JP2017174057A - Method for predicting partial cubic structure of protein - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method capable of easily predicting a cubic structure of a protein with sufficient accuracy.SOLUTION: A cubic structure of a protein can be predicted in a manner that: overlaps cubic structures of at least two types of known proteins having high amino-acid sequence homology with respect to a prediction object protein so as to maximize an overlapped portions and identifies plural overlapped portions and plural non-overlapped portions; adopts the overlapped portion directly as a predicted partial (core) structure of the prediction object protein; and extracts plural predicted partial (component) structures by executing partial structural prediction with the non-overlapped portions of the prediction object protein as molds and converts combinations of the core and plural types of components into a thermodynamically stable structural group by simulating the same at a temperature in a body.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a method for predicting a partial tertiary structure of a protein.

  Currently, Ligand-Based Drug Design (LBDD) method and Structure-Based Drug Design (SBDD) method are widely used as in silico screening techniques.

  In the LBDD method, some existing drugs having a pharmacological effect are required, and drug-drug interaction characteristics are predicted to generate new candidate compounds.

  On the other hand, in the SBDD method, the protein structure to be a drug target needs to be clarified, and a new candidate compound is generated based on the target site of the protein.

  Thus, in the conventional method, when screening for a new target protein whose steric structure is not known, not only the SBDD method but also the LBDD method can be performed if multiple existing drugs are not known. There wasn't.

  Several methods for predicting protein tertiary structures have been proposed (Non-Patent Documents 1 and 2), but there has been no method that can be easily predicted with sufficient accuracy to be used in the SBDD method.

D. Petrey, et al., Using Multiple Structure Alignments, Fast model Building, and Energetic Analysis in Fold Reecognition and Homology Modeling. Proteins. 53, 430-435 (2003) K. Zhu, et al., Antibody structure determination using a combination of homology modeling, energy-based refinement, and loop prediction.Proteins. 82, 1646-1655 (2014)

  An object of the present invention is to provide a method capable of easily predicting a protein tertiary structure with sufficient accuracy.

As a result of intensive studies to solve the above problems, the present inventors have
(I) By overlapping the three-dimensional structures of at least two known proteins having high amino acid sequence homology with the protein to be predicted so that the overlapping portions are maximized with each other, Determine multiple parts that do not overlap,
(Ii) For overlapping portions, adopt as the predicted partial structure (nucleus) of the prediction target protein corresponding to that portion as it is,
(Iii) For non-overlapping parts, a plurality of types of predicted partial structures (parts) are extracted by performing partial structure prediction using the partial structure of the protein to be predicted corresponding to the part as a template,
(Iv) The combination of the nucleus and multiple types of parts is the predicted overall structure of the protein,
(V) The predicted overall structure obtained in this way is converted into a stable structure group by moving (simulating) the temperature in the living body.
By adopting the methodology, it has been found that the structure can be predicted with high accuracy and simplicity, and the present invention has been completed.

That is, the present invention includes the following embodiments.
Item 1.
A method for predicting a partial three-dimensional structure of a protein, comprising the following steps:
(1) By overlapping the three-dimensional structures of at least two known proteins having high amino acid sequence homology with the protein to be predicted so that the overlapping portions are maximized with each other, Determining one or more non-overlapping parts;
(2) A step of determining each portion having a three-dimensional structure determined in step (1) as a primary predicted partial structure of a protein to be predicted;
(3) An amino acid partial sequence of at least one kind of known protein having high homology with respect to each part of the protein to be predicted corresponding to the amino acid sequence of each part having a non-overlapping three-dimensional structure determined in step (1) Determining and determining each of the partial structures corresponding to the amino acid partial sequence among the three-dimensional structures of the known proteins as primary predicted partial structure candidates for the part of the protein to be predicted;
(4) By combining each of the primary predicted partial structure determined in step (1) and the primary predicted partial structure candidate determined in step (3) with each other, a primary predicted overall structure candidate is obtained. (5) The molecular dynamics calculation is performed for each of the primary predicted overall structure candidates determined in (5) and (4), and the most stable structure is the most common among the candidates. A step of extracting a partial structure and determining the partial structure as a second predicted partial structure of the protein to be predicted.
Item 2.
Item 2. The method according to Item 1, wherein the thermodynamic behavior in the living body is simulated by the molecular dynamics calculation in the step (5).
Item 3.
Item 3. The method according to Item 1 or 2, wherein the extraction of the partial structure in the step (5) is performed by clustering.

  According to the present invention, it is possible to provide a method capable of easily predicting a protein tertiary structure with sufficient accuracy.

  In addition, unlike the conventional homology modeling method, the method of the present invention can predict a three-dimensional structure even when a known protein exhibiting high amino acid sequence homology over the entire core domain of the protein to be predicted cannot be identified. This is advantageous.

It is the figure which showed the method of this invention typically. It is a schematic diagram of the drug screening using the method of the present invention. 1 is a drawing comparing a predicted structure obtained by the method of the present invention with a three-dimensional structure obtained by X-ray crystal analysis.

1. Process (1)
In step (1), the three-dimensional structures of at least two known proteins having high amino acid sequence homology with the protein to be predicted are overlapped so that the overlapping portion is maximized. This is a step of determining one or more portions that do not overlap with each other.

  The three-dimensional structure of a known protein having high amino acid sequence homology with the protein to be predicted can be searched for by a known means. Although not particularly limited, for example, it can be searched by homology search from protein structures registered in Protein Data Bank (PDB) using BLAST or the like. In addition, FASTA or the like can be used for the search.

2. Process (2)
Step (2) is a step of determining each portion having a three-dimensional structure determined in step (1) as the primary predicted partial structure of the protein to be predicted. For convenience, the primary prediction partial structure determined in this way may be referred to as a “nucleus”.

3. Step (3)
Step (3) is an amino acid of at least one kind of known protein having high homology with respect to each part of the protein to be predicted corresponding to the amino acid sequence of each part that does not overlap in the three-dimensional structure determined in step (1). In this step, a partial sequence is determined, and each of the partial structures corresponding to the amino acid partial sequence in the three-dimensional structure of the known protein is determined as a primary predicted partial structure candidate for that part of the protein to be predicted.

  For convenience, the primary predicted partial structure determined in this way may be referred to as a “component”.

  In order to determine the amino acid partial sequence of a known protein having high homology for each of these parts, a known means can be used. Although not particularly limited, for example, it can be searched by homology search from protein structures registered in PDB using BLAST or the like. In addition, FASTA or the like can be used for the search.

  At this stage, it is preferable to determine at least two or more types of parts corresponding to each part because prediction accuracy can be improved when structure prediction is performed in combination with a nucleus as described later.

4). Process (4)
In the step (4), the primary prediction partial structure determined in the step (1) and the primary prediction partial structure candidate determined in the step (3) are combined with each other, so that the entire primary prediction This is a step of determining structure candidates.

  The primary overall prediction structure candidate is a combination of a nucleus and a part. If the number of types increases, the number of combinations increases, and the number of candidates increases. It is preferable that the number of candidates is larger because prediction accuracy can be improved when performing structure prediction described later.

  For example, if there is one core and there are two parts where there are parts, and two types of parts are obtained at each part, there are four combinations of cores and parts. Will be obtained (FIG. 1). If there are three cores and there are three parts, and two parts are obtained at each part, six combinations of cores and parts can be obtained. It will be.

  For the primary prediction overall structure candidate obtained at this stage, it does not matter even if there are many regions that enter the forbidden region in the Ramachandran plot. This is because even if there are many such regions, an optimal structure can be determined by subsequent structure prediction.

5. Step (5)
In step (5), molecular dynamics calculation is performed for each primary predicted overall structure candidate determined in step (4), and a partial structure that is most stable among the candidates is obtained. This is a step of extracting and determining the partial structure as the second predicted partial structure of the protein to be predicted.

  Although this process is not specifically limited, For example, it can carry out as follows.

  First, the structure is stabilized. Although stabilization is not specifically limited, For example, it can carry out by the Energy Minimization method. Here, the unnatural structure of the protein is relaxed to some extent without considering the temperature fluctuation (molecular vibration) in the living body. The internal energy is expressed as a function described by the interatomic distance, atomic nuclei and dihedral angle, and the differential equation is solved so as to decrease the energy value. At this time, for example, the steepest descent method and the Newton method can be combined according to the magnitude of the force obtained by the atoms.

  Moreover, when relaxing the primary predicted overall structure candidates in the direction of lower energy as described above, for example, buffer water is arranged around each primary predicted overall structure candidate. Can do. More specifically, 12 [Å] of buffer water (TIP3P) can be arranged. Condensation caused by the formation of hydrogen bonds inside the object can be mitigated by the arrangement of buffered water.

  Then, using the molecular dynamics calculation, the first predicted overall structure candidate is shaken. Although not particularly limited, it can be usually performed at a temperature corresponding to the in-vivo temperature. As an example, it can be performed at 310 [K]. Although not particularly limited, it is preferable to increase the temperature slowly. For example, slowly increase the temperature to 310 [K] over 500,000 steps with a step width of 0.0002 [ps].

  After the temperature increase, the number of structural trajectories corresponding to the number of temperature increase steps is accumulated. For example, when the temperature is increased over 5,000,000 steps as in the above example, a structural trajectory of 5,000,000 steps is accumulated. If a sufficient amount of time is simulated in the process of thermal fluctuation, generally unnatural three-dimensional structures can be eliminated.

In this structural relaxation process, molecular dynamics calculation is continued until the amount of mutation of RMSD (Root Mean Square Deviation) of the entire structure becomes sufficiently small. Whether or not a sufficient amount of sampling has been performed can be determined based on whether or not the accumulated RMSD deviation is constant regardless of the number of divisions (Reblocking Analysis). (J. Chem. Phys. 91, 461, 1989)
The partial structure can be extracted by clustering. The clustering is not particularly limited, and a commonly used structure clustering method can be used. For example, after overlapping only the residues that exist in the space 5 [Å] around the drug binding space (to minimize RMSD), similar structures in the order of small RMSD and large overlap It is possible to sort by.

  In this step, the entire structure, not the partial structure, may be determined by a similar process as necessary and as much as possible.

  In the present invention, further, drug screening can be performed using the partially predicted structure obtained as described above, for example, a representative catalyst structure (FIG. 2). The drug screening includes a primary screening including a step of determining whether or not a drug candidate compound can be bound, a secondary screening of a drug candidate compound in consideration of dehydration action and / or structural change, and a similarity clustering of drug candidate compounds. It may consist of

  Although not particularly limited, similarity clustering of drug candidate compounds can be performed in consideration of Finger Print, volume, and the like.

  EXAMPLES Hereinafter, although an Example is given and this invention is demonstrated further more concretely, this invention is not limited to these Examples at all.

Example 1: Preparation of protein predicted structure
The core domain structure of Jarid1B protein was predicted without using any of the family proteins 1A, 1C, and 1D that have high structural similarity, and the three-dimensional structure required for Structure-Based Drug Design (SBDD) was extracted.

  The amino acid sequence of the core domain of Jarid1B has already been reported to be as shown in SEQ ID NO: 1 (117 bases).

  Here, using BLAST, a protein group having a high homology with the amino acid sequence of Jarid1B was searched from protein structures registered in the PDB. Among them, the three-dimensional structures of a plurality of proteins having relatively high homology were superimposed, and the three-dimensional structure was adopted for the portion where both the arrangement and the three-dimensional structure were high.

  On the other hand, with respect to the portion having a low overlap, a highly homologous protein was similarly searched using only a reference sequence having a low overlap, and a three-dimensional structure was extracted. This three-dimensional structure is called “component”. Through this process, it is possible to extract a relatively large protein three-dimensional structure as a nucleus and a three-dimensional partial structure as a small component in the core domain.

  With respect to these large protein three-dimensional structures, a plurality of protein structures were prepared by combining small parts. At this point, it was decided that there would be no problem if there were many forbidden areas in the Ramachandran plot.

  Next, the thermal fluctuation simulation of the protein structure was performed as follows.

  The created predicted structure was shaken at in-vivo temperature (310 [K]) using molecular dynamics method. For each predicted structure, 12 [Å] buffered water (TIP3P) was placed around the protein. After stabilizing the structure by the energy minimization method, the temperature was slowly raised to 310 [K] over 500,000 steps with a step width of 0.0002 [ps]. Later, 5,000,000 steps of structural trajectories were accumulated. If a sufficient time is simulated in the process of this thermal fluctuation, generally unnatural three-dimensional structure can be eliminated.

  Furthermore, protein structure extraction for In Silico drug discovery was performed as follows.

  The three-dimensional structure required for SBDD was extracted by clustering each structural trajectory simulated from each initial structure based on the three-dimensional orientation of residues around the catalytic site.

Test Example 1: Evaluation of the Predicted Structure Obtained The structure of the average catalyst site selected from the structure trajectory obtained by the above-described method and the structure extracted by X-ray crystal analysis (PDB ID: 5A3N) A comparison of these is shown in FIG. In addition, we succeeded in extracting the representative protein structure essential for SBDD by applying structural clustering from the previous trajectory only with the amino acid at the catalytic site.

  When the predicted structure obtained by the present invention is compared with the three-dimensional structure obtained by X-ray crystal analysis, it can be seen that a very similar structure has been successfully extracted. In the predicted structure obtained by the present invention, molecular motion at in-vivo temperature is expressed, so the distance between amino acid residues, that is, the catalytic site space is slightly wider than the X-ray crystal structure. However, the orientation of the residues is slightly different, but the orientation between the residues is maintained. As obtained by the present invention, it can be inferred that the extraction of a plurality of representative structures from the conformation considering the thermal fluctuation at the in vivo temperature is more suitable for analyzing the actual drug interaction by SBDD. The structure prediction method of the present invention can predict a protein structure with high accuracy without using a family protein, and is considered to be a very useful method even in in silico drug discovery.

Claims (3)

A method for predicting a partial three-dimensional structure of a protein, comprising the following steps:
(1) By overlapping the three-dimensional structures of at least two known proteins having high amino acid sequence homology with the protein to be predicted so that the overlapping portions are maximized with each other, Determining one or more non-overlapping parts;
(2) A step of determining each portion having a three-dimensional structure determined in step (1) as a primary predicted partial structure of a protein to be predicted;
(3) Determine the amino acid partial sequence of at least one kind of known protein having high homology with respect to the amino acid sequences of the respective parts that do not overlap in the three-dimensional structure determined in step (1), and determine the three-dimensional structure of the known protein Determining each of the partial structures corresponding to the amino acid partial sequence as a primary predicted partial structure candidate for that part of the protein to be predicted;
(4) By combining each of the primary predicted partial structure determined in step (1) and the primary predicted partial structure candidate determined in step (3) with each other, a primary predicted overall structure candidate is obtained. (5) The molecular dynamics calculation is performed for each of the primary predicted overall structure candidates determined in (5) and (4), and the most stable structure is the most common among the candidates. A step of extracting a partial structure and determining the partial structure as a second predicted partial structure of the protein to be predicted. The method according to claim 1, wherein thermodynamic behavior in a living body is simulated by the molecular dynamics calculation in step (5). The method according to claim 1 or 2, wherein the extraction of the partial structure in the step (5) is performed by clustering.