Analysis method, information processing apparatus, and program

Abstract

To easily analyze a three-dimensional structure of a phosphorylated protein.SOLUTION: An analysis method according to an embodiment includes a three-dimensional structure prediction step and a classification step. In the three-dimensional structure prediction step, a three-dimensional structure of a phosphorylated protein is predicted by substituting phosphorylated serine and phosphorylated threonine in an amino acid sequence of the phosphorylated protein in which a target protein is phosphorylated with aspartic acid and glutamic acid, respectively, and inputting the amino acid sequence after substitution to a calculation model for three-dimensional structure prediction. In the classification step, the phosphorylated protein is classified on the basis of the three-dimensional structure obtained in the three-dimensional structure prediction step.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The embodiments disclosed in this specification and the drawings relate to an analysis method, an information processing device, and a program. [Background technology]

[0002] Proteins are biopolymers consisting of one or more polypeptide chains, which are important components involved in the cell structure in living organisms and play a major role in vital activities. Each polypeptide chain that constitutes a protein is composed of 20 to several hundred amino acid residues bound by peptide bonds. There are about 20 types of amino acids that constitute peptide chains, which are called basic amino acids. The order of amino acid residues in a peptide chain of a protein is called the primary structure of the protein, and the primary structure (amino acid sequence) of a protein determines the three-dimensional structure of the protein (including the secondary structure, tertiary structure, and quaternary structure), and the function of a protein is closely related to the three-dimensional structure. Changes in the three-dimensional structure of a protein are associated with diseases, and similar pathological conditions generally have similar protein structures (see Non-Patent Document 1).

[0003] The reversible process of protein phosphorylation and dephosphorylation regulates all life activities, including cell proliferation, development, differentiation, signal transduction, apoptosis, neuronal activity, muscle contraction, and tumor development. Protein phosphorylation occurs mainly on two types of amino acids, serine and threonine, and tyrosine. These residues have free hydroxyl groups and do not carry a charge themselves. When these residues are phosphorylated, the protein carries a charge, which changes the three-dimensional structure and further changes the activity of the protein. For example, when the neutral amino acids serine (S) and threonine (T) are phosphorylated, negatively charged phosphorylated serine (pS) and phosphorylated threonine (pT) are formed, which changes the three-dimensional structure of the protein and changes its function.

[0004] Abnormal protein phosphorylation often causes the onset of disease. For example, Alzheimer's disease is closely associated with hyperphosphorylated Tau protein (microtubule-associated protein tau) (see Patent Documents 1 and 4), hereditary Parkinson's disease is associated with phosphorylated LRRK2 (see Non-Patent Document 2), and the proliferation of the new coronavirus in the body and damage to the body are also closely associated with protein phosphorylation (see Non-Patent Document 3).

[0005] Therefore, research and detection of phosphorylated proteins have important clinical significance, and phosphorylated amino acid residues can be used as biomarkers for disease diagnosis. For example, phosphorylated proteins T181, S202, and T217 can be used as biomarkers for diagnosing Alzheimer's disease (see Patent Document 1), and phosphorylated DCBLD2 Y750 can be used as a biomarker for determining the malignancy of glioma (see Patent Document 2). Phosphorylated amino acid residues may be used as a biomarker for determining drug responsiveness, for example, phosphorylated Akt may be used as a biomarker for determining drug resistance of tumors to microtubule targeting reagents (see Patent Document 3), or may be used as an action target of drug treatment, for example, a specific antibody that recognizes the phosphorylated epitope of Tau protein (see Patent Document 4), an antibody against the phosphorylated tyrosine residue at position 74 of TOPK, and the like (see Patent Document 5).

[0006] In the prior art, phosphorylated proteins are usually detected by experimental methods. Specifically, the sequence and phosphorylation site of the target protein are analyzed, and the phosphorylated protein is obtained by experimental methods, after which the structure of the phosphorylated protein is identified by experimental methods, such as X-ray, MR, and Cryo-EM.

[0007] However, the experimental method is time-consuming and difficult to study all phosphorylated proteins. For example, Tau protein has a total of 441 amino acids, of which 80 are serine (S) and threonine (T). All of these serines and threonines can be phosphorylated. Therefore, theoretically, the possible states of phosphorylated Tau protein are 2 80 (about 10 24 ) species, and even if only two or less (i.e., one or two) amino acid sites are phosphorylated, the number is 3241. Therefore, it is almost impossible to comprehensively study the structure of phosphorylated Tau protein by simple experimental methods. Moreover, there are about 20,000 kinds of proteins in the human protein group, and most of them are phosphorylated, so simple experimental methods cannot thoroughly study them all. In addition, some proteins are difficult to extract and purify, and the structures of some proteins are difficult to identify by experimental methods, so it is not preferable to study them by experimental methods.

[0008] In light of these circumstances, methods for analyzing phosphorylated proteins using deep learning AI models have attracted widespread attention, and for example, various methods for calculating and predicting phosphorylation sites in proteins have been proposed (see Non-Patent Document 6, and Patent Documents 6 to 10).

[0009] In addition, two AI models proposed in 2021, AlphaFold 2 (see Non-Patent Document 4) and RosseTTAFold (see Non-Patent Document 5), have attracted widespread attention for predicting the three-dimensional structure of proteins. Both of these models are algorithms that predict the three-dimensional structure of proteins from amino acid sequences. Although these algorithms provide rapid and accurate prediction results for predicting protein structure, they cannot predict the three-dimensional structure of phosphorylated proteins. [Prior art documents] [Patent documents]

[0010] [Patent Document 1] China Patent Application Publication No. 114761807 [Patent Document 2] China Patent Application Publication No. 104083775 [Patent Document 3] China Patent Publication No. 103502467 [Patent Document 4] China Patent Application Publication No. 103502272 [Patent Document 5] China Patent Application Publication No. 105646714 [Patent Document 6] JP 2018-206374 A [Patent Document 7] China Patent Application Publication No. 113539364 [Patent Document 8] China Patent Application Publication No. 113450872 [Patent Document 9] China Patent Publication No. 113096722 [Patent Document 10] China Patent Application Publication No. 101710365 [Non-patent literature]

[0011] [Non-Patent Document 1] Shi, Y. et al. “Structure-based classification of tauopathies”, Nature, 598, 359-363 (2021). [Non-Patent Document 2] Kyle B. Fraser et al. "Urinary LRRK2 phosphorylation predicts parkinsonian phenotypes in G2019S LRRK2 carriers", Neurology, 86(11):994-9, (2016) [Non-Patent Document 3] Ryan M. Hekman et al. "Actionable Cytopathogenic Host Responses of Human Alveolar Type 2 Cells to SARS-CoV-2", Molecular Cell, 80(6), 1104-1122, (2020) [Non-Patent Document 4] Jumper et al., "Highly accurate protein structure prediction with AlphaFold", Nature, 596, 583-589, (2021) [Non-Patent Document 5] Baek,et al., "Accurate prediction of protein structures and interactions using a three-track neural network", Science, 373, 871-876, (2021) [Non-Patent Document 6] Fenglin Luo, et al., "DeepPhos: prediction of protein phosphorylation sites with deep learning", Bioinformatics, 35, 2766-2773(2021) Summary of the Invention [Problem to be solved by the invention]

[0012] As described above, the current state of the art is that no AI model or calculation method capable of predicting the three-dimensional structure of phosphorylated proteins has been established to date.

[0013] The present embodiment has been made to address this technical problem, and the purpose of the present embodiment is to provide a computational method for analyzing phosphorylated proteins, which involves computationally analyzing the three-dimensional structures of phosphorylated proteins, classifying and clustering the three-dimensional structures, and identifying disease-related phosphorylated protein biomarkers or drug action targets from the three-dimensional structures.

[0014] Specifically, this embodiment employs a phosphorylation simulation method to convert phosphorylated serine (pS) and phosphorylated threonine (pT) into amino acid sequences that can be analyzed by a conventional AI model, such as AlphaFold 2 or the RosseTTAFold algorithm, by replacing them with negatively charged amino acids, such as aspartic acid and glutamic acid, respectively, and then predict the three-dimensional structure of the replaced protein. This provides a prediction result of the three-dimensional structure of the phosphorylated protein, the structure of which is similar to that of the replaced protein. This embodiment provides a simple, rapid and accurate prediction method for predicting the three-dimensional structure of a phosphorylated protein.

[0015] In addition, by further comparing the three-dimensional structure of the phosphorylated protein with the three-dimensional structure of the non-phosphorylated protein, for example by classification and cluster analysis, phosphorylated proteins with large structural differences before and after phosphorylation are selected based on the comparison results, and a candidate library is constructed. Then, by referring to disease-related information, phosphorylated proteins associated with the disease are recognized as biomarkers or drug action targets.

[0016] One of the problems to be solved by the embodiments disclosed in this specification and the drawings is to easily analyze the three-dimensional structure of phosphorylated proteins. However, the problems to be solved by the embodiments disclosed in this specification and the drawings are not limited to the above problem. Problems corresponding to the effects of each configuration shown in the embodiments described later can also be positioned as other problems. [Means for solving the problem]

[0017] The analysis method according to the embodiment includes a three-dimensional structure prediction step and a classification step. In the three-dimensional structure prediction step, phosphorylated serine and phosphorylated threonine in the amino acid sequence of a phosphorylated protein obtained by phosphorylating a target protein are replaced with aspartic acid and glutamic acid, respectively, and the amino acid sequence after the replacement is input into a calculation model for predicting a three-dimensional structure, thereby predicting the three-dimensional structure of the phosphorylated protein. In the classification step, the phosphorylated protein is classified based on the three-dimensional structure obtained by the three-dimensional structure prediction step. [Brief description of the drawings]

[0018] [Figure 1] FIG. 1 is a schematic diagram showing the steps of the method for analyzing phosphorylated proteins according to this embodiment. [Diagram 2] FIG. 2 is a comparative diagram of the structures of phosphorylated serine (pS) and aspartic acid (D), and phosphorylated threonine (pT) and glutamic acid (E). [Diagram 3] FIG. 3 is a schematic diagram of the steps of analyzing phosphorylated Tau protein in Example 1 by the method according to this embodiment. [Figure 4A] FIG. 4A is a schematic diagram of the prediction step of predicting phosphorylated Tau protein sequences with high probability of existence in Example 1 and the prediction results. [Figure 4B] FIG. 4B is a schematic diagram of the prediction step of predicting phosphorylated Tau protein sequences with high probability of existence in Example 1 and the prediction results. [Figure 5A] FIG. 5A is a schematic diagram of the prediction steps for predicting the three-dimensional structure of phosphorylated Tau protein in Example 1 and the prediction results. [Figure 5B] FIG. 5B is a schematic diagram of the prediction step of predicting the three-dimensional structure of phosphorylated Tau protein in Example 1 and the prediction results. [Figure 6] FIG. 6 shows a visual comparison between the experimental results of the three-dimensional structure of the Tau protein fragment in Example 1 and the predicted results by AlphaFold 2 before and after amino acid substitution by the method according to the present embodiment. [Figure 7A] FIG. 7A is a schematic diagram of the clustering step for classifying and clustering the three-dimensional structures of phosphorylated Tau proteins in Example 1 and the clustering results. [Figure 7B] FIG. 7B is a schematic diagram of the clustering step for classifying and clustering the three-dimensional structures of phosphorylated Tau proteins in Example 1 and the clustering results. [Figure 8A] FIG. 8A shows data comparing predicted results and experimental results before and after amino acid substitution in 11 phosphorylated Tau protein sequences in Example 2. [Figure 8B] FIG. 8B shows data comparing predicted results and experimental results before and after amino acid substitution in the 11 phosphorylated Tau protein sequences of Example 2. [Figure 8C] FIG. 8C shows data comparing predicted results and experimental results before and after amino acid substitution in the 11 phosphorylated Tau protein sequences of Example 2. [Figure 8D] FIG. 8D shows data comparing predicted results and experimental results before and after amino acid substitution of 11 phosphorylated Tau protein sequences in Example 2. [Figure 9] FIG. 9 is a diagram illustrating an example of an information processing device according to the present embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0019] A specific embodiment of the present invention will be described in detail below. Note that the following description of the embodiment is merely for the purpose of explaining the concept of the present invention, and is not intended to limit the present embodiment.

[0020] (Method of Analyzing Phosphorylated Proteins) One embodiment of this embodiment relates to a method for analyzing phosphorylated proteins by computation (hereinafter, sometimes simply referred to as the "method of this embodiment"), which includes a phosphoprotein sequence prediction step of predicting the probability of existence of each phosphorylation combination of a target protein using a computational model and selecting phosphorylated proteins whose probability of existence is equal to or higher than a predetermined probability, a phosphoprotein three-dimensional structure prediction step of replacing phosphorylated serine and phosphorylated threonine in the phosphoprotein sequence obtained by the phosphoprotein sequence prediction step with aspartic acid and glutamic acid, respectively, and predicting the three-dimensional structure of the protein after the replacement using a computational model, and a classification / clustering step of classifying / clustering phosphorylated proteins based on the three-dimensional structure obtained by the phosphoprotein three-dimensional structure prediction step.

[0021] In this embodiment, the "amino acid sequence of a protein" may be simply referred to as the "sequence of a protein" or the "protein sequence".

[0022] Fig. 1 is a schematic diagram showing the steps of the phosphorylated protein analysis method of this embodiment. As shown in Fig. 1, the phosphorylated protein analysis method of this embodiment includes a step of selecting a target protein (10), a step of predicting a phosphorylated protein sequence (20), a step of predicting a phosphorylated protein three-dimensional structure (30), and a classification / clustering step (40).

[0023] (Step (10) of selecting a target protein) The target protein is not particularly limited and may be any protein to be studied, such as various kinases associated with diseases, phosphatases, and proteins in which abnormal phosphorylation mutations may occur, etc. For example, a person (user) who performs the analysis method of this embodiment specifies (selects) a target protein.

[0024] (Phosphoprotein sequence prediction step (20)) In the phosphorylated protein sequence prediction step (20), a computational model is used to predict the probability of existence of each phosphorylation combination of the target protein, and phosphorylated proteins whose existence probability is equal to or greater than a predetermined probability are selected.

[0025] The phosphorylation combination refers to all combinations of phosphorylated proteins in which one, two or more of the serine and threonine residue sites that can be phosphorylated in the amino acid sequence of the protein are phosphorylated. For example, Tau protein itself is a phosphate-containing protein, and research has shown that in a normal mature brain, Tau protein molecules contain 2 to 3 phosphate groups, whereas Tau protein in the brain of a patient with Alzheimer's disease (senile dementia) may be abnormally phosphorylated, with one Tau protein molecule containing 5 to 9 phosphate groups. Therefore, for Tau protein, for example, all combinations of phosphorylated proteins in which 2, 3, 4, 5, 6, 7, 8 and 9 serine and threonine residue sites are phosphorylated, respectively, may be considered, and the probability of existence of each phosphorylated protein may be analyzed.

[0026] The computational model used in the phosphorylated protein sequence prediction step (20) may be the DeepPhos algorithm described in Non-Patent Document 6, but is not limited thereto. For example, the methods described in Patent Documents 6 to 10 may be used.

[0027] DeepPhos is a novel multi-layer CNN (Convolutional Neural Network) architecture, which can accurately predict phosphorylation sites with protein sequence information. Different from previous deep learning methods, DeepPhos is composed of densely connected convolutional neural network (DC-CNN) blocks, and the convolutional layers are simultaneously connected by serial layers within blocks (Intra-BCL), effectively enhancing the flow of phosphorylation information. At the same time, multiple DC-CNN blocks with different window and filter sizes are used to automatically capture important sequence representations of protein phosphorylation sites, which are further accumulated by serial layers between blocks (Inter-BCL) to make the final prediction. DeepPhos outperforms traditional methods in general phosphorylation prediction and also has good performance in predicting kinase-specific phosphorylation sites. Predictions using DeepPhos are not affected by the kinase levels of groups, families, subfamilies, and individuals.

[0028] From the viewpoint of the amount of model calculations and the expected number of candidate phosphorylated proteins, it is preferable to select phosphorylated proteins with a probability of existence of 70% or more, for example, 75% or more, 80% or more, 85% or more, 90% or more, 95% or more, etc.

[0029] That is, the prediction step (20) corresponds to a phosphorylated protein prediction step of predicting the amino acid sequence of a phosphorylated protein in which an amino acid site that can be phosphorylated with a predetermined probability (probability of existence) or more is phosphorylated among the amino acid sequence of the target protein, using a computational model for predicting phosphorylated proteins. For example, the phosphorylated protein prediction step outputs a group of candidates of phosphorylated proteins in which one or more phosphorylation sites are phosphorylated among a plurality of amino acid sites that can be phosphorylated with a predetermined probability or more.

[0030] (Prediction step of three-dimensional structure of phosphorylated protein (30)) In the phosphorylated protein three-dimensional structure prediction step (30), phosphorylated serine and phosphorylated threonine in the phosphorylated protein sequence with high probability of existence obtained in the phosphorylated protein sequence prediction step are replaced with aspartic acid and glutamic acid, respectively, and the three-dimensional structure of the protein after replacement is predicted using a computational model.

[0031] The method of using non-phosphorylated amino acids instead of phosphorylated amino acids by replacing the amino acids at the phosphorylation sites with amino acids having carboxyl side chains is similar to phosphorylation simulation (phosphomid metics) used in experimental methods. As shown in Figure 2, aspartic acid (D) and glutamic acid (E) are structurally similar to phosphorylated serine (pS) and phosphorylated threonine (pT), respectively, and can be substituted for each other in experiments. The phosphorylation of serine and threonine phosphorylates the neutral amino acids serine and threonine to negatively charged amino acid groups. This causes the protein to acquire an electric charge and changes the three-dimensional structure, so that the three-dimensional structure of the phosphorylated protein is different from that of the protein before phosphorylation. The negatively charged amino acids aspartic acid or glutamic acid can be substituted for serine phosphate or threonine phosphate, the former being able to simulate the phosphorylation state of a protein (Pearlman, SM, Serber, Z. & Ferrell, JE A Mechanism for the Evolution of Phosphorylation Sites. Cell 147, 934-946 (2011).).

[0032] The computational model used in the phosphorylated protein three-dimensional structure prediction step (30) can be, for example, the RoseTTAFold algorithm or the AlphaFold 2 algorithm, but is not limited to these two algorithms, and other algorithms that predict protein three-dimensional structures from amino acid sequences used in the field can also be used. The RoseTTAFold algorithm, the AlphaFold 2 algorithm, and other current algorithms cannot directly predict the three-dimensional structure of phosphorylated proteins based on amino acid sequences, but the phosphorylation simulation method can replace phosphorylated amino acid residues with negatively charged amino acids, and predict the three-dimensional structure of the replaced simulated protein using an AI model, thereby indirectly analyzing the three-dimensional structure of phosphorylated proteins.

[0033] RoseTTAFold learns coevolution information directly from MSA (multiple sequence comparison pairs) using nlp (natural language processing) methods. MSA sequences are cropped based on the length of 260. This is because if the sequence is too long, attention calculations in the network will put a heavy burden on the GPU memory. MSA feature preprocessing embeds 21 types of residues in MSA (including empty residues in comparison pairs) into a vector of a certain length in the residue dimension, and adds sine positional encoding to provide positional information, and adds relationship information between MSA and the predicted sequence (query sequence) in the sequence dimension. The information provided by the searched template is 2D residue pair information in the comparison pair domain, including the distance and orientation between residue pairs. In addition, some indices generated by HHSearch search are also added to the 2D template information. The template information is preprocessed with axial attention (template direction and residue direction) and pixel attention, and adds embedding and positional encoding information of the query sequence.

[0034] Compared with RoseTTAFold, MSA in AlphaFold 2 has more processing. According to the perspective with the identity of the query sequence, MSA sequences are randomly deleted in the sequence direction in blocks, which is to save memory and maintain sequence diversity. Then, the remaining sequences are clustered, and only the sequences at the clustering center are input as the MSA of the model, and the sequences that are not at the clustering center are randomly extracted and input to the network as extra MSA. Here, AlphaFold 2 also learns co-evolution information directly in MSA, but AlphaFold 2 utilizes a mask policy, where each position is masked with a probability of 15%, then replaced by one specific rule, and finally looks at the recovery degree of this masked residue. This allows further co-evolution information to be mined and the embeddina to be better. The impact of this loss can be further reflected in the loss function given by it. AlphaFold 2 also performs cropping on sequences, and in the embedding process, pair features (i.e., pair representations) are initialized directly by query sequence information, and add template distance information by pointwise attention. MSA features (i.e., MSA representations) can be jointly projected by query sequence and MSA, and add template angle information. Moreover, the position coding in AlphaFold 2 is a relative position coding, and such coding can handle longer-distance text signals.

[0035] AlphaFold 2 and RoseTTAFold each have their own advantages, and while RoseTTAFold has a faster calculation speed, AlphaFold 2's calculation results are more accurate, so the two can be combined and applied in actual use. For example, Ian R. Humphreys and others have made useful attempts on this (Ian R. Humphreys et al., "Computed structures of core eukaryotic protein complexes", Science, Vol 374, Issue 6573, 11 Nov 2021). Of course, AlphaFold 2 or RoseTTAFold may be used in combination with other algorithms.

[0036] That is, the prediction step (30) corresponds to a three-dimensional structure prediction step in which phosphorylated serine and phosphorylated threonine in the amino acid sequence of a phosphorylated protein, which is a phosphorylated target protein, are replaced with aspartic acid and glutamic acid, respectively, and the amino acid sequence after replacement is input into a computational model for predicting the three-dimensional structure, thereby predicting the three-dimensional structure of the phosphorylated protein.

[0037] (Classification / clustering step (40)) In the classification / clustering step (40), phosphorylated proteins are classified and clustered based on the three-dimensional structures obtained in the phosphorylated protein three-dimensional structure prediction step.

[0038] Classification and clustering of phosphorylated proteins can be performed, for example, by hierarchical clustering.

[0039] Hierarchical clustering is an unsupervised learning clustering method that divides a data set into different groups according to similarity and forms a tree structure. Hierarchical clustering can be divided into two methods: aggregation clustering and divisive clustering. Aggregation clustering combines similar data points from each data point to successively form larger groups until all data points are in the same group. Dividual clustering continuously divides dissimilar data points from all data points into new groups until each group has only one data point. In hierarchical clustering, a similarity metric needs to be defined, and methods such as Euclidean distance, Manhattan distance, cosine similarity, etc. can be used to evaluate the degree of similarity between data points. At the same time, it is necessary to select an appropriate clustering method and stopping conditions to achieve optimal clustering results. The advantages of hierarchical clustering are that it can generate a tree structure, making it easy to visualize and interpret the data set, it does not require a preset number of clusters, it automatically identifies groups in the data set, and it can be applied to small data sets and data with hierarchical structures. In addition, hierarchical clustering can be combined with other clustering methods, such as K-means, DBSCAN, etc., to achieve better clustering results.

[0040] Each category obtained by classification and clustering is treated as a possible candidate phosphoprotein group, and a list of candidate phosphoproteins can be generated.

[0041] That is, the classification / clustering step (40) corresponds to a classification step of classifying phosphorylated proteins based on the three-dimensional structures obtained by the three-dimensional structure prediction step. For example, the classification step classifies a group of phosphorylated protein candidates based on the characteristics of the three-dimensional structures.

[0042] (Method for screening phosphorylated proteins as biomarkers / drug action targets) Another embodiment of this embodiment relates to a method for screening phosphorylated proteins that are biomarkers / drug action targets, in which the three-dimensional structure of a candidate phosphorylated protein obtained by the phosphorylated protein analysis method of this embodiment is compared with the three-dimensional structure of an unphosphorylated protein, and phosphorylated proteins that have large structural differences are selected as biomarkers / drug action targets.

[0043] The method for screening phosphorylated proteins may be performed as a classification step. For example, the classification step compares the three-dimensional structure of a phosphorylated protein (phosphorylated target protein) with the three-dimensional structure of a non-phosphorylated target protein to identify phosphorylated proteins with large structural differences. The classification step also identifies phosphorylated proteins with large structural differences as biomarkers or drug action targets.

[0044] If necessary, a combination of previous research results on the disease can also be selected.

[0045] According to the above-described analysis method of the present embodiment, the three-dimensional structure of a phosphorylated protein can be analyzed easily.

[0046] In the following examples, the method for analyzing phosphorylated proteins according to this embodiment will be specifically described using Tau protein as an example.

[0047] The microtubule system is a skeletal component of nerve cells and is involved in multiple cellular functions. Microtubules consist of microtubule proteins and microtubule-associated proteins, and Tau protein is the microtubule-associated protein with the highest content. The cellular functions of Tau protein in normal brains are to bind to microtubule proteins to promote their assembly to form microtubules, and to bind to formed microtubules to maintain the stability of microtubules, reduce the dissociation of microtubule protein molecules, and induce microtubule bundling. In healthy individuals, six types of isomers can be expressed depending on the cutting method of Tau protein mRNA. Tau protein is a phosphate-containing protein, and Tau protein molecules in normal mature brains contain 2 to 3 phosphate groups, whereas Tau protein in brains with Alzheimer's disease (senile dementia, AD) is abnormally phosphorylated, with one Tau protein molecule containing 5 to 9 phosphate groups, and may lose normal biological functions.

[0048] After abnormal phosphorylation, Tau protein in the brains of AD patients (PHF-Tau) has only 1 / 10 the binding strength with microtubule proteins as normal Tau protein, and loses its biological function of promoting microtubule formation and maintaining microtubule stability. PHF-Tau (paired helical filament-Tau) competes with microtubule proteins to bind to normal Tau protein and other large microtubule-associated proteins, depriving these proteins from microtubules and disassembling them, destroying the normal microtubule system. Abnormally phosphorylated Tau protein assembles itself into PHF / NFT (nerve fiber entanglement) structures. In the brains of AD patients, the microtubule structure of related neurons is widely destroyed, normal axonal transport is impaired, leading to synapse loss and neuronal functional damage, resulting in brain neurodegenerative diseases.

[0049] Based on the concept of the present invention of discovering phosphorylated proteins with biological significance by predicting the structure of modified proteins and performing cluster analysis, a person skilled in the art can further apply the method of the present invention to other types of post-translational modifications of proteins, such as glycosylation and acetylation of proteins. For example, a method for predicting the three-dimensional structure of glycosylated proteins is provided, and different glycosylated proteins are classified and clustered based on the predicted structures, and each clustering result is used as a candidate biomarker, from which new biomarkers are discovered.

[0050] Although the phosphorylated protein analysis method of the present embodiment has been described above based on the embodiment, the present embodiment is not limited thereto. As long as it does not deviate from the spirit of the present embodiment, various modifications conceived by a person skilled in the art and other forms constructed by combining some of the components of the embodiment are also included in the scope of the present embodiment.

[0051] (Example) The method for analyzing phosphorylated proteins according to the present embodiment will be specifically described below using a specific protein (Tau protein) sequence as an example. Note that the present embodiment is not limited to these examples, and may be used for analyzing post-translational modifications of other proteins.

[0052] Example 1: Prediction of phosphorylated Tau protein 1 FIG. 3 is a schematic diagram showing the steps of analyzing phosphorylated Tau protein in Example 1 by the method of this embodiment.

[0053] As shown in FIG. 3, the method for analyzing phosphorylated Tau protein specifically includes a step of selecting Tau protein (100), a step of predicting a phosphorylated Tau protein sequence with a high probability of existence (200), a step of predicting a three-dimensional structure of phosphorylated Tau protein based on the phosphorylated Tau protein sequence with a high probability of existence (300), and a step of classifying and clustering phosphorylated Tau protein based on the three-dimensional structure (400).

[0054] 4A and 4B are schematic diagrams of the prediction step of predicting phosphorylated Tau protein sequences with high probability of existence in Example 1 and the prediction results, where FIG. 4A is a schematic diagram of the prediction step and FIG. 4B is a schematic diagram of the prediction results.

[0055] As shown in FIG. 4A, in the step of predicting the phosphorylated Tau protein sequence with high existence probability (200), first, all possible phosphorylation combinations of two or less serine and threonine phosphorylation sites are listed (210). Then, the existence probability of each phosphorylated protein is predicted by the DeepPhos algorithm (220). Among them, phosphorylated Tau proteins with existence probability of 90% or more are selected (230). Note that the above "two or less" is merely an example of phosphorylation combinations given to explain the method of this embodiment, and as mentioned above, for Tau protein, all phosphorylated protein combinations in which 2, 3, 4, 5, 6, 7, 8 and 9 serine and threonine residue sites are phosphorylated, respectively, can be considered to analyze the existence probability of each phosphorylated protein.

[0056] As shown in Figure 4B, all phosphorylation combinations in which one phosphorylation site and two phosphorylation sites in Tau protein may be phosphorylated are listed, and then phosphorylated Tau proteins with a probability of existence lower than a certain probability (e.g., 90%), such as pTau-205, are eliminated, and phosphorylated Tau proteins with a probability of existence of 90% or more, such as pTau-195, pTau-202, and pTau-202 / 205, are retained. Note that pTau-195, pTau-202, and pTau-202 / 205 respectively indicate a protein phosphorylated at position 195 of Tau protein, a protein phosphorylated at position 202 of Tau protein, and a protein phosphorylated at positions 202 and 205 of Tau protein.

[0057] 5A and 5B are schematic diagrams of the prediction step of predicting the three-dimensional structure of phosphorylated Tau protein in Example 1 and the prediction results, where FIG. 5A is a schematic diagram of the prediction step and FIG. 5B is a schematic diagram of the prediction results.

[0058] As shown in Figure 5A, in the step of predicting the three-dimensional structure of phosphorylated Tau protein (300), first, phosphorylation simulation is performed to replace phosphorylated serine and phosphorylated threonine with aspartic acid (D) and glutamic acid (E), respectively (310). Then, the three-dimensional structure of Tau protein is predicted using the RosseTTAFold or AlphaFold 2 algorithm (320).

[0059] As shown in Figure 5B, phosphorylated serine (pS) in the phosphorylated protein is replaced with aspartic acid (D), and phosphorylated threonine (pT) is replaced with glutamic acid (E), respectively, and then the three-dimensional structure of the protein after the replacement is predicted using the AlphaFold 2 algorithm.

[0060] To verify whether the three-dimensional structure predicted by the method of this embodiment is accurate, it is compared with the experimentally obtained three-dimensional structure provided in the RCSB PDB (Worldwide Protein Data Bank, wwPDB, www.rcsb.org).

[0061] FIG. 6 is a diagram showing a visual comparison between the experimental results of the three-dimensional structure of the Tau protein fragment in Example 1 (https: / / www.rcsb.org / structure / 5e2v) and the predicted results by AlphaFold 2 before and after amino acid substitution by the method of this embodiment.

[0062] As can be seen from FIG. 6, there is a significant difference between the predicted structures before and after the amino acid substitution, and the predicted structure after the substitution is closer to the experimental three-dimensional structure.

[0063] In addition, since the 3D viewer has coordinate plotting, in order to compare the three-dimensional structures more accurately, the structures of the protein fragments in the xy plane, xz plane, yz plane, and another random angle are selected and compared. The results show that at some angles, the predicted results after replacement are closer to the experimental results, and the difference between the predicted results before and after replacement is obvious, but at some angles, the three structures are not very similar. Overall, the predicted structure after replacement is closer to the experimental result than before replacement.

[0064] 7A and 7B are schematic diagrams of the clustering step for classifying and clustering the three-dimensional structures of phosphorylated Tau protein in Example 1 and the clustering results, where FIG. 7A is a schematic diagram of the clustering step and FIG. 7B is a schematic diagram of the clustering results.

[0065] As shown in Figure 7A, in the classification and clustering step (400), first, the three-dimensional structural features of phosphorylated Tau protein are extracted (410). Based on the predicted three-dimensional structural features of phosphorylated Tau protein, cluster analysis is performed on phosphorylated Tau protein, specifically, the phosphorylated Tau protein is classified and clustered according to the similarity of the structural features by the Hierarchical K-means algorithm (420). Then, a candidate biomarker list is generated with each category as a possible candidate biomarker (430).

[0066] As shown in Figure 7B, pTau-195 is structurally similar to pTau-202 / 205 and can be divided into one type, and pTau-202 is divided into one type. These are all clearly different from the structure of pre-phosphorylated Tau protein, and by performing structural difference analysis with pre-phosphorylated Tau protein of known structure, phosphorylated proteins with large structural differences among pTau-195, pTau-202 / 205, and pTau-202 are calculated and used as candidate biomarkers for the diagnosis or treatment of Alzheimer's disease.

[0067] Example 2: Prediction of phosphorylated proteins 2 To further verify the reliability of the method of this embodiment, 11 more phosphoprotein sequences (SEQ ID NOs: 1 to 11) were randomly selected from the RCSB PDB for further study. The phosphorylation information of the 11 phosphoprotein sequences is shown in Table 1.

[0068] [Table 1]

[0069] In Table 1, "SEP" stands for phosphorylated serine, "TPO" stands for phosphorylated threonine, and "ptm" stands for post-translational modifications, including phosphorylation, glycosylation, ubiquitination, nitrosation, methylation, acetylation, lipidation, and proteolysis.

[0070] The phosphorylated serine and phosphorylated threonine in the 11 phosphorylated protein sequences were replaced with aspartic acid and glutamic acid, respectively, and the three-dimensional structures of the amino acid sequences obtained after the replacement using AphaFold 2 were predicted (hereinafter also referred to as the "predicted results") and compared with the three-dimensional structures of the phosphorylated proteins obtained through experiments provided in the database (hereinafter also referred to as the "experimental results").

[0071] 8A to 8D are diagrams (tables) showing data on the comparison of predicted results and experimental results before and after amino acid substitution of 11 phosphorylated Tau protein sequences in Example 2. Here, "u" indicates the calculation result of the predicted structure after amino acid substitution, and "o" indicates the calculation result of the predicted structure before amino acid substitution, which are prediction results assuming that phosphorylated residues are not phosphorylated residues, for example, phosphorylated serine and phosphorylated threonine are serine and threonine as they are in the structural prediction results.

[0072] In AphaFold 2 prediction, five model results (rank 0~rank 4) are output for each sequence, and the experimental result is used as the gold standard. After alignment, RMSD (Root Mean Square Deviation) is calculated, and the smaller the RMSD value, the closer the predicted result is to the experimental result. "Min RMSD, u≦o" indicates whether the predicted result after substitution is closer to the experimental result than the predicted result before substitution when judged by the minimum RMSD, "1" indicates "Yes" and "0" indicates "No". Similarly, "Avg RMSD, u≦o" indicates whether the predicted result after substitution is closer to the experimental result than the predicted result before substitution when judged by the average RMSD, "1" indicates "Yes" and "0" indicates "No".

[0073] As can be seen from Figures 8A to 8D, when judging by the minimum RMSD, the predicted results after substitution for 8 of the 11 proteins are closer to the experimental results than the predicted results before substitution. When judging by the average RMSD, the predicted results after substitution for 7 of the 11 proteins are closer to the experimental results than the predicted results before substitution.

[0074] As can be seen from Figures 8A to 8D, the method of this embodiment is a feasible and relatively reliable prediction method for the three-dimensional structure of phosphorylated proteins, which can provide useful reference information for research on protein structure and function, etc., and can significantly reduce the amount of experimental work.

[0075] As described above, the analysis method of the present embodiment allows the analysis of the three-dimensional structure of phosphorylated proteins to be easily performed. For example, the present embodiment allows the three-dimensional structure of phosphorylated proteins to be predicted quickly and at low cost without relying on experiments, and further allows screening of phosphorylated proteins that are biomarkers or drug action targets based on the obtained three-dimensional structure.

[0076] (Information processing device) The analysis method according to the embodiment described above can also be executed by an information processing device 500 equipped with a program for executing the analysis method according to the embodiment. The information processing device 500 corresponds to, for example, a personal computer, a tablet terminal, a smartphone, a workstation, a PACS (Picture Archiving Communication System) viewer, etc. The information processing device 500 is also called, for example, an analysis device.

[0077] 9 is a diagram showing an example of an information processing device according to the present embodiment. The information processing device 500 is an information processing device equipped with a program for executing the analysis method according to the present embodiment. The information processing device 500 includes an input interface 501, a display 502, a memory circuit 510, and a processing circuit 520. The input interface 501, the display 502, the memory circuit 510, and the processing circuit 520 are connected to each other so as to be able to communicate with each other.

[0078] The input interface 501 is an input device for receiving various instructions and setting requests from an operator, such as a mouse, a keyboard, a touch panel, etc. The display 502 is a display device for displaying various information (images) and a GUI for the operator to input various setting requests using the input interface 501.

[0079] The storage circuit 510 is, for example, a NAND (Not AND) type flash memory or an HDD (Hard Disk Drive), and stores various information (images) and various programs for displaying GUIs, as well as information used by the programs.

[0080] The processing circuit 520 is an electronic device (processor) that controls the entire processing in the information processing device 500. The processing circuit 520 executes a reception function 521, a phosphorylated protein prediction function 522, a three-dimensional structure prediction function 523, a classification function 524, and an output control function 525. The reception function 521, the phosphorylated protein prediction function 522, the three-dimensional structure prediction function 523, the classification function 524, and the output control function 525 are recorded in the storage circuit 510 in the form of a program (analysis program) that can be executed by a computer (information processing device 500), for example. The processing circuit 520 reads out and executes each program to realize a function (the reception function 521, the phosphorylated protein prediction function 522, the three-dimensional structure prediction function 523, the classification function 524, and the output control function 525) corresponding to each program that has been read out.

[0081] The reception function 521 (reception unit) executes the target protein selection step (10) shown in Fig. 1. For example, the reception function 521 receives, via the input interface 501, information indicating the target protein designated by the operator.

[0082] The phosphorylated protein prediction function 522 (phosphorylated protein prediction unit) executes the phosphorylated protein sequence prediction step (20) shown in Fig. 1. For example, the phosphorylated protein prediction function 522 predicts, using a computational model for phosphorylated protein prediction, the amino acid sequence of a phosphorylated protein in which an amino acid site that can be phosphorylated with a predetermined probability or higher is phosphorylated, from among the amino acid sequences of a target protein designated by an operator.

[0083] The three-dimensional structure prediction function 523 (three-dimensional structure prediction unit) executes the prediction step (30) of the phosphorylated protein three-dimensional structure shown in Fig. 1. For example, the three-dimensional structure prediction function 523 predicts the three-dimensional structure of the phosphorylated protein by replacing phosphorylated serine and phosphorylated threonine in the amino acid sequence of the phosphorylated protein in which the target protein is phosphorylated with aspartic acid and glutamic acid, respectively, and inputting the amino acid sequence after the replacement into a calculation model for predicting the three-dimensional structure.

[0084] The classification function 524 (classification unit) executes the classification and clustering step (40) shown in Fig. 1. For example, the classification function 524 classifies phosphorylated proteins based on the three-dimensional structures obtained by the three-dimensional structure prediction step.

[0085] The output control function 525 (output control unit) controls the output of various information (images) generated in the information processing device 500. For example, the output control function 525 displays the classification result by the classification function 524 on the display 502 or stores it in the memory circuit 510. The output control function 525 may also transmit the classification result to an external device (information processing device). The external device that receives the classification result displays the received classification result on a specified display device or stores it in a specified memory circuit (memory, etc.).

[0086] Moreover, the information output by the output control function 525 is not limited to the classification result by the classification function 524. For example, the output control function 525 may output the prediction result by the phosphorylated protein prediction function 522 (amino acid sequence of phosphorylated protein) or the prediction result by the three-dimensional structure prediction function 523 (three-dimensional structure of phosphorylated protein).

[0087] That is, in the information processing device 500, the three-dimensional structure prediction function 523 predicts the three-dimensional structure of the phosphorylated protein by replacing phosphorylated serine and phosphorylated threonine in the amino acid sequence of the phosphorylated protein in which the target protein is phosphorylated with aspartic acid and glutamic acid, respectively, and inputting the amino acid sequence after replacement into a calculation model for predicting the three-dimensional structure. The classification function 524 classifies the phosphorylated protein based on the three-dimensional structure obtained by the three-dimensional structure prediction step. This allows the information processing device 500 to easily analyze the three-dimensional structure of the phosphorylated protein.

[0088] (program) Furthermore, this embodiment may be provided as a program for causing the information processing device 500 shown in Fig. 9 to execute each process. That is, the program (analysis program) according to this embodiment causes a computer to execute each process of replacing phosphorylated serine and phosphorylated threonine in the amino acid sequence of a phosphorylated protein in which a target protein is phosphorylated with aspartic acid and glutamic acid, respectively, predicting the three-dimensional structure of the phosphorylated protein by inputting the amino acid sequence after replacement into a calculation model for predicting a three-dimensional structure, and classifying the phosphorylated protein based on the three-dimensional structure obtained by the prediction process. In this way, the program can easily analyze the three-dimensional structure of a phosphorylated protein.

[0089] The present embodiment provides the technical solutions described in the following appendix.

[0090] (Appendix 1) a phosphorylated protein sequence prediction step of predicting the probability of existence of each phosphorylation combination of a target protein using a computational model, and selecting a phosphorylated protein having an existence probability equal to or higher than a predetermined probability from among the predicted combinations; a phosphorylated protein three-dimensional structure prediction step of replacing phosphorylated serine and phosphorylated threonine in the phosphorylated protein sequence obtained by the phosphorylated protein sequence prediction step with aspartic acid and glutamic acid, respectively, and predicting the three-dimensional structure of the protein after the replacement using a computational model; A computational analysis method for phosphorylated proteins, comprising: a classification / clustering step of classifying / clustering phosphorylated proteins based on the three-dimensional structures obtained in the phosphorylated protein three-dimensional structure prediction step.

[0091] (Appendix 2) The phosphorylated protein sequence prediction step includes: Enumerating phosphorylation combinations that include all possible phosphorylation sites; The analytical method described in Appendix 1, comprising: calculating and predicting the probability of the existence of a phosphorylated protein; and selecting a phosphorylated protein whose probability of existence is equal to or higher than a predetermined probability.

[0092] (Appendix 3) The analysis method according to claim 1 or 2, wherein the computational model used in the phosphorylated protein sequence prediction step is the DeepPhos algorithm.

[0093] (Appendix 4) The analysis method according to any one of appendix 1 to 3, wherein the calculation model used in the phosphorylated protein three-dimensional structure prediction step is the RosseTTAFold algorithm or the AlphaFold 2 algorithm.

[0094] (Appendix 5) The analysis method according to any one of appendix 1 to 3, wherein the calculation model used in the phosphorylated protein three-dimensional structure prediction step is a combination algorithm of the RosseTTAFold algorithm and the AlphaFold 2 algorithm.

[0095] (Appendix 6) The analytical method according to any one of Appendices 1 to 5, wherein the method for classifying and clustering the phosphorylated proteins in the classification and clustering step is a hierarchical clustering method.

[0096] (Appendix 7) The analytical method according to any one of appendixes 1 to 6, further comprising generating a list of candidate phosphorylated proteins by treating each category obtained in the classification / clustering step as a possible group of candidate phosphorylated proteins.

[0097] (Appendix 8) The analysis method according to any one of claims 1 to 7, wherein the predetermined probability is 90%.

[0098] (Appendix 9) A method for screening phosphorylated proteins as biomarkers, comprising comparing the three-dimensional structure of a candidate phosphorylated protein obtained by any one of the analytical methods described in any one of Appendices 1 to 7 with the three-dimensional structure of an unphosphorylated protein, and selecting a phosphorylated protein having a large structural difference as a biomarker.

[0099] (Appendix 10) A method for screening phosphorylated proteins that are targets of drug action, comprising comparing the three-dimensional structure of a candidate phosphorylated protein obtained by the analytical method described in any one of Appendices 1 to 7 with the three-dimensional structure of an unphosphorylated protein, and selecting a phosphorylated protein that has a large structural difference as a target of drug action. [Explanation of symbols]

[0100] 500 Information processing equipment 501 Input Interface 502 Display 510 Memory circuit 520 Processing Circuit 521 Reception Function 522 Phosphoprotein Prediction Function 523 Three-dimensional structure prediction function 524 Classification function 525 Output Control Function

Claims

1. a three-dimensional structure prediction step of predicting the three-dimensional structure of the phosphorylated protein by replacing phosphorylated serine and phosphorylated threonine with aspartic acid and glutamic acid, respectively, in the amino acid sequence of the phosphorylated protein obtained by phosphorylating the target protein, and inputting the amino acid sequence after the replacement into a calculation model for predicting the three-dimensional structure; a classification step of classifying the phosphorylated protein based on the three-dimensional structure obtained by the three-dimensional structure prediction step; , an analytical method.

2. The method further includes a phosphorylated protein prediction step of predicting, using a computational model for predicting phosphorylated proteins, an amino acid sequence of a phosphorylated protein in which an amino acid site that can be phosphorylated with a predetermined probability or higher is phosphorylated from the amino acid sequence of the target protein. The analytical method according to claim 1.

3. The phosphorylated protein prediction step outputs a group of candidates of the phosphorylated protein in which one or more phosphorylation sites are phosphorylated among a plurality of amino acid sites that can be phosphorylated with a predetermined probability or higher. The method of analysis according to claim 2.

4. The computational model for predicting phosphorylated proteins is the DeepPhos algorithm; The method of analysis according to claim 2.

5. The computational model for predicting three-dimensional structure is the Rosse TTAFold algorithm or the AlphaFold 2 algorithm; The analytical method according to claim 1.

6. The computational model for predicting the three-dimensional structure is a combination of the Rosse TTAFold algorithm and the AlphaFold 2 algorithm; The analytical method according to claim 1.

7. The classifying step classifies the phosphorylated proteins by a hierarchical clustering method. The analytical method according to claim 1.

8. The classification step classifies the candidate group of phosphorylated proteins based on the three-dimensional structural features. The analytical method according to claim 3.

9. The predetermined probability is 90%. The method of analysis according to claim 2.

10. The classification step includes comparing the three-dimensional structure of the phosphorylated protein with the three-dimensional structure of a non-phosphorylated target protein, and identifying phosphorylated proteins having a large structural difference. The analytical method according to claim 1.

11. The classification step identifies phosphorylated proteins with large structural differences as biomarkers or targets for drug action. The analytical method according to claim 10.

12. a three-dimensional structure prediction unit that predicts the three-dimensional structure of the phosphorylated protein by replacing phosphorylated serine and phosphorylated threonine with aspartic acid and glutamic acid, respectively, in the amino acid sequence of a phosphorylated protein obtained by phosphorylating a target protein, and inputting the amino acid sequence after the replacement into a calculation model for predicting a three-dimensional structure; a classification unit that classifies the phosphorylated proteins based on the three-dimensional structure obtained by the three-dimensional structure prediction unit; An information processing device comprising:

13. a phosphorylated serine and a phosphorylated threonine in an amino acid sequence of a phosphorylated protein obtained by phosphorylating a target protein are replaced with aspartic acid and glutamic acid, respectively, and the amino acid sequence after the replacement is input into a computational model for predicting a three-dimensional structure, thereby predicting a three-dimensional structure of the phosphorylated protein; Classifying the phosphorylated proteins based on the three-dimensional structures obtained by the prediction process. A program that causes a computer to execute each process.

Images