix as in (a), and the number of cycles of the helix is three. Further, the structures 2 to 3 and 6 to 7 have the same structure as (a). Therefore, this experimental result was considered good.

Similar to FIG. 6, the evaluation of researchers specializing in protein structure for this experimental result was the same as FIG. Furthermore, it is said that the features are good for the complicated structure. From this, it can be concluded that the experimental results shown in FIG. 12 are good.

(4) Experimental results for 1 i x a

In (4) following (3), an experiment was conducted on a protein having a more complicated structure than in (1) and (2). The result of this experiment is shown in FIG.

Next, FIG. 13 will be described. In the figure, numbers are assigned from 1 to 9.
First, as a whole, (b) in the figure has a structure almost the same as (a). Next, in each part, in the case of 2 to 6, (b) in the figure has the same structure as (a) if the shape of the dogleg shape between 3 and 4 is the same. Furthermore, the structure of the loop portion of 8 is the same. Therefore, this experimental result was considered good.

  Similar to FIG. 6, the evaluation from the researchers who specialize in protein structure for this experimental result was the same as FIG. From this, it can be concluded that the experimental results shown in FIG. 13 are good.

  The evaluation of researchers who specialize in protein structure for the experimental results in this section was "We believe that relatively good results were obtained throughout." From this it can be concluded that the experimental results presented in this section are generally good.

  Finally, when an experiment was conducted using this method at 2.26 GHz of Pentium (registered trademark) 4, the results could be obtained in about 2 hours on average.

The following three can be given as considerations regarding the experimental results.
(1) Many predicted structures differed from the naturally occurring protein at both ends. It is considered that an appropriate structure was not formed because other amino acids were less likely to be present around both ends than other parts and were not easily affected by the surroundings.
(2) From the results shown in FIG. 11, this method has a certain degree of dependence on random numbers.
(3) Since the overall experimental results were good, it is considered that the various approximations introduced were valid.

  Although there are points to be improved, the results of the experiment were generally good despite the introduction of various approximations. This suggests that this method may be effective as a method for analyzing the three-dimensional structure of proteins.

  If the author's proposed method is used, results can be obtained in a shorter time than the existing method that does not use structural information. In other words, this method suggests the possibility of being effective as a method for analyzing the three-dimensional structure of proteins.

The following three issues will be addressed.
(1) The reliability of this method is improved by examining the calculation process using energy calculation.
(2) Consider a framework that can correctly form both ends of a protein and reduce the dependence on random numbers.
(3) In order to further improve accuracy, information such as hydrophobic interaction is incorporated into this method.

Division diagram of analysis space. Neighborhood 1 range. Flow of the proposed method. The initial state of the amino acid sequence. Examples of proteins expressed by the proposed method. Experimental result 1 for 1 d x g A Experimental result 2 for 1 d x g A. Experimental result 3 for 1 d x g A Experimental result 1 for 2 er l. Experimental result 2 for 2 er l. Experimental result 3 for 2 er l 3. Experimental results for 1 g pt. Experimental results for 1 i x a.

Claims (19)

  This is performed on the structure of a substance composed of a plurality of connected elements using a computer, and the space where the target structure exists is divided, the state is embedded in the divided space, and the structure is made using mathematical formulas. A three-dimensional structure estimation method for estimating the whole while controlling the formation process.   The three-dimensional structure estimation method according to claim 1, wherein the element is an amino acid.   The three-dimensional structure estimation method according to claim 1, wherein the substance is a protein.   The three-dimensional structure estimation method according to claim 1, wherein the space is divided by a lattice.   The three-dimensional structure estimation method according to claim 4, wherein the lattice is a hexagonal lattice.   The three-dimensional structure estimation method according to claim 4, wherein the lattices are layered.   The three-dimensional structure estimation method according to claim 1, wherein information including presence / absence of an element, charge when an element is present, and positions of adjacent elements are stored as the state.   The three-dimensional structure estimation method according to claim 1, wherein the mathematical expression is a metropolis standard of annealing method.   The three-dimensional structure estimation method according to claim 1, wherein the mathematical expression is a function of a vector (F) and a control parameter (T).   The three-dimensional structure estimation method according to claim 9, wherein the vector (F) is a Coulomb force.   The three-dimensional structure estimation method according to claim 1, wherein the state is changed so that an element is moved when the mathematical formula exceeds a threshold value.   The three-dimensional structure estimation method according to claim 11, wherein the threshold value is calculated by a function.   The method according to claim 12, wherein the function generates a random number.   The three-dimensional structure estimation method according to claim 13, wherein the random number is a uniform random number. The formula is

The three-dimensional structure estimation method according to claim 9, wherein:   The three-dimensional structure estimation method according to claim 1, wherein the state is based on a local rule.   The three-dimensional structure estimation method according to claim 1, wherein the state depends on a nearby state.   11. The method for estimating a three-dimensional structure according to claim 10, wherein the Coulomb force is calculated in the nearest neighborhood or a range of a constant multiple thereof.   This is performed on the structure of a substance composed of a plurality of connected elements using a computer, and the space where the target structure exists is divided, the state is embedded in the divided space, and the structure is made using mathematical formulas. A three-dimensional structure estimation device that estimates the whole while controlling the formation process.

Images