JP7583450B2 - Method, device and program for predicting chemical shift values of nuclear magnetic resonance - Google Patents

Description

The present invention relates to a method, a prediction device, and a prediction program for predicting chemical shift values in nuclear magnetic resonance analysis.

Currently, in the field of drug discovery, attempts are being made to develop new drugs that target intrinsically disordered proteins (also known as IDPs) and proteins with intrinsically disordered regions (IDRs).

Nuclear Magnetic Resonance (NMR) is a useful tool for IDPs (Non-Patent Document 1). In NMR analysis, chemical shift (δ/ppm) is the most basic and important observation quantity (parameter), and is used to predict the three-dimensional structure of peptides represented by the three-dimensional coordinates of atoms (Non-Patent Document 2). Attempts have also been made to identify the structure of IDPs by predicting the chemical shift in NMR analysis. In the structural analysis of IDPs, it is necessary to sample as many predicted structures of the target protein as possible and evaluate the experimental reproducibility (Non-Patent Document 3).

M. R. Jensen, et al.:Chem. Rev., 2014, 114, 6632-6660 Y. Shen and A. Bax:J. Biomol. NMR, 2007, 38, 289-302 M. Schwalbe, et al.: Structure, 2014, 22, 238-249

In the method described in Non-Patent Document 3, when performing structural analysis of Tau protein and α-synuclein, which are types of IDPs, the structures of individual peptides contained in the target peptide population must be analyzed by conducting numerous experiments using a combination of small angle X-ray scattering (SAXS) analysis, NMR analysis, etc.

In addition, the method described in Non-Patent Document 2 is useful in that it can predict the chemical shift from the three-dimensional structure of the target peptide. For example, the chemical shift can be predicted from the dihedral angle values (Φ and Ψ values) of the main chains of the amino acids constituting the peptide and the dihedral angle value (χ ¹ value) of the side chains of the peptide, and compared with the experimental value. However, this method requires a huge amount of time for analysis when the peptide structure is diverse, such as IDPs and IDRs.

The objective of the present invention is to predict the three-dimensional structure of a target peptide in a shorter period of time.

As a result of extensive research, the present inventors have found that by performing cluster analysis on the dihedral angle application frequency of amino acid residues and using the obtained cluster distribution, it is possible to predict the chemical shift value of a target peptide in a shorter time.
The present invention has been completed based on this finding, and includes the following aspects.
Item 1.
A method for predicting a nuclear magnetic resonance chemical shift value of a peptide, comprising the steps of:
A step of clustering the occurrence frequency distribution of the dihedral angle values (represented by values of Φ and Ψ) of the main chain of each amino acid residue constituting the target peptide to obtain a distribution of multiple clusters;
calculating a similarity between the distribution of each of the obtained clusters and a reference main chain dihedral angle value registered in an amino acid three-residue database using a score function;
and predicting the nuclear magnetic resonance chemical shift value of the target peptide from the calculated similarity.
Item 2.
Item 2. The prediction method according to item 1, further comprising performing pre-clustering based on the dihedral angle values of the side chains (represented by the value of χ ¹ ) before clustering the frequency distribution of the dihedral angle values of the main chains of each amino acid residue.
Item 3.
Item 3. The method for predicting a nuclear magnetic resonance chemical shift value of an unchained protein according to Item 1 or 2.
Item 4.
4. The prediction method according to any one of claims 1 to 3, wherein the clustering is performed using a Gaussian mixture distribution.
Item 5.
5. The prediction method according to item 4, wherein the similarity is calculated by an integral method.
Item 6.
On the computer,
A step of clustering the occurrence frequency distribution of the main chain dihedral angle values (expressed as values of Φ and Ψ) of each amino acid residue constituting the target peptide to obtain a distribution of a plurality of clusters; A step of calculating the similarity between the distribution of each obtained cluster and a reference main chain dihedral angle value registered in an amino acid three-residue database using a score function;
predicting a nuclear magnetic resonance chemical shift value of the target peptide from the calculated similarity;
A prediction program for the nuclear magnetic resonance chemical shift value of the peptide.
Item 7.
A prediction device for predicting a nuclear magnetic resonance chemical shift value of a peptide, comprising:
The prediction device (10, 20) includes a processing unit (101, 201),
The processing unit (101, 201)
The frequency distribution of the dihedral angle values (represented by values of Φ and Ψ) of the main chain of each amino acid residue constituting the target peptide is clustered to obtain a distribution of multiple clusters;
Using a score function, the similarity between the distribution of each obtained cluster and the reference main chain dihedral angle value registered in the amino acid three residue database is calculated;
predicting a nuclear magnetic resonance chemical shift value of the target peptide from the calculated similarity;
The prediction device (10, 20).
Item 8.
A method for predicting a nuclear magnetic resonance chemical shift value of a peptide, comprising the steps of:
A step of clustering the occurrence frequency distribution of the dihedral angle values of the main chains (represented by values of Φ and Ψ) and the dihedral angle values of the side chains (represented by values of ^χ1 ) of each amino acid residue constituting the target peptide, and obtaining a three-dimensional distribution of multiple clusters;
calculating a similarity between the distribution of each of the obtained clusters and a reference main chain dihedral angle value registered in an amino acid three-residue database using a score function;
and predicting the nuclear magnetic resonance chemical shift value of the target peptide from the calculated similarity.

According to the present invention, the NMR chemical shift values of a target peptide can be predicted in a shorter time.

The present invention is outlined below. Fig. 1(a) shows a distribution map of the frequency of occurrence of dihedral angles of a target peptide. Fig. 1(b) shows a part of a molecular model of the target peptide. Fig. 1(c) shows a cluster distribution obtained by clustering the distribution map of the frequency of occurrence of dihedral angles with a mixed Gaussian distribution. Fig. 1(d) shows a combination of clusters of the frequency of occurrence of dihedral angles of the i-1th, ith, and i+1th amino acid residues of the mixed Gaussian distribution. 1 shows the hardware configuration of the prediction systems 1000 and 2000. 2 shows a hardware configuration of the prediction device 10. 2 shows functional blocks of the prediction device 10. 13 shows the process flow of a prediction program 1042. 5 and step S26 shown in FIG. 9. FIG. 5 shows a flowchart of the chemical shift value prediction process. 2 shows a hardware configuration of the prediction device 10. 2 shows functional blocks of the prediction device 10. 13 shows the process flow of a prediction program 1042. The figure shows a comparison of the nuclear magnetic resonance chemical shift values of unfolded apomyoglobin predicted by the method of the present invention and the nuclear magnetic resonance chemical shift values predicted by the SPARTA program. In the figure, C, CA, CB, N, NH, and HA represent ¹³ C', ¹³ C ^α , ¹³ C ^β , ¹⁵ N, ¹ H ^N , and ¹ H ^α , respectively. Three clustering results obtained by pre-clustering with the ^χ1 value of glutamine residues are shown. The left column shows "gauche+", the middle column shows "trans", and the right column shows "gauche-". The top row shows the clustering results, and the bottom row shows the predicted main chain dihedral angle data. 1 shows the reproducibility of chemical shift experimental data of urea-denatured apomyoglobin. In the figure, C, CA, CB, and HA represent ¹³ C', ¹³ C ^α , ¹³ C ^β , and ¹ H ^α, respectively.

1. Overview of the present invention and explanation of terms In this specification, the term "target peptide" refers to a peptide for which a chemical shift value is to be predicted. Hereinafter, the "target peptide" may be simply referred to as the "target peptide".

The "residue" of each amino acid is a structural unit of an amino acid that constitutes a peptide, and refers to a group formed by removing a hydrogen atom from the amino group in the main chain and/or removing an -OH from the carboxyl group in the main chain of the amino acid.

The amino acid is not particularly limited, but is preferably a natural amino acid. More preferably, it is selected from valine, isoleucine, leucine, methionine, lysine, phenylalanine, tryptophan, threonine, histidine, arginine, glycine, alanine, serine, tyrosine, cysteine, asparagine, glutamine, proline, aspartic acid, and glutamic acid.

"Each amino acid residue constituting the target peptide" refers to each of the amino acid residues constituting the target peptide. In other words, for example, if the target peptide is composed of 100 amino acid residues, each of the 100 amino acid residues has a main chain dihedral angle Φ and a main chain dihedral angle Ψ. In addition, in the case of an amino acid having a side chain, a side chain dihedral angle χ ¹ exists. However, since glycine residues and alanine residues do not have a side chain, there is no side chain dihedral angle χ ^{1. In addition, proline residues have substantially only one side chain dihedral angle χ 1 .}

In this specification, "chemical shift value" refers to nuclear magnetic resonance chemical shift value.

An outline of the prior art in the technical field and the present invention will be described with reference to FIG.
The conventional method shown in the upper box of FIG. 1 is a method for predicting the chemical shift values of main chain atoms of peptides constituting a protein using the three-dimensional structure of the peptide (three-dimensional coordinates of the constituent atoms) by using the SPARTA program (Non-Patent Document 2).

In conventional prediction methods, the three-dimensional structure of a sequence of three consecutive amino acid residues (hereinafter, amino acid residues may be simply referred to as "residues") in a target peptide is quantitatively compared with structures in a three-residue database that includes experimental chemical shift values.
The score function S is shown in the following formula (1).

式中、ｉは目的ペプチドを構成するアミノ酸の残基番号をｊは３残基データベース中の残基番号を示す。ｒは原子の核種（^１５Ｎ，^１Ｈ^Ｎ，^１Ｈ^α，^１３Ｃ^α，^１３Ｃ^β，^１３Ｃ）を表す。

In the formula, i represents the residue number of an amino acid constituting the target peptide, j represents the residue number in the three-residue database, and r represents the nuclide of the atom ( ¹⁵ N, ¹ H ^N , ¹ H ^α , ¹³ C ^α , ¹³ C ^β , ¹³ C).

The sum on the right side indicates that the scores of each of the three consecutive residues are added together. The terms on the right side indicate the similarity of the amino acid residue type (ResType), the main chain dihedral angle Φ, the main chain dihedral angle Ψ, and the side chain dihedral angle χ ^1, respectively, from the left. In addition, k _{n, r} in the formula are parameters, and values optimized for chemical shift prediction of natural globular proteins are shown in Non-Patent Document 2. 24,166 pieces of data are registered in the three-residue database, and a score function is calculated for all of them, and the weighted average of the 20 chemical shift values with the lowest score value S is used as the estimated value. When applying SPARTA, a conventional method, to an unchained protein, it is first necessary to construct a structure population of the unchained protein and calculate the ensemble average of the predicted values of SPARTA for the three-dimensional structures of each peptide contained in the population, which requires a huge amount of calculation.

Here, a globular protein is intended to mean a peptide that forms a unique three-dimensional structure. Moreover, an unchained protein is intended to mean a peptide in a denatured state in which the chains have been unfolded. Unchained proteins may include denatured proteins, etc. Here, the globular protein is preferably a natural globular protein. Moreover, the denatured protein is preferably a naturally denatured protein. Here, "natural" is intended to mean a protein or peptide that has not been artificially manipulated using a denaturing agent, etc.

The dihedral angles (Φ, Ψ) of the peptide backbone are intended to be the angles of rotation around the covalent bond. Therefore, their values are defined within the range of -π to π, with a periodicity of 2π.

In addition, because the main chain structure of a peptide can be well represented by the (Φ, Ψ) value (hereinafter also referred to as "dihedral angle value"), its frequency distribution (hereinafter referred to as main chain dihedral angle frequency distribution) is widely used in the classification of local structures of natural globular proteins and quality control of three-dimensional structures. This main chain dihedral angle frequency distribution can be used in the structural analysis of unchained proteins. As shown in Figure 1(a), the main chain dihedral angle frequency distribution has several high frequency regions, but also regions where almost no structure appears.

The dihedral angles of the main chain or side chain of a peptide can be determined by measurements such as X-ray diffraction or neutron diffraction, or by computer simulations such as quantum chemical calculations.

In the conventional method, first, about 10 ⁴ to 10 ⁶ dihedral angle values of the target peptide are sampled from the distribution of dihedral angle values of the main chain structure of the target peptide shown in FIG. 1(a) (hereinafter, also referred to as "target peptide dihedral angle values"). The dihedral angle values of the main chain structure of the target peptide include values obtained from a plurality of target peptides. Next, the similarity between all the sampled target peptide dihedral angle values and the dihedral angle values of each structure of 24,166 peptides consisting of three amino acid residues recorded in the three-residue database (hereinafter, also referred to as "reference dihedral angle values") registered in the amino acid three-residue database (Tripeptide Database: TPDB) 30 shown in FIG. 1 is calculated. The similarity can be calculated by a score function. Next, based on the similarity calculated ^by the score function, the chemical shift values of atoms ^13C ', ^13Cα , ^13Cβ , ^15N , ^1HN , or ^1Hα of the similar structures shown in Figure ¹ (b ⁾ are extracted from a database and used as predicted chemical shift ^values of the target peptide.

An example of a three-residue amino acid database is a database in which a three-dimensional structure is registered in a protein structure database such as RCSB Protein Data Bank (RCSB PDB: https://www.rcsb.org/) and a chemical shift value is registered in a database such as BioMagResBank (BMRB: http://www.bmrb.wisc.edu/), and a database is created of the chemical shift values of each atomic nuclide ( ¹³ C′, ¹³ C ^α , ¹³ C ^β , ¹⁵ N, ¹ H ^N , or ¹ H ^α ) of each amino acid residue contained in the protein and the dihedral angle values of the main chain and side chain of three amino acid residues including one amino acid residue before and after the amino acid residue, as described in Non-Patent Document 2. The three-residue amino acid database may include the dihedral angle values of the main chain and side chain of the three amino acid residues, the names of the amino acid residues constituting the three residues, the chemical shift values of the nuclei of each atom constituting each amino acid residue, and the like. The three-residue amino acid database described in Non-Patent Document 2 is constructed from three amino acid residues contained in 200 proteins that satisfy the above requirements.

1-2. Prediction Method of the Present Invention One embodiment of the present invention relates to a method for predicting a nuclear magnetic resonance chemical shift value of a peptide. This embodiment may include the steps of: (1) clustering the occurrence frequency distribution of the main chain dihedral angle values (represented by the values of Φ and Ψ) of each amino acid residue constituting a peptide of interest to obtain a distribution of a plurality of clusters; (2) calculating the similarity between the distribution of each of the obtained clusters and a reference main chain dihedral angle value registered in an amino acid three-residue database using a score function; and (3) predicting the nuclear magnetic resonance chemical shift value of the peptide of interest from the calculated similarity.

(1) Obtaining the cluster distribution In the present invention, as shown in FIG. 1(c), the frequency distribution of the dihedral angle values of the main chain of the target peptide is clustered, and based on the cluster distribution, structures similar to the reference dihedral angle values recorded in the three-residue database described in 1-1 above are extracted. In this clustering, the number of clusters in one dihedral angle frequency distribution diagram may be multiple. Preferably, the number of clusters may be about four to six. In this specification, the dihedral angle value of the main chain may be abbreviated as "main chain dihedral angle value". In addition, the frequency distribution of the dihedral angle value of the main chain may be abbreviated as "main chain dihedral angle distribution".

In addition, prior to the clustering, a pre-clustering may be performed in which the values of the dihedral angles (χ ¹ ) of the side chains of the amino acid residues constituting the target peptide are clustered in advance. In this specification, the dihedral angle of the side chain may be abbreviated as "side chain dihedral angle". The value of the dihedral angle (χ ¹ ) of the side chain may also be referred to as "side chain dihedral angle value" or "χ ¹ value".

The pre-clustering is intended to obtain an occurrence frequency distribution in three peaks (angles) of χ ¹ for amino acid residues other than glycine, alanine, and proline residues among the amino acid residues constituting the target peptide. Specifically, there are three peaks of χ ¹ : "gauche +" in the vicinity of +60°, "gauche -" in the vicinity of -60°, and "trans" in the vicinity of +180°. Here, "+" means "plus", and "-" means "minus". Considering the values of the side chain dihedral angle (χ ¹ ) indicated by these three χ ¹ peaks, when obtaining the main chain dihedral angle value of the target peptide for each amino acid residue, three main chain dihedral angle distributions that differ according to the value of χ ¹ can be obtained for one amino acid residue. The main chain dihedral angle distribution according to the value of χ ¹ may be obtained for at least one peak of χ ¹ and used for clustering. Preferably, the main chain dihedral angle distribution according to the value of χ ¹ is obtained for at least two peaks of χ ¹ and used for clustering. More preferably, the main chain dihedral angle distribution according to the value of χ ¹ is obtained for the three peaks of χ ¹ and used for clustering. A group of main chain dihedral angle distributions of the amino acids constituting the target peptide can be obtained according to the side chain dihedral angles (χ ¹ ) of the amino acid residues constituting the target peptide, and clustering can be performed for each group of main chain dihedral angle distributions of the amino acids.

There is a correlation between the value of χ ¹ and the frequency of occurrence of the main chain dihedral angle. Therefore, even for the same amino acid residue, by changing the frequency distribution of the dihedral angle of the main chain depending on the structure of the side chain (χ ¹ ), it is expected that the correlation between the calculated predicted chemical shift value and the experimental value will be improved. On the other hand, it is considered that the frequency distribution of the χ ¹ value does not change significantly depending on the main chain dihedral angle value.

Here, for glycine residues, alanine residues, and proline residues, χ ¹ cannot be taken into consideration, and therefore for these amino acids, it is preferable to define k^χ ¹ _n,r as "0 (zero)" in formula (2) described later.

The clustering method is not limited as long as it is possible to cluster the distribution of each dihedral angle value within one dihedral angle frequency distribution. For example, clustering can be performed using a mixed Gaussian distribution model, a von Mises distribution model, etc. From the perspective of cost, it is preferable to use a mixed Gaussian distribution model. When performing clustering, parameters can be estimated using the EM algorithm, maximum likelihood estimation, MAP estimation, Bayesian estimation, etc. When performing clustering using a mixed Gaussian distribution model, it is preferable to estimate parameters using the EM algorithm.

Here, the clustering of the distribution of the main chain dihedral angle values is performed two-dimensionally using the Φ value and the Ψ value. In addition, the pre-clustering of the distribution of the side chain dihedral angle values is performed one-dimensionally using only the ^χ1 value.

Furthermore, it is also possible to perform three-dimensional clustering of Φ, Ψ, and ^χ1 values using the distribution of main chain dihedral angle values and side chain dihedral angle values without performing pre-clustering. In three-dimensional clustering, the number of clusters in one dihedral angle frequency distribution map may be multiple. Preferably, the number of clusters may be about 12 to 18.

(2) Calculation of Similarity Hereinafter, the present embodiment will be described taking as an example a case where a Gaussian mixture distribution is used.
In this embodiment, the term related to the dihedral angle of the main chain of the score function shown in formula (1) is expanded to a score function for one Gaussian distribution (one region) constituting the mixed Gaussian distribution. The expansion to the score function is carried out by using an integration method for one Gaussian distribution (one region) to compare the dihedral angle values of each structure of a peptide consisting of three amino acid residues recorded in a three-residue database according to the following formula.

式中、ｉは、式（１）とは異なり、主鎖２面角分布の領域（ガウス分布）を示している。他の記号は、式（１）と同様である。

In the formula, i indicates the region of the main chain dihedral angle distribution (Gaussian distribution), which is different from that in formula (1). The other symbols are the same as those in formula (1).

ここで、θは主鎖２面角（Φ，Ψ）を表している。ＦＷＮは、下記Ｗｒａｐｐｅｄ Ｎｏｒｍａｌ Ｄｉｓｔｒｉｂｕｔｉｏｎ関数を表す。 Moreover, the similarity regarding Φ and Ψ in the formula (2) is expressed by the following integral formula.

Here, θ represents the dihedral angle (Φ, Ψ) of the main chain, and FWN represents the following Wrapped Normal Distribution function.

ＦＷＮのパラメータである平均値μ_θと分散σ_θ ^２は、主鎖２面角分布の各領域についてそれぞれ与えられる。式（３）の積分を実行する。

The average value _μθ and the variance _σθ2 ^, which are parameters of FWN, are given for each region of the main chain dihedral angle distribution.

ここでｘは周期的なので、ｘ’＝ｘ－μ_θとおいても積分結果は変わらない。

Here, since x is periodic, the integration result does not change even if we set x' = x - μ _θ .

ただし、θ’_ｊ＝θ_ｊ－μ_θとおいた。積分の結果を以下に示す。なお、計算の詳細は後述する。 So

Here, θ' _j = θ _j - μ _θ . The integration results are shown below. The details of the calculation will be described later.

_θj for the amino acid three-residue database relates only to the first term of formula (7) and has the same format as formula (1). The second and subsequent terms can be calculated using only the variance σ _θ ² of the Gaussian distribution, so they can be calculated in advance regardless of comparison with the database. Furthermore, if the main chain dihedral angle frequency distribution is assumed to be a "non-periodic Gaussian distribution," the integral result will have the same form as the first two terms of formula (7). That is, the third and fourth terms of formula (7) are terms that appear by taking into account the periodicity of the Gaussian distribution, and these terms can be ignored if σ _θ is sufficiently smaller than the period 2π. As will be described later, if the maximum value of σ _θ in the main chain dihedral angle frequency distribution is set to 30°, sufficient accuracy can be obtained by calculating k=1 alone.

As a result, the calculation time for predicting one peptide using the SPARTA program is equivalent to the calculation time for predicting an unchained protein population for one main chain dihedral angle distribution pattern. In other words, when the SPARTA program is used, the calculation time for predicting one peptide using the SPARTA program is required for a population containing multiple unchained proteins and the unchaining patterns are various, which is times the number of structures. However, with the prediction method of the present invention, even a population containing multiple unchained proteins can predict the structure in a short time, and the time is equivalent to the time when the SPARTA program is used to predict the structure of one peptide. If the main chain dihedral angle frequency distribution can be approximated by 4 to 6 regions, the main chain dihedral angle frequency distribution pattern of three residues is 64 (= 4 ³ ) to 216 (= 6 ³ ). This number is overwhelmingly smaller than the number of structures in the structure population of unchained proteins, which is 10 ⁴ to 10 ⁶ , and the calculation time can be significantly reduced.

Next, an algorithm for calculating the similarity s θ (j) of the main chain dihedral angle from the main chain dihedral angle distribution (μ _θ , σ _θ ² ) of the target peptide for which chemical shifts are to be predicted and the corresponding dihedral angle value ( _{θ j} ) in the amino acid three-residue database will be described. The algorithm is shown in the following steps 1 to 3. The angles in the formula are in radians except for some cases.

分布の標準偏差の単位を度（ｄｅｇｒｅｅ：ｄｅｇ）で表示すると

例えば現実的なσ_{θ，ｄｅｇ}≦３０°の場合、ｋ＝１である。
この計算は、σ_θが変わらなければ再度する必要はない。 Step 1: Estimate k _max that satisfies the calculation accuracy. If the error between the exponential function and the error function is 10 ^-7 or less,

The standard deviation of the distribution is expressed in degrees.

For example, in the realistic case of σ _θ,deg ≦30°, k=1.
This calculation does not need to be repeated if σ _θ does not change.

主鎖２面角頻度分布の領域数（例えば４～６個の領域）だけ計算し、記録する。
この式は、アミノ酸３残基データベースとは無関係である。 Step 2: Perform precomputation.

The number of regions in the main chain dihedral angle frequency distribution (for example, 4 to 6 regions) is calculated and recorded.
This formula is independent of the amino acid three residue database.

得られた値を２乗し、ステップ２の値を加えてｓ_θ（ｊ）を計算する。
この計算を、アミノ酸３残基データベースの全てについて計算する。 Step 3: The dihedral angle value (θ _j ) from the center of the distribution is obtained, and the score function s _θ (j) is calculated.

The resulting value is squared and the value from step 2 is added to calculate s _θ (j).
This calculation is carried out for the entire three-residue amino acid database.

Next, a specific calculation of the above-mentioned formula (6) is shown below, where the prime is omitted.

First, perform the variable substitution x=t-2πk.

The first term is the same as the second moment calculation of the Gaussian distribution, since it simply involves dividing the integral range, integrating, and then summing.

The second term then performs the integration normally.

となる。 Here, k=0 is 0, so if we write down the same absolute value of k, we get

It becomes.

を使って変換する。 Finally, the third term is the error function

Convert using .

を実行する。

i.e. variable transformation

Execute.

となる。 Here, when k=0,

It becomes.

となる。ここで、θ_ｊ ^２が現れる２つの式は、和記号の後半の式は、常にｋが一つ前の前半とキャンセルする。 Furthermore, if we write down the same absolute value of k as in the second term, we get

Here, in the two equations in which θ _j ² appears, the latter half of the sum symbol always cancels out the first half of the equation with the previous k.

となる。 For example, if we write down number 21 up to k=4,

It becomes.

となる。 therefore,

It becomes.

となる。 Therefore, using the results of paragraphs 1, 2, and 3,

It becomes.

となる。 Finally, returning to the definition of _θj , we rewrite equation 24 as follows:

It becomes.

In this way, reference amino acids that have dihedral angle values similar to those of each amino acid residue in the target peptide can be extracted.

(3) Prediction of Chemical Shift Values In the method (2) above, for each combination of amino acid residues of the target peptide and main chain dihedral angle distribution, the chemical shift values of the nuclei of each atom (e.g., ^15N , ^{1HN, 1Hα , 13Cα} , ^13Cβ , ^{13C )} corresponding to three amino acid residues extracted from the amino acid three-residue database as being similar are obtained. The ^sum of the chemical shift values of all combinations of the main chain dihedral angle distribution of the target peptide multiplied by the frequency probability of each ^combination is taken, and ^this is used as the predicted chemical shift value of the nuclide of each atom in each amino acid residue of the target peptide.

When performing pre-clustering, the frequency probability of each of "gauche+", "gauche-", and "trans" is multiplied by the chemical shift values of all combinations of the main chain dihedral angle distribution of the target peptide, and the sum is taken to obtain the predicted chemical shift value of the nuclide of each atom in each amino acid residue of the target peptide.

2. Prediction system 1000 and prediction device 10
2-1. Prediction system 1000
2 shows a hardware configuration of a prediction system 1000 (hereinafter, also simply referred to as "prediction system 1000") for predicting a nuclear magnetic resonance chemical shift value of a peptide, which includes a prediction device 10 (hereinafter, also simply referred to as "prediction device 10") for predicting a nuclear magnetic resonance chemical shift value of a peptide according to the first embodiment. In the prediction system 1000, the prediction device 10 is communicably connected to an amino acid three-residue database 30 via a wired or wireless network.

2-2. Prediction device 10
3 shows a hardware configuration of the prediction device 10. The prediction device 10 may be connected to an input device 111, an output device 112, and a storage medium 113.

In the prediction device 10, the CPU 101, memory 102, ROM (read only memory) 103, recording device 104, communication interface (I/F) 105, input interface (I/F) 106, output interface (I/F) 107, and media interface (I/F) 108 are connected to each other via a bus 109 so that data can be communicated between them. The memory 102 and recording device 104 are sometimes collectively referred to simply as a storage unit. The recording device 104 non-volatilely records an operation system (OS) 1041 that provides a graphical user interface environment such as Windows (registered trademark) manufactured and sold by Microsoft Corporation in the United States, a prediction program 1042 that is the application software of the present invention, and a target peptide dihedral angle value database 1043 that stores the target peptide dihedral angle values.

The CPU 101 is a processing unit of the prediction device 10. The CPU 101 executes the OS 1041 and the prediction program 1042 stored in the recording device 104 or the ROM 103 in cooperation with each other, and performs the processing of steps S1 to S5 shown in FIG. 5, whereby the computer functions as the prediction device 10.

ROM 103 is composed of a mask ROM, PROM, EPROM, EEPROM, etc., and stores computer programs executed by CPU 101 and data used therein. CPU 101 may be MPU 101. ROM 103 stores a boot program executed by CPU 101 when prediction device 10 is started, and programs and settings related to the operation of the hardware of prediction device 10.

The memory 102 is composed of a RAM (random access memory) such as an SRAM or a DRAM. The memory 102 is used to read out computer programs recorded in the ROM 103 and the recording device 104. The memory 102 is also used as a working area when the CPU 101 executes these computer programs.

The communication I/F 105 is composed of serial interfaces such as USB, IEEE 1394, RS-232C, parallel interfaces such as SCSI, IDE, IEEE 1284, analog interfaces such as D/A converters and A/D converters, a network interface controller (NIC), etc. Under the control of the CPU 101, the communication I/F 105 receives data from the measurement unit 30 or other external devices, and transmits or displays information stored or generated by the prediction device 10 to the outside as necessary. The communication I/F 105 receives data from an external database via a network.

The input I/F 106 is composed of, for example, serial interfaces such as USB, IEEE 1394, RS-232C, parallel interfaces such as SCSI, IDE, IEEE 1284, and analog interfaces including D/A converters and A/D converters. The input I/F 106 accepts character input, clicks, voice input, and the like from the input device 111. The accepted input contents are stored in the memory 102 or the recording device 104.

The input device 111 is composed of a touch panel, a keyboard, a mouse, a pen tablet, a microphone, etc., and is used to input characters or voice to the prediction device 10. The input device 111 may be connected from outside the prediction device 10 or may be integrated with the prediction device 10.
The output I/F 107 is configured, for example, by an interface similar to that of the input I/F 106. The output I/F 107 outputs information generated by the CPU 101 to the output device 112. The output I/F 107 outputs information generated by the CPU 101 and stored in the recording device 104 to the output device 112.

The output device 112 is composed of, for example, a display, a printer, etc., and displays the measurement results transmitted from the measurement unit 30 as well as various operation windows and analysis results in the prediction device 10.

The media I/F 108 reads, for example, application software etc. stored in the storage medium 113. The read application software etc. is stored in the memory 102 or the recording device 104. The media I/F 108 also writes information generated by the CPU 101 to the storage medium 113. The media I/F 108 writes information generated by the CPU 101 and stored in the recording device 104 to the storage medium 113.

The storage medium 113 is composed of a flexible disk, a CD-ROM, a DVD-ROM, etc. The storage medium 113 is connected to the media I/F 108 by a flexible disk drive, a CD-ROM drive, a DVD-ROM drive, etc. The storage medium 113 may store application programs, etc. for the computer to execute operations.

The CPU 101 may obtain application software and various settings necessary for controlling the prediction device 10 via a network instead of reading them from the ROM 103 or the recording device 104. The application program may be stored in an auxiliary storage unit of a server computer on the network, and the prediction device 10 may access the server computer to download the computer program and store it in the ROM 103 or the recording device 104.

In addition, an operation system that provides a graphical user interface environment, such as Windows (registered trademark) manufactured and sold by Microsoft Corporation, is installed in the ROM 103 or the recording device 104. The application program according to the second embodiment runs on the operating system. In other words, the prediction device 10 may be a personal computer or the like.

(2) Functions of Prediction Apparatus 10 Fig. 4 shows the functional configuration of the prediction apparatus 10. The CPU 101 of the prediction apparatus 10 functions as a sample dihedral angle value acquisition unit 1, a reference dihedral angle value acquisition unit 3, a clustering unit 5, a cluster distribution comparison unit 7, and a chemical shift prediction unit 9. The sample dihedral angle value acquisition unit 1 corresponds to step S1 in Fig. 5 described later, the reference dihedral angle value acquisition unit 3 corresponds to step S2 in Fig. 5 described later, the clustering unit 5 corresponds to step S3 in Fig. 5 described later, the cluster distribution comparison unit 7 corresponds to step S4 in Fig. 5 described later, and the chemical shift prediction unit 9 corresponds to step S5 in Fig. 5 described later.

(3) Processing Executed by Prediction Program 1042 FIGS. 5 and 6 show the flow of processing executed by the prediction program 1042. FIG.

5, the CPU 101 acquires the main chain dihedral angle values of the target peptide from the sample dihedral angle value database 1043 stored in the recording device 104. This process is started when the user issues a request to acquire the main chain dihedral angle values of the target peptide from the input device 111.

In step S2 in Fig. 5, the CPU 101 acquires the reference dihedral angle values stored in the amino acid three-residue database 30 shown in Fig. 2 via the communication I/F 107. This process is started when the user issues a request to acquire the reference dihedral angle values from the input device 111.

Here, steps S1 and S2 in Fig. 5 may be reversed. Also, the CPU 101 may acquire the reference dihedral angle values stored in the amino acid three-residue database 30 shown in Fig. 2 in advance and record them in the recording device 104.

In step S3 of Fig. 5, the CPU 101 performs clustering of the main chain dihedral angle values of the target peptide obtained in step S1 to obtain a cluster distribution. The clustering method is as described in 1-2. (1) above.

In step S4 of Fig. 5, the CPU 101 compares the cluster distribution obtained by the clustering in step S3 with the reference dihedral angle value obtained in step S2 to calculate the similarity. The method of calculating the similarity is as described in 1-2. (2) above.

In step S5 of FIG. 5, based on the similarity obtained in step S4, the CPU 101 obtains the chemical shift values of the nuclei of each atom corresponding to the reference dihedral angle values similar to the cluster distribution obtained in step S3 as predicted values of the chemical shifts of the nuclei of each atom contained in each amino acid of the target peptide.

Using Figure 6, the processing of step S5 shown in Figure 5 by CPU 101 will be explained in more detail.

In step S51 shown in FIG. 6, the CPU 101 extracts structures having reference dihedral angle values that were highly similar in step S4 shown in FIG. 5 from the amino acid three-residue database 30.

Next, in step S52 shown in FIG. 6, the CPU 101 obtains chemical shift values corresponding to the structure extracted in step S51 from the amino acid three-residue database 30.

Next, in step S53 shown in FIG. 6, the CPU 101 determines whether chemical shift values have been obtained for all atomic nuclei, and if not all chemical shift values have been obtained (if "No"), the CPU 101 returns to step S52 and obtains chemical shift values for atomic nuclei for which chemical shifts have not yet been obtained.

If the CPU 101 has obtained all chemical shift values for all atomic nuclides in step S53 shown in FIG. 6 (if "Yes"), the CPU 101 proceeds to step S54 and records the chemical shift values obtained for the nuclides in the recording device 104 as predicted chemical shift values of the target peptide.

Details of steps S51 to S54 shown in Fig. 6 follow 1-2.(2) above.
Here, the extraction of structures having highly similar reference dihedral angle values in step S51 shown in Fig. 6 may be performed by inputting an extraction instruction from the input device 111 by the user after step S4 shown in Fig. 5, and the extraction instruction may be accepted by the CPU 101. Alternatively, the end of step S4 shown in Fig. 5 may be used as a trigger to cause the CPU 101 to automatically extract structures having highly similar reference dihedral angle values from the amino acid three-residue database 30.

6 may be acquired by a user inputting an acquisition instruction from the input device 111 after step S51, and the acquisition instruction may be accepted by the CPU 101. Alternatively, the end of step S51 may be used as a trigger to automatically acquire the chemical shift values from the amino acid three-residue database 30 based on the extracted structure by the CPU 101.
3. Prediction system 2000 and prediction device 20
3-1. Prediction System 2000
2 shows a hardware configuration of a prediction system 2000 (hereinafter, also simply referred to as "prediction system 2000") for predicting a nuclear magnetic resonance chemical shift value of a peptide, which includes a prediction device 20 (hereinafter, also simply referred to as "prediction device 20") for predicting a nuclear magnetic resonance chemical shift value of a peptide according to a second embodiment. In the prediction system 2000, the prediction device 20 is communicably connected to an amino acid three-residue database 30 via a wired or wireless network.

3-2. Prediction device 20
7 shows the hardware configuration of the prediction device 10. The prediction device 20 may be connected to an input device 211, an output device 212, and a storage medium 213.

In the prediction device 20, the CPU 201, memory 202, ROM (read only memory) 203, recording device 204, communication interface (I/F) 205, input interface (I/F) 206, output interface (I/F) 207, and media interface (I/F) 208 are connected to each other via a bus 209 so that data can be communicated between them. The memory 202 and recording device 204 are sometimes collectively referred to simply as a storage unit. The recording device 204 non-volatilely records an operation system (OS) 2041 that provides a graphical user interface environment such as Windows (registered trademark) manufactured and sold by Microsoft Corporation in the United States, a prediction program 2042 that is the application software of the present invention, and a target peptide dihedral angle value database 2043 that stores the target peptide dihedral angle values.

The CPU 201, memory 202, ROM 203, recording device 204, communication I/F 205, input I/F 206, output I/F 207, media I/F 208, bus 209, operation system (OS) 2041, and target peptide dihedral angle value database 2043 correspond to the CPU 101, memory 102, ROM 103, recording device 104, communication I/F 105, input I/F 106, output I/F 107, media I/F 108, bus 109, operation system (OS) 1041, and target peptide dihedral angle value database 1043, respectively, in the prediction device 10.

(2) Functions of Prediction Device 20 Fig. 8 shows the functional configuration of the prediction device 20. The CPU 201 of the prediction device 20 functions as a sample dihedral angle value acquisition unit 21, a reference dihedral angle value acquisition unit 23, a side chain dihedral angle value pre-clustering unit 24, a main chain dihedral angle value clustering unit 25, a cluster distribution comparison unit 27, and a chemical shift prediction unit 29. The sample dihedral angle value acquisition unit 21 corresponds to step S21 in Fig. 9 described later, the reference dihedral angle value acquisition unit 23 corresponds to step S22 in Fig. 9 described later, the side chain dihedral angle value pre-clustering unit 24 corresponds to step S23 in Fig. 9 described later, the main chain dihedral angle value clustering unit 25 corresponds to step S24 in Fig. 9 described later, the cluster distribution comparison unit 27 corresponds to step S25 in Fig. 9 described later, and the chemical shift prediction unit 29 corresponds to step S26 in Fig. 9 described later.

(3) Processing Executed by the Prediction Program 2042 FIGS. 9 and 6 show the flow of processing executed by the prediction program 2042. FIG.

9, the CPU 201 acquires the main chain dihedral angle values and the side chain dihedral angle values of the target peptide from the sample dihedral angle value database 2043 stored in the recording device 204. This process is started when the user issues a request to acquire the main chain dihedral angle values and the side chain dihedral angle values of the target peptide from the input device 211.

In step S22 in Fig. 9, the CPU 201 acquires the reference dihedral angle values stored in the amino acid three-residue database 30 shown in Fig. 2 via the communication I/F 207. This process is started when the user issues a request to acquire the reference dihedral angle values from the input device 211.

Here, steps S21 and S22 in Fig. 9 may be reversed. Also, the CPU 201 may acquire the reference dihedral angle values stored in the amino acid three-residue database 30 shown in Fig. 2 in advance and record them in the recording device 204.

In step S23 of FIG. 9, the CPU 201 performs pre-clustering using the side chain dihedral angle values. The method of pre-clustering is as described in 1-2. (1) above.

9, the CPU 201 performs clustering of the main chain dihedral angle values of the target peptide for each χ ¹ according to the χ ¹ of the side chain obtained in step S23, and obtains a cluster distribution. The clustering method is as described in 1-2. (1) above.

9, the CPU 201 compares the cluster distribution obtained by the clustering in step S24 with the reference dihedral angle value obtained in step S22 to calculate the similarity. The method of calculating the similarity is as described in 1-2. (2) above.

In step S26 of FIG. 9, the CPU 201 obtains the chemical shift values of the nuclei of each atom corresponding to the reference dihedral angle values similar to the cluster distribution obtained in step S24 as predicted values of the chemical shifts of the nuclei of each atom contained in each amino acid of the target peptide based on the similarity obtained in step S25.

Using Figure 6, the processing of step S26 shown in Figure 9 by CPU 201 will be explained in more detail.

In step S51 shown in FIG. 6, the CPU 101 extracts structures having reference dihedral angle values that were found to have a high degree of similarity in step S25 shown in FIG. 9 from the amino acid three-residue database 30.

Next, in step S52 shown in FIG. 6, the CPU 201 obtains chemical shift values corresponding to the structure extracted in step S51 from the amino acid three-residue database 30.

Next, in step S53 shown in FIG. 6, the CPU 201 determines whether chemical shift values have been obtained for all atomic nuclei, and if not all chemical shift values have been obtained (if "No"), the CPU 201 returns to step S52 and obtains chemical shift values for atomic nuclei for which chemical shifts have not yet been obtained.

If the CPU 201 has obtained all chemical shift values for all atomic nuclides in step S53 shown in FIG. 6 (if "Yes"), the CPU 201 proceeds to step S54 and records the chemical shift values obtained for the nuclides in the recording device 204 as predicted chemical shift values of the target peptide.

Details of steps S51 to S54 shown in Fig. 6 follow 1-2.(2) above.
Here, the extraction of structures having highly similar reference dihedral angle values in step S51 shown in Fig. 6 may be performed by inputting an extraction instruction from the input device 211 by the user after step S25 shown in Fig. 9, and the extraction instruction may be accepted by the CPU 201. Alternatively, the CPU 201 may automatically extract structures having highly similar reference dihedral angle values from the amino acid three-residue database 30, triggered by the end of step S25 shown in Fig. 9.

6, the user may input an instruction to acquire the chemical shift value from the input device 211 after step S51, and the CPU 201 may accept the instruction to acquire the chemical shift value. Alternatively, the end of step S51 may be used as a trigger to automatically acquire the chemical shift value from the amino acid 3 residue database 30 based on the extracted structure by the CPU 201.

4. Storage medium having prediction program recorded thereon The prediction program 1042 and the prediction program 2042 can be executed on a computer as a computer program for predicting the nuclear magnetic resonance chemical shift value of the peptide according to the present invention.

The computer program may be provided as a program product such as a storage medium. The computer program is stored in a storage medium such as a hard disk, a semiconductor memory element such as a flash memory, an optical disk, etc. The format of the program stored in the storage medium is not limited as long as the control unit can read the program. It is preferable that the storage in the storage medium is non-volatile.

5. Verification of the Effects of the Invention 5-1. Verification of the First Embodiment Figure 10 shows a comparison of the patterns of the nuclear magnetic resonance chemical shift values predicted by the prediction method according to the first embodiment and the nuclear magnetic resonance chemical shift values predicted by the SPARTA program for unfolded apomyoglobin. In the figure, C, CA, CB, N, NH, and ^HA respectively represent ^13C ', ^13Cα , ^{13Cβ , 15N} , ^1HN , and ^1Hα . The red line (gray line in gray scale) represents the predicted value of the present invention ^, and the black line represents the predicted ^value by the SPARTA program. The vertical axis represents the chemical shift value (ppm), and the horizontal axis represents the residue number. The residue number is set to 1 for the N-terminus.

The chemical shift values obtained by the prediction method of the first embodiment were almost identical to the chemical shift values predicted by the SPARTA program.

Therefore, it was considered that the prediction method of the present invention has predictive function comparable to that of the SPARTA program.

5-2. Verification of the Second Embodiment The effect of the prediction method according to the second embodiment was verified.
Figure 11 shows three clustering results obtained by pre-clustering with the ^χ1 value of glutamine residues. The left column shows "gauche+", the middle column shows "trans", and the right column shows "gauche-". The top row shows the clustering results, and the bottom row shows the predicted main chain dihedral angle data.

Figure 12 shows the reproducibility of chemical shift experimental data of urea-denatured apomyoglobin. Calculations were performed for four nuclides, C, CA, ^CB , and ^HA (representing ^13C ', ^13Cα , ^{13Cβ ,} and ^1Hα , respectively). Open circles indicate predicted values calculated without considering the value of ^χ1 . Black circles indicate predicted values calculated with the value of ^χ1 considered. The predicted values calculated with the value of ^χ1 considered were closer to the experimental values than the predicted values calculated without considering the value of ^χ1 . Therefore, by correcting the weight of each Gaussian distribution of the main chain dihedral angle distribution of each amino residue with the value of ^χ1 , it became possible to bring the predicted values closer to the experimental values.

10, 20 Prediction device 101, 201 Processing unit

Claims (8)

A method for predicting a nuclear magnetic resonance chemical shift value of a peptide, comprising the steps of:
A step of clustering the occurrence frequency distribution of the dihedral angle values (represented by values of Φ and Ψ) of the main chain of each amino acid residue constituting the target peptide to obtain a distribution of multiple clusters;
calculating a similarity between the distribution of each of the obtained clusters and a reference main chain dihedral angle value registered in an amino acid three-residue database using a score function;
and predicting the nuclear magnetic resonance chemical shift value of the target peptide from the calculated similarity. The prediction method according to claim 1, further comprising performing pre-clustering based on the dihedral angle values of the side chains (represented by the value of ^χ1 ) before clustering the frequency distribution of the dihedral angle values of the main chains of each amino acid residue. The prediction method described in claim 1 or 2, which is for predicting the nuclear magnetic resonance chemical shift value of an unchained protein. A prediction method according to any one of claims 1 to 3, wherein clustering is performed using a Gaussian mixture distribution. The prediction method described in claim 4, wherein the calculation of similarity is performed by an integral method. On the computer,
A step of clustering the occurrence frequency distribution of the dihedral angle values (represented by values of Φ and Ψ) of the main chain of each amino acid residue constituting the target peptide to obtain a distribution of a plurality of clusters;
calculating a similarity between the distribution of each cluster obtained and a reference main chain dihedral angle value registered in an amino acid three-residue database using a score function;
predicting a nuclear magnetic resonance chemical shift value of the target peptide from the calculated similarity;
A prediction program for the nuclear magnetic resonance chemical shift value of the peptide. A prediction device for predicting a nuclear magnetic resonance chemical shift value of a peptide, comprising:
The prediction device includes a processing unit,
The processing unit includes:
The frequency distribution of the dihedral angle values (represented by values of Φ and Ψ) of the main chain of each amino acid residue constituting the target peptide is clustered to obtain a distribution of multiple clusters;
Using a score function, the similarity between the distribution of each obtained cluster and the reference main chain dihedral angle value registered in the amino acid three residue database is calculated;
predicting a nuclear magnetic resonance chemical shift value of the target peptide from the calculated similarity;
The prediction device. A method for predicting a nuclear magnetic resonance chemical shift value of a peptide, comprising the steps of:
A step of clustering the occurrence frequency distribution of the dihedral angle values of the main chains (represented by values of Φ and Ψ) and the dihedral angle values of the side chains (represented by values of ^χ1 ) of each amino acid residue constituting the target peptide, and obtaining a three-dimensional distribution of multiple clusters;
calculating a similarity between the distribution of each of the obtained clusters and a reference main chain dihedral angle value registered in an amino acid three-residue database using a score function;
and predicting the nuclear magnetic resonance chemical shift value of the target peptide from the calculated similarity.

Images