JP2010134514A - Protein three-dimensional structure prediction method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method of determining which sequence fragment is firstly subjected to structure optimization by using the entropy of the sequence fragments of all the length and all the positions calculated from the sequence of protein whose structure is predicted when performing protein three-dimensional structure prediction. <P>SOLUTION: The structure optimization of protein having short sequence fragments is given priority over protein having long sequence fragments, and the optimization of the sequence fragments having the same length and having small entropy is sequentially given priority, and the energy optimization of the sequence fragments to be optimized afterwards including the sections of the fragments optimized beforehand is performed by scheduling. As for the optimization of each fragment, structure prediction is performed by optimizing the structure for minimizing total energy by calculating the sum of the mean power field potentials of all pairs of residual bases in the fragments and the sum of molecular dynamics potentials among all atoms in the fragments supposing that the all the residual bases are glycine, wherein CO and HH are added to the residual bases at both ends of each fragment. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a protein tertiary structure prediction method based on protein variable length fragment entropy analysis, which predicts a tertiary structure of a protein from amino acid residue sequences of natural and non-natural proteins.

  A natural protein is an organic polymer in which a peptide in which 20 kinds of biological amino acids are linked in a chain form from several tens to several thousands is folded into a specific structure. Since proteins are identified by their amino acid sequences, the types of proteins can be said to be infinite considering the diversity of length and the diversity of sequences. Proteins are present throughout the body and form the elemental structure of biological structures (structural proteins), act as catalysts for promoting chemical reactions in the body (enzymes), and are represented by the immune system It has various functions such as recognizing and binding specific molecules. The amino acid sequence of a natural protein in a living body is associated with the corresponding gene as a DNA base sequence, and when necessary, amino acids are combined based on the DNA base sequence to synthesize a peptide. The synthesized peptide finally takes a stable three-dimensional structure and has a specific function. Note that some proteins have a structure that is not partially or wholly constant and still has a specific function.

  Since the function of each protein is closely related to its three-dimensional structure, it is indispensable to know the three-dimensional structure of the protein in order to know the mechanism by which the protein functions. However, obtaining the protein's three-dimensional structure by X-ray crystallography or NMR method requires much effort and time compared to obtaining the protein's amino acid sequence. Is necessary, that is, a protein three-dimensional structure prediction method. In order to develop a highly accurate protein tertiary structure prediction method, it is necessary to explore what the relationship between the amino acid sequence and structure of the protein is, and this is to explore the mechanism of protein tertiary structure formation. Obtained by.

  Research on protein structure prediction methods began in the first half of the 1960s when the myoglobin structure was first elucidated by X-ray crystallography, and was initially a regular helical structure stabilized by hydrogen bonds in the structure. Attention has been focused on the helix and the sheet structure in which a plurality of linearly extending strand structures are joined together in a plate shape by hydrogen bonding, and the amino acid residue sequence indicating which part of the sequence takes the helix or strand structure There have been many studies on secondary structure prediction methods that try to infer from the sequence pattern. In addition, turn structures were found in some of the irregular structures stabilized by hydrogen bonds, although they were not regular, and research on the types of characteristic turn structures was also actively conducted. In the 1980s, statistical physics researchers used the theory of statistical physics to analyze the stability of protein steric structures, and to develop statistical physics folding mechanisms with simplified models of proteins. There has been a lot of research to elucidate. Under these circumstances, it has been found that the amino acid sequence of a protein having a stable three-dimensional structure is adjusted so that the structure is easily formed, and most of the random sequences do not have a stable three-dimensional structure.

  In the 1990s, the number of proteins whose three-dimensional structures were already known exceeded 1000, and protein structure prediction methods using the structures of these known proteins were proposed. In Non-Patent Document 1, the sequence of a protein for which a three-dimensional structure is to be predicted is applied to the three-dimensional structure of a protein with a known structure, based on the concept of mean force field potential statistically extracted from the protein with a known structure. From the average force field potential, when the fitness is high, it was found that there is a high possibility that the protein whose structure is to be predicted is similar to the structure of a known protein having a high fitness. Here, the protein structure prediction technique is interpreted as a structure recognition technique, and at this stage, the three-dimensional structure of a protein can often be predicted with practical accuracy. The mainstream of current protein structure prediction methods is a significant extension of this method, and roughly predicts the three-dimensional structure of proteins with unknown structure as possible structures obtained by the structure recognition method, which is then molecular dynamics. There are many things that are corrected in detail by calculation. At present, the rate at which the three-dimensional structure of a newly identified protein is similar to the predicted structure is said to be 60% or more, and it is considered that the prediction technology has entered a stage where it can be practically used.

  However, on the other hand, it is far from being able to say that the mechanism of protein structure formation has been found. If a newly elucidated protein has a three-dimensional structure that has little similarity to a known three-dimensional structure, and the protein sequence is completely different from that of a known protein, the structure prediction is successful. There are almost no examples, and there is only a prediction accuracy of 10% to 20% even if a case where a similar structure can be predicted to some extent can be included as a successful example. The true method for predicting the three-dimensional structure of a protein is not a method of inferring from a known protein structure with a structure recognition technique or a similar sequence, but a precise prediction method after elucidating the structure formation mechanism of the protein. It is.

  Under such circumstances, Patent Document 1 suggests that the order of formation of the three-dimensional structure of the protein may be determined by the entropy of partial fragments of the sequence. In the prior literature, for a small number of proteins for which the order of folding in structure formation has been clarified by experiments, the entropy for the structural diversity of partial sequence fragments of those proteins is calculated. Many cases were found that could be considered ongoing. These low entropy parts are not secondary structural parts such as regular helices and sheets that are finally stabilized by hydrogen bonds, but rather irregular structural parts corresponding to the breaks between secondary structure and secondary structure. Often corresponds to the part of the turn structure. This explains the reason why the secondary structure prediction method, which tried to estimate the helix and sheet parts, which was popular in the 1970s, was not very successful. The regular and stable structures such as these helixes and sheets frequently exist as a result of the natural folding of these stable structures regardless of the amino acid sequence pattern. Can be either a strand or a helix. Rather, the structure is considered to depend on the sequence of the part that creates an irregular structure between these stable structural parts on the sequence. It is. Since this discovery and this discovery are rooted in the principles of statistical mechanics, it has become possible to elucidate a very important part of the mechanism of protein structure formation.

  Therefore, focusing on the small entropy portion and examining the characteristics of the sequence pattern, there are many quite special ones among the 20 amino acids such as glycine, proline, tryptophan, and cysteine. Glycine can have a very flexible structure due to the absence of side chains, which characterizes this amino acid. On the other hand, proline has a 5-membered ring in the main chain, so that the angle φ of the main chain dihedral angle can only be around −60 °. Since tryptophan has a huge side chain, there are many restrictions on the angles of φ and ψ. On the other hand, when attention is focused on a portion with a large entropy, a very ordinary amino acid represented by alanine appears there. Alanine itself is often present in the helix, but when entropy is calculated, alanine is not willing to make a helix by itself, and even if it happens to be a helix, it is often not energetically unstable. It can be seen that the helix structure is often used and the helix structure is not triggered.

  What is estimated from the results of the entropy analysis as described above is as follows. A partial fragment with a large entropy in a protein sequence has a small local energy change in the partial sequence fragment, regardless of the structure of the partial sequence fragment. Depending on the influence of this part, it can have any structure. On the other hand, where the entropy is small, when that part takes a small number of specific structures, the local energy becomes very small and other structures are taken. The energy is so great that it tries to take a local structure with low energy by itself even if there is no influence from the surroundings. In other words, where the entropy is small compared to the surrounding area, it takes a specific structure prior to the formation of the structure of other parts, and as a result, the surrounding entropy area is affected and a relatively stable helix or sheet structure. I will take. This means that the protein structure formation process follows the Boltzmann principle of statistical mechanics.

The entropy S ^s of the sequence fragment is the relative arrangement k = j−i and the two types of amino acids a, i for the i-th amino acid residue in the fragment and the j-th amino acid residue in the fragment. The sequence-dependent pair entropy S ^sab _k determined from b ( ^assuming that the type of the i-th residue is a and the type of the j-th residue is b) is to determine all possible i ≦ j in the fragment. The total sum S ^s of the combinations to be satisfied can be calculated by the following equation.

[Equation 1]
S ^s = Σ _{i ≦ j} S ^sab _{k = j−i}

Here, the sequence fragment entropy S ^s means the sequence-dependent entropy S ^s of the total entropy S = S ^c + S ^s of the sequence fragment. Although entropy S ^c of sequence-independent in principle incalculable, since the sequence length is to take independent constant value sequence differences for the same sequence fragment, in order to see the entropy difference in due sequence, sequence As long as the lengths of the fragments are the same, it is only necessary to consider the sequence-dependent entropy S ^s of the sequence fragments.

As described above, an entropy map in which the sequence fragment entropy S ^s is obtained for fragments of various lengths and various positions in the entire protein sequence predicted in the three-dimensional structure prediction method of the protein is created. By determining the folding order, it is expected that the three-dimensional structure of the protein can be predicted including the folding process.

Japanese Patent Application Laid-Open No. 2008-146529. Sippl M.J., `` Calculation of Conformational Ensemble from Potentials of Mean Force: An Approach to the Knowledge-based Prediction of Local Structure in Globular Proteins '', Journal of Molecular Biology, 213, pp.850-883, 1990. Bowie J.U., et al, "A Method to Identify Protein Sequence That Fold into a Known Three-Dimensional Structure." Science, 256, pp.164-170, 1991. Onizuka K., et al., "Using Data Compression for Multi-dimensional Distribution Analysis", IEEE Intelligent Systems 17 (3), pp.48-54, 2002.

  In such a protein three-dimensional structure prediction method, it is necessary to be able to calculate the total energy value of a sequence fragment of any length when the fragment takes a specific three-dimensional structure fragment. Furthermore, it is required from the structure prediction algorithm that this total energy value depends only on the three-dimensional structure of the main chain and does not change depending on the side chain conformation. The first issue is how to calculate the total energy of structural fragments that satisfy this requirement.

  Next, in the energy optimization of structural fragments, the issue is how to obtain the optimal structure of the structural fragments, and this includes an initial structure setting method for gradually changing from the initial structure to the final optimal structure. And a method of deforming the structure fragment at each stage is a problem. That is, the initial structure setting method for the structure fragment is the second problem, and the third problem is a method for deforming the structure fragment.

  In order to predict the entire structure, it is necessary to perform local structure optimization in order from the smallest entropy after looking at the entropy of sequence fragments of various lengths and positions in the sequence of the protein to be predicted. . At this time, the entropy of structural fragments having different lengths due to constraints from the entropy calculation method cannot be compared because the entropy (incapable of calculation) is independent of sequence. Therefore, the magnitude of entropy must be compared only between sequence fragments of the same sequence length. The fourth problem is how to optimize the structure of the sequence fragments in any of these constraints.

  The object of the present invention is to solve the above problems and to calculate the total energy of protein structural fragments with higher accuracy than in the prior art, to appropriately set the initial structure, to deform the structural fragment, The object is to provide a protein tertiary structure prediction method capable of optimizing the structure of a fragment.

The protein three-dimensional structure prediction method according to the present invention calculates the sequence-dependent entropy S ^s of all possible partial sequence fragments of a protein sequence that predicts a three-dimensional structure, and folds from which partial fragment according to the magnitude of the value. Decide the folding order (structure optimization scheduling), and optimize the total energy value of the fragments by changing the variable-length fragments in each fold (structure fragment optimization) and finally the whole A protein tertiary structure prediction algorithm that predicts the structure is used. The protein tertiary structure prediction algorithm is intended to solve the protein tertiary structure prediction problem.

  The protein three-dimensional structure prediction method according to the present invention is a protein three-dimensional structure prediction method that predicts a three-dimensional structure of a protein, and calculates sequence-dependent entropy of all sequence fragments in the entire protein sequence to be predicted three-dimensional structure. The order of predicting or optimizing the structure of the structure fragment corresponding to each sequence fragment and the condition for considering the surrounding sequences are scheduled according to the magnitude relationship of the sequence-dependent entropy. And

  In the protein three-dimensional structure prediction method, the order of optimization of the structural fragments of the protein is optimized by giving priority to fragments having a short length, and in the optimization of structural fragments of the same length, sequences corresponding to the structural fragments If the structure fragments that are optimized in a given order have a portion that overlaps with a structure fragment of the same length that is to be optimized in the previous order, As a condition for optimization, the structure optimization is performed for the purpose of minimizing the total energy of the extended structural fragments including all of the structural fragments to be optimized first, which are continuous in sequence.

  In the protein tertiary structure prediction method, when the structure fragment optimization scheduling is applied in preference to a sequence fragment having a small length, the optimal solution for all residues is independent at length 1. Assuming that there is a length of 2 or more, the structure optimization at each fragment length is performed by a method in which the entire structure optimized at the stage where one length is small is used as an initial value. And

  Further, in the protein conformation prediction method, in the total energy optimization of the extended structural fragment of the protein, the sum of the average force field potentials for all pairs of residues in the extended structural fragment and the extended Sum of the sum of the Leonard Jones potential and the electrostatic potential between all atoms in the structure fragment when CO and NH groups are added to the residues at both ends of the structure fragment and all the residues are glycine And the sum of the dihedral potentials related to the main chain dihedral angle, and the energy difference caused by glycine as the total residue was extended with the non-glycine corrected by the mean force field potential of the glycine-glycine pair. The total energy of the structural fragment is used, and the structure of the extended structural fragment is optimized by modifying the structure to minimize this energy. It is characterized in.

  Furthermore, in the protein three-dimensional structure prediction method, in the optimization of the extended structural fragment of the protein, among the extended structural fragments, the main chain dihedral angles included in the structural fragment portion before being extended are continuous. The combination of the three dihedral angle angles ω, φ, ψ, or ψ, ω, φ is based on the size of the sequence fragment entropy of the partial structure fragment of sequence length 1 or 2 that involves the set of dihedral angles. The optimum dihedral angle ω, φ, ψ, or ψ, ω, φ is determined in order of increasing entropy.

  Therefore, according to the protein tertiary structure prediction method according to the present invention, the total energy of the protein structural fragment can be calculated with higher accuracy than in the prior art, and the initial structure is appropriately set and the structural fragment is deformed. The structure of the sequence fragment can be optimized.

  Hereinafter, embodiments according to the present invention will be described with reference to the drawings.

  The average force field potential proposed in Non-Patent Document 1 will be used to solve the first problem of calculating the total energy of sequence fragments. This mean force field potential is defined as a two-body potential obtained by statistically extracting the force field that acts between any two amino acid residues in the protein three-dimensional structure. b and the spatial arrangement in the case of relative arrangement k = j−i on the two sequences (assuming that the amino acid residue of a is i-th in the sequence and the amino acid residue of b is j-th) This is a statistical potential calculated from the distribution density of the relative arrangement r. This mean force field potential can be defined as a potential independent of the side chain conformation by ignoring the side chain conformation of the amino acid. However, due to a probabilistic problem, it is necessary to use the net mean force field potential proposed in Non-Patent Document 1, in which case the physicochemical interaction between the main chain atoms of amino acid residues is This has to be taken into account separately since it is not incorporated.

  Therefore, the force field potential of molecular dynamics is used as the interaction of the main chain atom of this amino acid residue. In other words, it is a Leonard Jones potential that approximates the classical electrostatic interaction and van der Waals interaction that is an intermolecular force, and a dihedral angle potential that acts on the main chain dihedral angle. The partial charge, van der Waals radius, and dihedral angle potential parameters required for each of these are the same as those used in a proven molecular dynamics calculation system (for example, the molecular dynamics calculation parameter set of CHARMM22) Such). A new issue at this stage is the handling of CB atoms in the side chain bonded to the CA atoms in the main chain. This CB atom strongly interacts not only with CA but also with other main chain atoms, and greatly affects the main chain dihedral angle φ. In order to solve this problem, when using the molecular dynamics potential, it was decided that all amino acids were glycine and there were no CB atoms. As a result, the interaction between the main chains is represented by the molecular dynamic interaction between glycine and glycine, which is not suitable as the main chain interaction of other amino acids.

  To avoid this, we decided to correct the mean force field potential with the mean force field potential of the glycine pair. In addition, since the target for calculating the total energy is in the middle of the algorithm when the sequence fragment has a specific structure, the situation at both ends of the fragment may have a bad influence. When a plurality of amino acids are combined to form a peptide, the C-terminal carboxyl group —COOH and the N-terminal amino group —NH 2 are dehydrated to form —CONH—. The bond connecting C and N is a double bond, and therefore the dihedral angle ω can only be around 0 ° and around 180 °, and is often 180 °. Then, CONH becomes a strong dipole because H has a positive partial charge and O has a negative partial charge. This dipole greatly affects the structure of the surrounding amino acids. In general, the boundary between two consecutive amino acids in the sequence is between CO and NH. However, in the structure fragment, CO and NH are added to the C-terminal side and N-terminal side, respectively, and the energy is such that CONH is at both ends. The calculation accuracy is better when the calculation is performed.

As described above, in the embodiment of the present invention,
(1) An average force field potential in Non-Patent Document 1 is introduced as a potential between amino acid residues,
(2) As the potential between main chains that are not expressed in this way, the molecular dynamics potential is introduced after all residues are glycine,
(3) To correct that all were glycine, the mean force field potential was corrected with the glycine pair,
(4) The total energy of the sequence fragment is calculated by adding CO and NH to the C-terminal side and N-terminal side of the sequence fragment, respectively.
The method of (1) to (4) solves how to calculate the total energy value when the sequence fragment, which is the first problem, has a specific structure.

  The second problem is naturally solved by means for solving the fourth problem.

  As a means for solving the third problem, it is assumed that a method of deforming the angles ω, φ, ψ of the main chain dihedral angles in the fragment is used. In the total energy calculation method devised by the means for solving the first problem, it is not desirable to change the distance or bond angle between the covalently bonded atoms in the main chain. It is desirable to change the structure by changing only the coupling dihedral angle. We think that it is impossible to change all dihedral angles in a fragment at the same time from the initial structure to the final optimum structure, so that it is impossible to optimize the structure. We will take a method of gradually approaching the optimum structure by changing the angle.

The problem is how to change the structure with a single structural deformation. It was confirmed that the method of changing only one of all the dihedral angles that can be changed in the fragment in one structural deformation does not lead to the final structure. The deformation method at this time does not lead to the final structure even if the dihedral angle can be changed by a maximum of 180 ° even when the dihedral angle is slightly deformed by about 1 ° to 5 °. Therefore, in one structural deformation, the angle ω, φ, ψ of one residue in the structural fragment (ω is considered to belong to the C-terminal residue among the residues on both sides) is changed greatly at the same time. We decided to use both the case and the case where the angles ψ, ω, and φ between two residues in the structural fragment are changed greatly at the same time. Also, which dihedral angle (ω, φ, ψ) of which residue or dihedral angle between two consecutive residues (ψ, ω, φ) is changed in one structural deformation is It was decided to use entropy S ^saa ₀ based on the internal structure of one closely linked residue and entropy S ^sab ₁ based on the relative arrangement of two consecutive residues (ψ, ω, φ). The continuous two-residue entropy S ^sab in the sequence fragment is represented by the following formula when the N-terminal residue is a and the C-terminal residue is b.

[Equation 2]
S ^sab = S ^saa ₀ + S ^sbb ₀ + S ^sab ₁

This entropy is calculated for all consecutive two residues in the sequence fragment, and is sequentially changed from the smallest two-residue sequence fragment. As a structural modification method for each consecutive two residue, S ^saa ₀ and S The angle ω, φ, ψ of the smaller residue of ^sbb ₀ is changed first, then the larger one is changed, and then the angle ψ, ω, φ between the two is changed. In order to obtain the optimal deformation by changing the continuous dihedral angle at the same time, the angle ω changes to 0 ° or 180 °, and φ and ψ change from 15 ° to 60 ° in all possible angle combinations. Each time, the total energy value of the structural fragment is calculated, and once set to an angle at which the smallest total energy value is obtained, correction by minute deformation is performed. If it is a large-scale deformation in 60 ° increments, the values of the angles ψ and φ that were optimal in the increment are changed by ± 32 ° respectively (there are 9 small deformations in total) and the optimum After obtaining a very small deformation, the change angle is set to 16 °, and this is repeated, and the deformation angle is halved and this is repeated every 1 °. When the total energy value of the structural fragment obtained by the deformation is smaller than that before the deformation, the deformation is accepted as the optimum deformation at that stage. If it is not smaller than before the deformation, the deformation is not performed, and the structure before the deformation is the optimum structure at this stage.

As described above, the third problem is solved in the following manner.
(1) The deformation from the initial structure to the optimum structure is realized by repeating the deformation of the structure fragment by changing the dihedral angle.
(2) In one structural deformation, three dihedral angles (ω, φ, ψ) of one residue in a fragment or three dihedral angles between two consecutive residues The angles (ψ, ω, φ) are changed simultaneously.
(3) Which dihedral angle of which residue or dihedral angle between two consecutive residues is changed is determined by the size of the sequence fragment entropy of the two consecutive residues, and is the smallest Change for two consecutive residues with entropy.
(4) For each successive two residues, first change the angle ω, φ, ψ with the smaller entropy of the residue, then change the larger one, then angle ψ, ω, φ between the two To change.
(5) As a method of simultaneously changing the dihedral angle in the vicinity, the energy value of the sequence fragment is calculated in all possible combinations of angles in increments of 15 ° to 60 °, and then changed to the best combination of angles. A combination of micro deformations of about half the angle is tried, then the step angles of the micro deformations are set to half each, and finally an optimal deformation angle combination in units of 1 ° is obtained.

Next, means for solving the fourth and second problems will be described. Entropy of sequence fragments can be compared for sequence fragments having the same sequence length (consisting of the same number of residues), since the sequence-independent entropy is constant. Can not. On the other hand, protein folding can be assumed to go from local structure formation to overall structure formation. Considering this, first of all, a sequence fragment is very short, for example, it consists of only one residue. It is considered that an algorithm for obtaining the entire structure while gradually lengthening the sequence fragment is suitable. Therefore, first consider the case where the length of the sequence fragment is 1, that is, one residue. The entropy to be considered in this case is S ^saa ₀ , (K = j−i = 0). Since the entropy of one residue is not affected by the type of surrounding residues, etc., the optimum structure can be determined independently for each residue of the protein sequence for which a three-dimensional structure is to be obtained. However, considering the influence of the above-mentioned dipole of CONH, the structure having the smallest energy (combination of angles ω, ψ, and φ) is selected by adding CO and NH to both sides of one residue. . This determines the initial value of protein tertiary structure prediction and can solve the second problem.

Next, consider the case of K = j−i = 1. The entropy case is ^{S s _i, i} _{+ 1} because it is entropy of contiguous two residues fragment pairs entropy as the elements sum ^{S s _i, i} _{+ 1} is expressed by the following equation.

[Equation 3]
_{^{S s i, i + 1 =}} S saa 0 + S saa 0 + S saa 1

The entropy S ^s _{i, i + 1} of all two consecutive residues (i-th residue and i + 1-th residue) in the sequence fragment is examined, and the optimum structure is determined in order of increasing entropy. At that time, when the entropy S ^s _{i, i + 1} is the local minimum, the optimum structure can be determined independently from the surroundings, but the entropy S ^s _{i−1, i} adjacent to the N-terminal side and the entropy adjacent to the C-terminal side are already present. When the optimum structure of S ^s _{i + 1, i + 2} is determined, the optimum structure of this part is affected by one or both of them. Therefore, in this case, when the optimum structure is obtained, an array solution including a neighboring portion in which the optimum structure has already been determined as a range for obtaining the total energy is calculated to obtain an optimum solution. In this way, in the case of K = 1, the optimum structure of fragments can be determined in order of increasing entropy, and the structure of the fragment having the smallest entropy is determined independently from the surroundings. The structure of the non-minimum fragment is determined in consideration of the influence of the surrounding predetermined structure. Furthermore, when the structure of the fragment having the maximum entropy S ^si _{, i + 1} is determined, the optimal structure is determined in consideration of the entire sequence having the entire sequence of the protein. At this stage, the provisional overall structure is determined. Desired.

  By repeating this method until K is the total sequence length -1, that is, the sequence fragment matches the entire sequence, it is shown that the entire structure is optimized. When K = 0, an optimum structure is obtained independently for each residue, and an optimum structure in the case of K = 1 is obtained with the result thus obtained as an initial state. Next, as for the structure in the case of K = 2, the optimum structure is obtained with the result of K = 1 as an initial state, and thereafter the entire structure is finally obtained by repeating until K becomes the total length of the sequence −1. This makes it possible to determine the structure in ascending order of entropy, taking into account the entropy of sequence fragments of all lengths, and obtaining the best method for determining the order of folding according to their size Thus, the fourth problem is solved.

  By this method, the order of optimization of the three-dimensional structure can be strictly determined by the entropy of all the lengths and positions of the entire sequence of the protein to be predicted, which is a problem to be solved by the present invention. As a result, a great progress can be made toward the realization of highly accurate protein tertiary structure prediction.

Embodiments according to the present invention
(1) Using the average force field potential between amino acid residues based on statistics extracted from proteins of known structure,
(2) Calculate the sequence-dependent entropy of sequence fragments of all possible lengths and positions of the amino acid residue sequence of a given protein that performs three-dimensional structure prediction,
(3) Determine the order of sequence fragments to be optimized based on the entropy of each sequence fragment and the dependency on other sequence fragments,
(4) When predicting the structure of each sequence fragment, the average force field potential for all pairs of residues in the range and the range of the range for which the structure determination of the sequence fragment depends on the structure prediction of the sequence fragment The total energy calculated as the sum of all the potential electrostatic potentials and Leonard Jones potentials between all the residues, assuming that all residues are glycine, and the correction term for the mean force field potential for residues other than glycine. This is a method for predicting the final three-dimensional structure by optimizing the fragments to minimize the size.

The embodiment according to the present invention is based on the above configuration.
(1) Mean force field potential calculation unit 1;
(2) a sequence entropy map calculation unit 2 for calculating sequence-dependent entropy of all sequence fragments of the entire given protein sequence;
(3) A structure optimization scheduling unit 3 that determines the order of structure optimization of sequence fragments and the dependency on other partial sequences at the time of each structure fragment optimization using the calculated entropy result of each sequence fragment When,
(4) When optimizing each sequence fragment, the fragment structure optimization unit 4 deforms the partial structure fragment, calculates the total energy of the related portion, and minimizes it.
It is roughly divided into four parts.

  First, since the average force field potential calculation unit 1 uses a known method as it is, it does not include the gist of the present invention. However, in an embodiment of the present invention, in order to calculate a highly accurate pair entropy, the relative configuration r in the mean force field potential is the relative distance d in the residue pair space and the direction of the other residue from one side. What is defined by three degrees of freedom (two degrees of freedom of zenith angle θ and azimuth angle φ) is unacceptable. That is, a multidimensional mean force field potential is used. As a statistical process, it is necessary to create a histogram of r for the residue pairs in the three-dimensional structure database that are the types of amino acids a and b constituting the residue pair and the relative arrangement k on the sequence. As a method for dividing r, if a histogram is created in increments of 100 pm (picometers) for distances and increments of 30 ° to 60 ° for angles θ and φ, a good balance is obtained in consideration of the number of data and statistical accuracy. .

  Further, the distance k on the sequence of two amino acid residues constituting the residue pair of the average force field potential must be defined when k = 0, that is, the average force field potential of one amino acid instead of the pair. There is. In this case, since the relative configuration within one amino acid cannot be defined, a set of main chain dihedral angles ω, φ, ψ represents the internal state as the internal state of one amino acid, and the angle ω, φ , Ψ is used to determine the mean force field potential. Similarly, when k = 1, it is the average force field potential of two adjacent amino acids, and the relative arrangement of the two adjacent amino acids is almost strictly determined by the angle ψ, ω, φ of the main chain dihedral angle between them. Therefore, the average force field potential is defined by using the angles ψ, ω, and φ as the relative arrangement. When creating a histogram in the statistical processing of these dihedral angles, the angle ω takes only 0 ° or 180 ° depending on the characteristics of the chemical bond represented by this, and the angles φ and ψ are 15 ° to 60 °, respectively. Divide by ticks to create a three-dimensional histogram.

When k> 1, a three-dimensional average force field potential is used. If a protein having a very long sequence is taken into consideration, k can be made as large as possible. However, for k above a certain value, the distribution f ^ab _k (r) is not significantly different, and k is a constant value K In the case of larger k> K, it may be statistically indistinguishable. K is preferably about 5 to 20 in consideration of statistical errors.

  Since the average force field potential is determined based on statistics as described above, the frequency is observed by dividing a continuous space. As a result, the calculated mean force field potential is discontinuous at the divided boundary. This potential discontinuity should be avoided when using the mean force field potential in the protein tertiary structure prediction method in which the optimal structure is obtained by repeating microdeformation. In the embodiment of the present invention, a secondary Bezier is used. The mean force field potential smoothed using interpolation was used.

  The method for calculating the sequence fragment entropy is not the gist of the present invention, but is related content, so the outline thereof will be described. The sequence fragment entropy calculation system used in the present invention includes a pair entropy calculation unit 8, a sequence fragment entropy calculation unit 10, and an array for calculating sequence fragment entropy for all the length positions of the whole sequence of a given protein. The entropy map calculation unit 2 includes three parts.

The entropy calculation unit 8 first calculates the value of the mean force field potential when a given amino acid pair a, b is separated by k on the sequence for various r, from which the amino acid pair a, b is arranged There is a step S1 of calculating the entropy S ^sab _k (r) when the distance is k above, and then a step S2 of performing this for all a, b and k combinations. Step S2 is a triple loop process in which step S1 is performed for combinations of different amino acid types a, b, and k. Further, the entropy calculation unit 8 includes a step S3 for storing the calculation result. That is, the calculated S ^sab _k (r) is stored as the entropy table 9 in step S3. When the present invention is performed on a normal computer system, the anti-entropy table 9 must be stored in a nonvolatile storage medium. Once this stored anti-entropy table 9 has been calculated, it need not be recalculated until it needs to be updated. The necessity of updating refers to the case where the statistical data that is the basis of the average force field potential is updated or the calculation method of the average force field potential is changed.

The sequence fragment entropy calculation unit 10 is a part that calculates the sequence-dependent entropy of a given sequence fragment. First, the sequence fragment entropy calculation unit 10 has a step S4 of reading the pair entropy table 9 calculated and stored by the pair entropy calculation unit 8 prior to the calculation. . Next, the sequence fragment entropy calculation unit 10 checks the types a and b of the i-th and j-th residues that satisfy i ≦ j in the input sequence fragment, and a, b, k = having step S5 referring to S ^sab _k corresponding to j−i and calculating the final sequence fragment entropy of entropy S ^sab _k for all i, j satisfying i ≦ j in the sequence fragment It has step S6 which calculates total. The sequence fragment entropy obtained in step S6 is calculated.

The sequence entropy map calculation unit 2 calculates all possible sequence fragments in the entire protein sequence (the number of residues or the sequence length is N), ie, i th to j th The sequence fragment entropy is calculated for all the sequence fragments satisfying 1 ≦ i ≦ N and i ≦ j ≦ N, and this is output as the map S ^s _ij .

The structure optimization scheduling unit 3 uses the sequence fragment entropy map obtained by the sequence entropy map calculation unit 2 to optimize which sequence fragment in any order based on the dependency with what other part. To decide. Let S ^s _{i, j} be the sequence fragment entropy of the sequence fragment from the i-th amino acid residue to the j-th amino acid residue in the entire residue sequence 5 of the protein whose structure is to be predicted. This entropy refers to the sequence-dependent entropy of a sequence fragment of length k + 1 (where k = j−i), which means that all i ′, j satisfying i ≦ j and j−i ≦ k. ^Is the sum of S ^sab _{k = j′−i ′} . Here, a is the type of the i'th residue, and b is the type of the j'th residue.

First, the simplest k = 0, that is, optimization of a structural fragment consisting of one residue of length 1 is performed first. In this case, it is considered that there is no influence of other residues in the optimization of each residue, and the optimal structure of all residues can be obtained independently. The optimum structure is tentatively determined so that the minimum total energy value is obtained by adding fragments CO and NH of adjacent residues to both ends of the fragment. That is, the structure optimization scheduling unit 3 has a step S1 for scheduling so that each residue in the entire sequence is optimized independently when k = 0. Next, when k = K (K> 0), that is, the structure optimization order of the sequence fragment having a length of K + 1 and the influence from a large number of sequence parts are determined as follows. First, among all the sequence fragments having a length of K + 1, for the smallest entropy S ^s _{i, j + K} , this means that the structure of this fragment part can be determined independently of the other parts, and the i th residue To optimize the structure fragment consisting of residues i + K to minimize the total energy. When calculating the total energy of the fragment of this part, CO is added to the N-terminal side when the i-th residue is not N-terminal, and NH is C-terminal side when it is not the i + K-th residue C-terminal. Find the total energy when added.

Next, when the entropy S ^s _{i, j + K} of the fragment from the i-th residue to the i + K-th residue is larger than the entropy of the same-length fragment surrounding this fragment and the overlapping portion, As a target for calculating the total energy, the range is extended to the N-terminal side and the C-terminal side as long as the sequence is continuous.

FIG. 2 is a diagram showing a range of structures (arrays) to be considered in calculating the total energy when optimizing the sequence fragment structure in the calculation system of FIG. The horizontal circles in FIG. 2 represent residues, the range indicated by 15 is a sequence fragment to be optimized in the current order, and the fragments indicated by 11 to 14 are fragments to be optimized in the preceding order. is there. In this case, among the regions (fragments) indicated by 11 to 14 that are optimized in the preceding order, except for 12 that does not overlap with the newly optimized 5 continuously, the sequence fragments of 11, 13, 14, and 15 are obtained. A total of 16 extended fragment portions are fragments to be subjected to the total energy calculation. The range of the sequence fragment thus extended is assumed to be the range from the I-th residue to the J-th residue. Optimization of the sequence fragment with the largest entropy S ^s _{i, j + K} is performed at the end when the sequence length is K + 1, and if the total length of this protein is N, then I = 1 (N-terminal) and J = N (J is C-terminal).

  When calculating the total energy, if the I-th is not the N-terminus, CO is added to the NH of the I-th residue, and if the J-th residue is not the C-terminus, NH is added to the CO. Perform under conditions. That is, the structure optimization scheduling unit 3 shown in FIG. 1 performs optimization on each sequence fragment based on the calculation result of the sequence fragment entropy when k = K (K> 0), and optimizes each sequence fragment. In step S12, it is determined how to set the ranges I and J of the array for calculating the total energy. The optimization order is determined by increasing K by 1 from 0 to N-1. This structure optimization scheduling unit 3 completely determines from which part the structure is optimized and the energy up to which sequence part is taken into consideration when the entire protein sequence whose structure is to be predicted is given. The Therefore, the structure optimization scheduling unit 3 first performs step S11 of the structure optimization scheduling unit 3 and then optimizes the fragments in each K from K = 1 to K = N−1 (total length−1). And step S13 for determining the total energy calculation ranges I and J, and how to optimize in this order for all sequence fragments in the entire sequence of the protein whose structure is to be predicted. decide.

  Next, the structural fragment optimization unit 4 has the sequence fragment to be optimized in the i-th to j-th range (i ≦ j), and the maximum total energy calculation range is the I (1 ≦ I ≦ i) -th range. When the residue is the Jth (j ≦ J ≦ N) th residue, the main chain dihedral angles ω, φ, ψ related to the ith to jth residues are changed to change the Ith This is the part that minimizes the total energy of the J-th residue from the residue and the structural fragment with CO and NH added to both ends respectively. When the angles ω, φ, ψ of one residue are changed, these three angles are changed at the same time, and the angles ψ, ω, φ between two consecutive residues in the sequence are also changed. A method is adopted in which the three angles ψ, ω, and φ are changed simultaneously. When changing from the i-th residue to the j-th residue, all the j-i + 1 residues in the meantime, the set of angles ω, φ, ψ, and all the j-i residues in this range The pair of dihedral angles ψ, ω, and φ is changed, and the order of change must be determined. Therefore, also from this, it is determined which set of angles is to be changed in order based on the entropy relationship of the partial fragments composed of two residues that are continuous with each residue. That is, the fragment structure optimizing unit 4 has a step S21 of calculating the entropy of the partial sequence fragment of length 2 (that is, k = 1) among the i th to j th sequence fragments, and then the smallest entropy Step S22 in which optimization is sequentially advanced from the partial sequence fragments. In step S22, the angles ω, φ, ψ of the two residues constituting the partial sequence fragment of length 2 and the angles ψ, ω, φ between the two residues are one and three in total. There are two variable angle pairs (however, the angles overlap). Therefore, in accordance with the principle that the angles should be optimized sequentially from the local area as much as possible, the order of optimization of the angles of the two residues constituting the partial fragment of k = 1 is first, the entropy of the residues out of the two is small The larger angle ω, φ, ψ is optimized, then the larger angle ω, φ, ψ is optimized, and then the angle ψ, ω, φ between the two residues is optimized. In this method, since it is considered that the final optimum structure is not reached after step S22 is performed once, step S22 is repeated until the convergence stage where the structural deformation is no longer possible. Since entropy does not change with structural changes, step S21 need only be performed once.

  The structural deformation method in step S22 determines which angle is optimal for the angle ω in either 0 ° or 180 °, and the angles φ and ψ in steps of 15 ° to 60 °, respectively. Judging from the total energy in the case of giving an angle, it has a step S23 for transforming it into the optimum angles ω, φ, ψ, and when the step size of the angles φ, ψ is 60 °, first, the angle φ, When each of ψ is changed by 0 ° and ± 32 ° (there are 3 ways for each of the two angles, there are 9 ways), step S24 is found to find a set of fine movements of the optimum angle, and then the deformation width is set to 16 Step S24 is performed, and step S25 is continued until the deformation width is halved until it reaches 1 °. When the step angle of the angles φ and ψ in step S23 is 15 °, the initial deformation width in step S24 is 8 °. That is, when the step angle is L °, the number is obtained by a power of 2 that does not exceed the step width L. As the step width L °, either 60 °, 30 °, or 15 ° is appropriate in consideration of the stability of chemical bonds. In the case of 30 °, the deformation width is 16 °. Thus, by repeating step S23 and step S24, the optimum angle ω, φ, ψ, or the pair of angles ψ, ω, φ between two residues is determined, and in this case, the I-th to J-th structures are determined. If the total energy value of the fragments becomes smaller than when the structure does not change at all, the deformation is effective. If not, the deformation is discarded, and the optimal deformation is obtained by step S25 to return to the original set of angles. The angle combination is obtained sequentially. In step S23, when obtaining the best combination of angles ω, φ, ψ or angles ψ, ω, φ with step size L °, the statistical processing of proteins with known structures is used to determine the angle that cannot be taken due to steric hindrance. By excluding combinations, it is possible to reduce the number of combinations and increase the calculation speed.

  As described above, according to the embodiment of the present invention, the structure optimization scheduling unit 3 that determines the order of structure optimization based on the results of the pair entropy calculation unit 8 and the sequence fragment entropy calculation unit 10, and the structure optimization A fragment structure optimizing unit 4 for optimizing each structural fragment in the order according to the result of the synthesizing / scaling unit 3. The structure optimization scheduling unit 3 determines which structural fragment is to be optimized in what order and under what conditions in steps S11 to S13, and the fragment structure optimization unit 4 Is the schedule and conditions determined by the third part of the structure optimization scheduling (range on the array for calculating the total energy) Along, optimized by step S25 the structure of each sequence fragments from step S21. When the sequence fragment length is finally equal to the total sequence length, the sequence fragment structure is optimized. The overall structure (stereostructure) obtained is the overall structure of the given protein (stereostructure). ), The structure prediction is completed.

  In FIG. 1, a counter entropy table 9 and mean force field potential calculation statistical data 7 are stored in a storage device such as a hard disk memory, for example, and a counter entropy calculation unit 8, a sequence fragment entropy calculation unit 10, a sequence entropy map calculation unit 2, The structure optimization scheduling unit 3, the structure fragment optimization unit 4, and the average force field potential calculation unit 1 are configured by a digital computer such as a computer.

  FIG. 3 shows the result of prediction using the protein tertiary structure prediction method by the calculation system according to the embodiment of the present invention. 3A is a diagram showing the partial structure 21 predicted by the calculation system of FIG. 1, and FIG. 3B is a diagram showing the actual structure 22. As shown in FIG.

  In FIG. 3A, 21 is the predicted three-dimensional structure of POIA1, and 22 in FIG. 3B is the corresponding three-dimensional structure of POIA1. Although it takes time to predict the whole structure and it is difficult to predict with high accuracy, it shows that it is possible to predict with high accuracy when a part of the whole protein is cut out. In this example, the turn structure by proline and glycine takes almost the same structure as the actual protein (the relevant part of POIA1), which induces the formation of a hairpin sheet structure. The correspondence of hydrogen-bonding residues in sheet formation is also consistent with the actual protein conformation. In this method, the structure shown in 1 was obtained at the stage of K = 8 (sequence fragment length 9).

  As described above in detail, according to the present invention, the accuracy of the three-dimensional structure prediction of a protein is improved and the order of the three-dimensional structure formation can be predicted. It is possible to estimate what design or sequence change has the desired function and the structure that realizes that function in the design of a non-natural artificial protein possessed or the sequence change of an existing natural protein. . This is useful for elucidating detailed mechanisms in biological activities, and for developing methods for controlling physicochemical reactions inside the body. Furthermore, in designing a new artificial protein, a lot of knowledge will be given at the stage of determining the details of the design.

  The mechanism of secondary structure formation of proteins has already been elucidated by this invention. The helix structure has a local structure in which φ and ψ are negative in the small entropy sequence fragment of 2 to 4 residues in length, so that the electrostatic potential generated by each dipole of each CONH It was found that a helix-like structure was created by making the region downstream of this portion mainly downstream of the C-terminal easy to become a helix, and a long helix was formed by extending this further to the C-terminal. It was also found that the region with small entropy residues, such as glycine and proline, produced hairpin turns mainly in the sheet structure, which induced hydrogen bonding in the regions on both sides, forming an antiparallel sheet structure. At this time, there is not a special combination of residues between residue pairs in regions separated from each other in the sheet structure in which hydrogen bonds are formed, and even when both are difficult to form a sheet, It was also found that it was embedded in the sheet structure depending on the surrounding environment.

  In addition, there are times when the entropy of the region that makes up the strands that ultimately make up the sheet is small, in which case it insists that the strands make strands tough and how helix the surroundings are, It was also found that other strands approached in the vicinity to form a sheet by not becoming. Furthermore, for the helix, the helix structure extends to the C-terminal side or the N-terminal side, but the regions with small entropy located at both ends of the helix finally formed an irregular turn structure or loop structure. It has also been found that making it stops the extension of the helix.

  The knowledge of structure formation obtained by the structure prediction according to the present invention is greatly different from the knowledge that has been attempted to be used in the conventional protein structure prediction method. By using these new knowledge, protein structure formation It is considered that the understanding of this mechanism is further advanced, and these findings are indispensable for designing artificial proteins having novel sequences and modifying existing natural proteins.

It is a block diagram which shows the structure of the calculation system which calculates entropy using the protein tertiary structure prediction method which concerns on embodiment of this invention. In the calculation system of FIG. 1, it is a figure which shows the range of the structure (sequence) which should be considered at the time of total energy calculation at the time of sequence fragment structure optimization. (A) is a figure which shows the partial structure 21 estimated by the calculation system of FIG. 1, (b) is a figure which shows the actual structure 22. FIG.

Explanation of symbols

1 ... Average force field potential calculation part,
2 ... Sequence entropy map calculator,
3. Structure optimization scheduling unit,
4 ... Structural fragment optimization unit,
5 ... Protein residue sequence,
6 ... Protein tertiary structure prediction result,
7 ... Statistical data for calculating mean force field potential,
8: Counter entropy table calculation unit,
9 ... Entropy table,
10: Sequence fragment entropy calculation section.

Claims (5)

  In the protein tertiary structure prediction method that predicts the three-dimensional structure of a protein, the sequence-dependent entropy of all the sequence fragments in the entire sequence of the protein for which the three-dimensional structure is to be predicted is calculated. A protein three-dimensional structure prediction method characterized by scheduling an order of predicting or optimizing a structure of a structure fragment corresponding to a sequence fragment and a condition on how to consider surrounding sequences.   The order of optimization of the structural fragments of the protein is optimized in preference to fragments having a short length. In the optimization of structural fragments of the same length, the sequence-dependent entropy of the sequence fragments corresponding to the structural fragments is in ascending order. If the structure fragment that is optimized in the predetermined order overlaps with the structure fragment of the same length to be optimized in the previous order, the structure fragment optimization condition is The protein three-dimensional structure prediction according to claim 1, wherein the structure optimization is performed for the purpose of minimizing the total energy of the extended structural fragment including all of the structural fragments to be optimized. Law.   In the structure fragment optimization scheduling, when applying in preference to a sequence fragment having a small length, it is assumed that the optimal solution of all residues is independent at length 1, and then the length is 3. In the case of 2 or more, the structure of each fragment length is optimized by a method in which the whole structure optimized in the stage where one length is small is used as an initial value. Prediction method.   In the total energy optimization of the extended structural fragment of the protein, the sum of the mean force field potentials for all pairs of residues in the extended structural fragment and the CO at both ends of the extended structural fragment , And NH, and the sum of the Leonard-Jones potential and electrostatic potential between all atoms in the structural fragment, assuming that the residue is all glycine, and the main chain dihedral angle. The sum of the angular potentials, and the energy difference caused by setting all residues to glycine, corrected for the non-glycine by the average force field potential of the pair of glycine and glycine, is the total energy of the extended structural fragment, and this is the minimum 3. The protein solid according to claim 2, wherein the structure of the extended structural fragment is optimized by modifying the structure so as to be Concrete prediction methods.   In the optimization of the extended structural fragment of the protein, three continuous dihedral angles ω and φ among the main chain dihedral angles included in the structural fragment portion before being extended in the extended structural fragment. , Ψ, or a combination of angles ψ, ω, φ, with the optimal dihedral angle in ascending order of entropy based on the size of the sequence fragment entropy of the partial structure fragment of sequence length 1 or 2 related to the set of dihedral angles. 5. The protein tertiary structure prediction method according to claim 4, wherein the angles ω, φ, ψ or the angles ψ, ω, φ are determined.