Method for visualizing trained protein solubility prediction model

By applying formal concept analysis to a trained protein solubility prediction model, the method identifies key attributes for enhancing protein solubility, allowing insoluble proteins to be synthesized as soluble proteins, thus improving production efficiency.

JP2024165114AInactive Publication Date: 2024-11-28松井 大亮 +1
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
JP2023080992
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-16
Publication Date
2024-11-28
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Existing methods for producing enzymes through genetic modification often result in protein precipitation, limiting the production of desired enzymes, and current solubility prediction models have low accuracy, making it difficult to reliably improve protein solubility.

Method used

A method using formal concept analysis to visualize the implications of a trained protein solubility prediction model, identifying protein attributes related to solubility by combining machine learning with domain knowledge, and modifying proteins based on these attributes to enhance solubility.

Benefits of technology

Enables the design of insoluble proteins to be synthesized as soluble proteins, improving production yields by identifying key attributes that affect solubility and applying targeted modifications.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024165114000026
    Figure 2024165114000026
  • Figure 2024165114000027
    Figure 2024165114000027
  • Figure 2024165114000028
    Figure 2024165114000028
Patent Text Reader

Abstract

To provide a method for identifying attributes of protein involved in protein solubility, using visualization of a trained protein solubility prediction model through formal concept analysis.SOLUTION: A method for identifying attributes of protein involved in protein solubility, comprises the steps of: setting attributes of protein; calculating attributes related to a degree of solubility based on output data obtained from input data by a trained protein solubility prediction model; and extracting rules that fit a premise: the attribute of the protein, a confidence level: 60% or more, and a conclusion: the attributes related to the degree of solubility.SELECTED DRAWING: None
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a method for identifying protein attributes that contribute to protein solubility using visualization by formal concept analysis of a trained protein solubility prediction model. The present invention also relates to a method for producing a protein with improved solubility, comprising a step of modifying the protein so that it has the attribute identified by the identification method. Furthermore, the present invention relates to a method for improving protein solubility using the attribute identified by the identification method. [Background technology]

[0002] Mass production technology is necessary for the practical use of enzymes, but when enzymes from other microorganisms, animals, and plants are produced by genetic recombination, the proteins often precipitate, making it impossible to produce the desired enzyme. Methods that have been attempted to avoid protein precipitation include (i) expression as a fusion protein, (ii) examination of culture conditions, (iii) examination of other expression systems, (iv) rewinding, and (v) production of highly expressed mutant enzymes, but a highly versatile method suitable for industrial use has not yet been established (Non-Patent Document 1).

[0003] (v) For the production of highly expressed mutant enzymes, the α-helix method and the HiSol method are known. However, the former is limited to proteins having an α-helix structure, and the latter is limited to enzymes with many highly homologous protein sequences, and therefore these are not considered to be highly versatile techniques (Non-Patent Document 2).

[0004] The inventors have discovered a method for efficiently producing a target enzyme by replacing a part of the amino acid sequence of an enzyme that precipitates and becomes insoluble with another amino acid, and have also developed a computer program that can predict the site for amino acid substitution (Patent Document 1).

[0005] However, in the conventional method, the accuracy of the prediction model for protein solubilization is about 50%, and therefore even if the prediction result output by the model is used as it is, the protein is not necessarily solubilized. Therefore, a technology with higher predictive power for protein solubilization has been desired. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] International Publication No. 2016 / 199898 [Non-patent literature]

[0007] [Non-Patent Document 1] Y. Asano, et al., Protein Engineering, Design and Selection, vol.24(8), pp. 607-616, 2011. [Non-Patent Document 2] D. Matsui, et al., Sci. Rep. 7, 9558, 2017. Summary of the Invention [Problem to be solved by the invention]

[0008] Therefore, an object of the present invention is to provide a method for identifying a protein attribute involved in protein solubility using visualization by formal concept analysis of a trained protein solubility prediction model. Another object of the present invention is to provide a method for producing a protein with improved solubility, which includes a step of modifying a protein so that the protein has the attribute identified by the above-mentioned identification method. [Means for solving the problem]

[0009] The present inventors have constructed a model that predicts protein solubility using machine learning and have achieved predictions with a certain degree of accuracy. However, since it is difficult to grasp the trends of learning obtained with machine learning (the so-called black box problem), the solubilization trends obtained by the prediction model are unknown, and the solubilization mechanism has not been elucidated. Therefore, the present inventors came up with the idea of ​​visualizing the trained protein solubility prediction model by formal concept analysis (a method for comprehensively extracting implication relationships between the features of a learning device) in order to search for and identify the attributes (features) of proteins involved in protein solubility.

[0010] Based on the above idea, as a result of intensive research, we focused on the input and output of the trained machine learning machine described above, and found a method of visualizing the relationship between the prediction result and the domain knowledge in the framework of the implication relationship between the features of the learning machine or the association rule by providing the domain knowledge of the prediction problem as an attribute of the formal concept analysis in addition to the set of various input and output results. As a result of further research based on these findings, we have completed the present invention (a method for identifying attributes of a protein involved in the solubility of a protein) characterized by applying formal concept analysis to a trained machine learning machine in which both the input and output of the machine learning machine are attributes. We have also completed the present invention (a method for producing a protein with improved solubility) characterized in that a desired protein has an attribute identified by the method for identifying attributes of a protein involved in the solubility of a protein of the present invention.

[0011] That is, the present invention is as follows. [1] A method for identifying a protein attribute that is involved in protein solubility, comprising: setting attributes of the protein; Calculating attributes related to the degree of solubility based on output data obtained from the input data by the trained protein solubility prediction model; Preamble: the attributes of said protein; Confidence level: 60% or higher, and Conclusion: Attributes related to the degree of solubility Extracting rules that fit the A method comprising: [2] The method according to [1], wherein the input data is a dataset of an amino acid sequence of a wild-type protein and a sequence of the wild-type protein in which one to several amino acids have been mutated. [3] The method according to [1] or [2], wherein the attribute of the protein is one or more (e.g., 1, 2, 3, 4, 5, 6, 7, 8) selected from the group consisting of the amino acid before the mutation, the properties of the amino acid before the mutation, the classification of the amino acid before the mutation, the amino acid after the mutation, the properties of the amino acid after the mutation, the classification of the amino acid after the mutation, the position of the amino acid mutation, and the secondary structure of the position of the amino acid mutation. [4] The method according to [3], wherein the amino acid has hydrophilic or hydrophobic properties. [5] The method according to [3] or [4], wherein the classification of the amino acids is aliphatic, negatively charged, positively charged, aromatic, or uncharged. [6] The method according to any one of [3] to [5], wherein the secondary structure at the position where the amino acid mutation is introduced is an α-helix, a β-sheet, or a coil. [7] The method according to any one of [1] to [6], wherein the attribute relating to the degree of solubility is based on the degree of solubility of a wild-type protein and the change in the degree of solubility of a protein in which one to several amino acids of the wild-type protein have been mutated. [8] A method for producing a protein with improved solubility, comprising the step of modifying the protein so that it has an attribute identified by the method described in any one of [1] to [7]. [9] A method for producing a protein with improved solubility, comprising: A step of substituting at least one amino acid selected from the group consisting of arginine, tyrosine, and glutamine among amino acids constituting a protein with an amino acid other than arginine, tyrosine, and glutamine; A step of substituting amino acids other than glycine, valine, threonine, cysteine, and lysine constituting a protein with glycine, valine, threonine, cysteine, or lysine; and Substituting aromatic amino acids in a disordered region of a protein with glycine, valine, or threonine; The method comprises one or more steps selected from the group consisting of:

[10] The method according to [9], wherein the amino acids constituting the protein are amino acids in the latter half of the amino acid sequence of the protein.

[11] The method according to [9] or

[10] , wherein aromatic amino acids in the disordered region of the protein are hydrophilic.

[12] A method for improving the solubility of a protein, comprising: A step of substituting at least one amino acid selected from the group consisting of arginine, tyrosine, and glutamine among amino acids constituting a protein with an amino acid other than arginine, tyrosine, and glutamine; A step of substituting amino acids other than glycine, valine, threonine, cysteine, and lysine constituting a protein with glycine, valine, threonine, cysteine, or lysine; and Substituting aromatic amino acids in a disordered region of a protein with glycine, valine, or threonine; A method comprising one or more steps (e.g., steps 1, 2, 3) selected from the group consisting of: Effect of the Invention

[0012] According to the present invention, since the attributes (characteristics) of a protein that are involved in the solubility of the protein can be identified, it becomes possible to design insoluble or soluble recombinant proteins, etc., based on the identified attributes. In particular, for a protein that forms a precipitate, it is possible to synthesize the protein as a soluble protein without forming a precipitate based on the identified attributes, thereby improving the production amount of the protein. [Brief description of the drawings]

[0013] [Figure 1] Figure 1 shows an outline of a concept in formal concept analysis, which is a pair of denotation and intension. [Diagram 2]Figure 2 shows the results of predicting the degree of solubility of the test data using a regression model trained on the training data. [Diagram 3] FIG. 3 shows a visualization of the regression model summarizing amino acid to amino acid mutations that result in increased solubility. [Figure 4] FIG. 4 shows a visualization of the regression model summarizing amino acid to amino acid mutations that result in decreased solubility. [Diagram 5] FIG. 5 shows a visualization of the regression model summarizing amino acid-to-amino acid changes in coiled secondary structures with increasing solubility. [Figure 6] FIG. 6 shows a visualization of the regression model summarizing amino acid-to-amino acid changes in the β-sheet secondary structure that increase solubility. [Figure 7] FIG. 7 shows a visualization of the regression model summarizing the amino acid-to-amino acid changes in α-helices of secondary structure that increase the degree of solubility. [Figure 8] FIG. 8 shows a visualization of the regression model summarizing amino acid-to-amino acid changes in coiled secondary structure that result in decreasing solubility. [Figure 9] FIG. 9 shows a visualization of the regression model summarizing amino acid-to-amino acid changes in β-sheet secondary structure that result in decreased solubility. [Figure 10] FIG. 10 shows a visualization of the regression model summarizing amino acid-to-amino acid changes in α-helices of secondary structure as the degree of solubility decreases. [Figure 11] FIG. 11 shows a visualization of the regression model, which summarizes the amino acid mutations from the amino acid at position-1 when the degree of solubility increases. [Figure 12]FIG. 12 shows a visualization of the regression model, which summarizes the amino acid mutations from the amino acid at position-2 when the degree of solubility increases. [Figure 13] FIG. 13 shows a visualization of the regression model, in which the amino acid mutation positions at which the degree of solubility increases are summarized for amino acid mutations starting from the amino acid at position-3. [Figure 14] FIG. 14 shows a visualization of the regression model, which summarizes the amino acid mutations from the amino acid at position-4 when the degree of solubility increases. [Figure 15] FIG. 15 shows a visualization of the regression model, which summarizes amino acid mutations from the amino acid at position-1 when the degree of solubility decreases. [Figure 16] FIG. 16 shows a visualization of the regression model, which summarizes amino acid mutations from the amino acid at position-2 when the degree of solubility decreases. [Figure 17] FIG. 17 shows a visualization of the regression model, which summarizes the amino acid mutations at position-3 and subsequent amino acid mutations that result in a decrease in solubility. [Figure 18] FIG. 18 shows a visualization of the regression model, in which the amino acid mutation positions at which the degree of solubility decreases are summarized for amino acid mutations starting from the amino acid at position-4. [Figure 19] FIG. 19 shows the relationship between multiple attributes and changes in solubility when the amino acid before mutation is Y. [Figure 20] FIG. 20 shows a summary of amino acid-to-amino acid mutations that increase the probability of soluble expression in a visualization of the classification model. [Figure 21] FIG. 21 shows a summary of amino acid-to-amino acid mutations that decrease the probability of soluble expression in a visualization of the classification model. [Figure 22]FIG. 22 shows a summary of amino acid-to-amino acid mutations in the disorder region in Disorder-1 when the probability of soluble expression increases in the visualization of the classification model. [Diagram 23] FIG. 23 shows a summary of amino acid-to-amino acid mutations in the disorder region of Disorder-5 when the probability of soluble expression increases in the visualization of the classification model. [Figure 24] FIG. 24 shows a summary of amino acid-to-amino acid mutations in the Disorder region in Disorder-1 when the soluble expression probability decreases in the visualization of the classification model. [Diagram 25] FIG. 25 shows a summary of amino acid-to-amino acid mutations in the disorder region of Disorder-5 when the probability of soluble expression is reduced in the visualization of the classification model. [Figure 26] FIG. 26 shows a visualization of the classification model that summarizes amino acid-to-amino acid mutations in the disorder region of Disorder-5 when the soluble expression probability decreases. [Figure 27] FIG. 27 shows a summary of amino acid mutations from the amino acid at position-2 when the probability of soluble expression increases in the visualization of the classification model. [Figure 28] FIG. 28 shows a summary of amino acid mutations from the amino acid at position-3 when the probability of soluble expression increases in the visualization of the classification model. [Figure 29] FIG. 29 shows a summary of amino acid mutations from the amino acid at position-4 when the probability of soluble expression increases in the visualization of the classification model. [Diagram 30] FIG. 30 shows a summary of amino acid mutations at amino acid positions starting from position-1 when the probability of soluble expression is reduced in the visualization of the classification model. [Diagram 31]FIG. 31 shows a summary of amino acid mutations from the amino acid at position-2 when the probability of soluble expression is reduced in the visualization of the classification model. [Diagram 32] FIG. 32 shows a summary of amino acid mutations at positions where the probability of soluble expression decreases, starting from the amino acid at position-3, in the visualization of the classification model. [Diagram 33] FIG. 33 shows a visualization of the classification model, which summarizes amino acid mutations that reduce the probability of soluble expression and whose amino acid mutation position is position-4. [Diagram 34] FIG. 34 shows the relationship between multiple attributes and the change in soluble expression probability when the amino acid before mutation is Y. [Diagram 35] FIG. 35 shows the relationship between multiple attributes and changes in solubility when the amino acid before mutation is W. [Diagram 36] FIG. 36 shows the relationship between multiple attributes and changes in solubility when the amino acid before mutation is F. [Figure 37] FIG. 37 shows the relationship between multiple attributes and changes in solubility when the amino acid after mutation is D. [Figure 38] FIG. 38 shows the relationship between multiple attributes and changes in solubility when the amino acid after mutation is E. [Figure 39] FIG. 39 shows the relationship between multiple attributes and changes in solubility when the amino acid after mutation is K. [Diagram 40] FIG. 40 shows the relationship between multiple attributes and the change in the probability of soluble expression when the amino acid before mutation is Q. [Diagram 41] FIG. 41 shows the relationship between multiple attributes and the change in the probability of soluble expression when the amino acid after mutation is G. [Diagram 42] FIG. 42 shows the relationship between multiple attributes and the change in the probability of soluble expression when the amino acid after mutation is V. [Diagram 43] FIG. 43 shows the relationship between multiple attributes and the change in the probability of soluble expression when the amino acid after mutation is T. [Diagram 44] Figure 44 shows where in the sequence the sites affected by amino acid substitutions are located. In Figure 44, the horizontal axis shows the number, and the vertical axis shows the position, with the N-terminus at the bottom and the C-terminus at the top. [Diagram 45] Figure 45 shows, at the top, amino acids before substitution that increase solubility, and, at the bottom, amino acids after substitution that similarly decrease solubility. [Figure 46] FIG. 46 shows that substitution of aromatic amino acids in the disorder region with G tends to increase soluble expression. [Figure 47] FIG. 47 shows that replacement of aromatic amino acids in the disorder region with T tends to increase soluble expression. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] 1. Identification method of the present invention The present invention provides a method for identifying protein attributes that are involved in protein solubility using visualization of a trained protein solubility prediction model by formal concept analysis. More specifically, the present invention provides a method for identifying protein attributes that are involved in protein solubility, comprising: setting attributes of the protein; Calculating attributes related to the degree of solubility based on output data obtained from the input data by the trained protein solubility prediction model; Preamble: the attributes of said protein; Confidence level: 60% or higher, and Conclusion: Attributes related to the degree of solubility Extracting rules that fit the The present invention provides a method comprising:

[0015] The identification method of the present invention is constructed by a technique for predicting mutations and visualizing the solubilization mechanism by combining machine learning (a trained machine learning model) and formal concept analysis. Specifically, the following steps (C) are performed. (C) Formal concept analysis is applied to the machine learning machine that has completed training, with both the input and output of the machine learning machine as attributes. In other words, both the soluble factors and the prediction results are set as attributes. This allows the factors that contribute to solubility to be output in the form of a rule, such as "if it has attribute A, it must also have attribute B."

[0016] Moreover, the following step (D) may be carried out. (D) A biochemistry expert examines the rules obtained in (C), extracts rules that match his / her own knowledge, selects mutations that match the rules and have high prediction values ​​based on machine learning, and verifies them through biochemical experiments.

[0017] Furthermore, in preparing a trained machine learning model, for example, the following steps may be performed. (A) Protein primary structure (amino acid sequence) data, structural information data, and biochemical attributes of amino acids are compiled in a database. These are formalized using formal concept analysis to organize attributes that may affect solubility as logical expressions. This allows the attributes that are factors in solubility to be extracted. (B) Using machine learning with the extracted attributes as feature input, solubility prediction is realized for each mutation site.

[0018] Machine learning models In the present invention, the type of the trained protein solubility prediction model is not particularly limited as long as it can be used for the formal concept analysis described later. For example, a neural network, more specifically, deep learning having a multi-layered neural network, can be used. The method for constructing the model is also not particularly limited, and may be performed by a method known per se. In this specification, the term "neural network" refers to a model used in machine learning, which is composed of artificial neurons (nodes) that form a network through synaptic connections and has a problem-solving ability. A neural network is defined by the connection pattern between neurons in other layers, the learning process for updating model parameters, the activation function for generating output values, and the like.

[0019] A neural network includes an input layer, an output layer, and optionally one or more intermediate layers (also called "hidden layers"). Each layer includes one or more neurons, and a neural network includes synapses that connect neurons. In a neural network, each neuron can output a function value of an activation function for an input signal, weight, and bias input through a synapse. Examples of such activation functions include a ReLU function, a Tanh function, a sigmoid function, a step function, an identity function, a softmax function, and derivatives thereof. A machine learning model includes model parameters that are optimized through learning, weights of synaptic connections, and biases of neurons. In addition, when training a machine learning model, it is necessary to set hyperparameters before learning. Examples of such hyperparameters include a learning rate, a number of iterations, and a size of a mini-batch.

[0020] Model information (program or data structure) that defines a machine learning model includes, for example, connection information on how units included in the input layer, one or more intermediate layers, and output layer that constitute each neural network are connected to each other, and various information such as weights and biases that are assigned to data input and output between the connected units. Examples of the connection information include the number of units included in each layer, information specifying the type of unit to which each unit is connected, activation functions that implement each unit, and information such as gates provided between neurons in the intermediate layer. The gates selectively pass or weight data transmitted between units depending on, for example, the value (e.g., 1 or 0) returned by the activation function.

[0021] Learning of a machine learning model using training data typically involves determining a loss function for the training data and searching for parameters (optimum values ​​of parameters) that minimize the loss function, and calculating the loss function for each training data and using the sum as an index. As a loss function, a squared sum error is typically used, but other loss functions such as mean absolute error and mean squared logarithmic error can also be used. In addition, the search for the optimal parameter value may be performed at high speed by approximating the loss function using surrogate modeling based on the Bayesian optimization method. By increasing the speed, it becomes possible to perform searches using a genetic algorithm or gradient method tens of thousands or millions of times, which can further improve accuracy. Furthermore, mini-batch learning or online learning can also be used to avoid stagnation in learning.

[0022] Examples of machine learning models using neural networks include regression models and classification models. The difference between the two is whether the predicted value of the network is a continuous value or a discrete value. Regression looks at the tendency of the features of the data and outputs a predicted value as a continuous value. Classification has a finite number of predetermined classes, and classifies the data into which class it falls. Each class corresponds to a different value, and classification is performed according to the value, so the predicted value of the network is a discrete value. The trained protein solubility prediction model used in the identification method of the present invention may be, for example, both a regression model that predicts the solubility degree of a protein and a classification model that classifies an input protein as soluble or insoluble, or it may be either one of them.

[0023] Examples of the neural network include a recurrent neural network (RNN) and a convolutional neural network (CNN), with the RNN being preferred. The RNN is a neural network that handles time-series data such as voice, text, and video. The RNN is good at appropriately handling data that has characteristics based on the arrangement of elements in a time series, and learning patterns and properties of the data. One of the characteristics of the RNN is that it has a looping path inside the network. A standard neural network cannot process the order of data, but the RNN has a looping path, so it can store past information by constantly circulating data information inside the network. Then, the information is updated based on that memory. In the present invention, protein structural information may be treated as time-series data, and a model that has learned the characteristics and patterns of the arrangement of protein structures using the characteristics of the RNN may be used.

[0024] In theory, a recurrent neural network takes all past inputs into account in the output, but in reality, only some past data is taken into account. RNNs circulate information in a looping circuit, but if the information is circulated for a long period of time, it may lose its value. Therefore, it becomes difficult for RNNs to store information over a long period of time, and there is a risk that past information cannot be considered. To solve this problem, gates and memory cells may be added to the RNN. This is called long short-term memory (LSTM). The trained protein solubility prediction model used in the present invention may be trained using LSTM.

[0025] For the trained protein solubility prediction model used in the present invention, all of the data used for training may be used for training the model, or a portion of the data may be used for training and the rest may be used as test data for evaluating the versatility of the trained model. Methods for dividing the data into training data and test data include, for example, the hold-out method, the cross-validation method, and the leave-one-out method.

[0026] The training data for constructing a trained protein solubility prediction model is not particularly limited as long as the desired prediction model is constructed. For example, the protein data to be trained may be the "Solubility database of all Escherichia coli proteins (eSOL)" (eSOL database developed in the Targeted Proteins Research Project, http: / / www.tanpaku.org / tp-esol / index.php?lang=en (accessed May 26, 2022)) or a data collection owned by the user.

[0027] In addition, for the trained protein solubility prediction model used in the present invention, the input and output data are not particularly limited as long as the prediction model can be applied to the formal concept analysis described later, and examples of the input and output data include information on the primary, secondary, or tertiary structure of a protein, and the output data includes information on the degree of solubility of the protein. More specifically, for example, the input data is a data set of the amino acid sequence of a wild-type protein and a sequence in which one to several (e.g., 1 to 9, 1 to 5, 1 to 3, 1 or 2, 1) amino acids of the wild-type protein are mutated, and the output data is the degree of solubility (%) of the protein and a classification of whether the protein is soluble or insoluble.

[0028] ·Formal concept analysis In this specification, formal concept analysis is a data analysis method that analyzes conceptual structures based on mathematical foundations. Formal concept analysis treats mathematically defined conceptual data as units of thought, and clarifies conceptual structures, analyzes phenomena, visualizes data, and clarifies data dependencies. The following are definitions of terms used in formal concept analysis.

[0029] (i) Object G An object that appears in a phenomenon or event (ii) Attribute M The properties of an object (iii) Context K Relation between objects and attributes K = (G, M, I),I ⊆ G × M (iv) Extension A A set of objects with a common set of attributes A ⊆ G (v) Intended B A set of attributes that the extension sets have in common B ⊆ M (vi) Concept (A, B) Denotation and Intent Pairs (vii) Concept lattice B(G, A, B) Complete bundle of concepts

[0030] In formal concept analysis, concepts that are pairs of extension and intension are extracted by applying formal concept analysis to a context, and the extracted concepts become components of a concept lattice (Figure 1). In addition, inclusion relationships between attributes such as entailment relationships and association rules can be extracted based on the extracted concept lattice. Formal concept analysis can be performed using a method known per se, for example, a formal concept analysis method using The Concept Explorer (Release 1.3) (https: / / conexp.sourceforge.net / ). The Concept Explorer (Release 1.3) can automatically generate concept lattices, entailment relationships, and association rules by inputting a context table. A context table refers to a table that summarizes contexts.

[0031] Concept Lattice In a concept lattice, concepts that are pairs of extension and intensification become nodes, and concepts that have an implication relationship are connected by arcs. Attributes are added toward the bottom of the concept lattice, and objects are added toward the top of the concept lattice. Therefore, the higher the node is located, the more objects are included in the concept's extension and the fewer attributes are included in its intensification.

[0032] Association Rules In a context table, for certain attributes A and B, a rule that "an object having attribute A has attribute B" may be established. This is called an association rule. In this association rule, attribute A is the antecedent, attribute B is the conclusion, and the percentage of objects having all the attributes of the conclusion among objects having all the attributes of the antecedent is called the reliability (%). In particular, an association rule with a reliability of 100% is called an implication relationship. It is necessary to determine the reliability of the implication relationship and association rule to be extracted. In the present invention, based on general knowledge in biology, the analysis focuses on association rules and implication relationships with a reliability of 60% or more. The reliability may be set, for example, to 70% or more, 75% or more, 80% or more, 85% or more, 90% or more, or 95% or more.

[0033] How to apply formal concept analysis An example of a method for applying formal concept analysis is to create a context table with attributes such as the properties of amino acids, the classification of amino acids, and the molecular weight of amino acids, and to generate a concept lattice and association rules.

[0034] · Creating a context table In the context table, attributes are represented by two values, so each attribute needs to be binarized. Specifically, it is as follows.

[0035] The properties of amino acids are classified into attributes based on, for example, Table 1, whether they are hydrophilic or hydrophobic.

[0036] [Table 1]

[0037] There are a number of standards for classifying amino acids, and the attributes may be determined by referring to the classification of amino acids into five categories, non-polar aliphatic, negatively charged, positively charged, aromatic, and polar uncharged, as set forth in Principles of biochemistry: with an extended discussion of oxygen-binding proteins / Albert L. Lehninger, David L. Nelson, Michael M. Cox. -- 2nd ed., [enl. with suppl.]. -- Worth Publishers, 1993. The classification of amino acids according to Lehninger is shown in Table 2. Based on Table 2, the amino acid is classified into each attribute depending on whether its nature is aliphatic, negatively charged, positively charged, aromatic, or uncharged.

[0038] [Table 2]

[0039] Molecular weight of amino acids Each attribute is classified according to whether the molecular weight of the amino acid is one of the four: less than 120 Da (mw-1), between 120 Da and 140 Da (mw-2), between 140 Da and 160 Da (mw-3), and over 160 Da (mw-4). Based on the above attributes, a context table is created with the features of the 20 amino acids classified into objects (A, C, D, E, F, G, H, I, K, L, M, N, P, Q, R, S, T, V, W, Y) into a total of seven types (hydrophilic, hydrophobic, aliphatic, negatively charged, positively charged, aromatic, uncharged, mw-1, mw-2, mw-3, mw-4) as attributes. A concept lattice is generated from the created context table. In addition, in the case of a small-scale context, the output concept lattice is not very complex, but as the number of concepts increases, the concept lattice becomes more complex, so it may be difficult to read the concept structure from the concept lattice. Therefore, in this case, the analysis may be performed using association rules rather than the concept lattice.

[0040] Association Rules To explain the association rules, Table 3 shows the association rules obtained by running formal concept analysis that have a confidence level of 60% or more and have 5 or more objects with the antecedent attribute. For example, the rule in the first row of Table 3 means that "80% of amino acids with a molecular weight of 160 Da or more also have the hydrophilic attribute." Table 4 shows the implication relationships among the relationships obtained from the concept lattice described above.

[0041] [Table 3]

[0042] [Table 4]

[0043] Visualization of regression models Regarding the specific procedure of the identification method of the present invention, in one embodiment, the solubility of a protein in which a mutation has been introduced comprehensively into a protein described in eSOL is predicted using a trained protein solubility prediction model (regression model), and the predicted result is subjected to formal concept analysis to identify the protein attribute involved in the solubility of the protein. In the present invention, a mutation is introduced comprehensively into a portion of the amino acid sequence of the protein, and all amino acid sequences are predicted using the above-mentioned regression model. As a rule of mutation, 19 patterns of mutant amino acid sequences are created by mutating a portion of the base amino acid sequence into 19 types excluding the original amino acid, and the solubility of the mutant amino acid sequence is predicted using the regression model. However, since the first character of the amino acid sequence is basically the amino acid M, typically, the first character is not mutated. The protein into which a mutation is introduced comprehensively may be a wild-type protein. In addition, the number of mutations to be introduced is 1 to several (2, 3, 4, 5, 6, 7, 8, 9, etc.), more specifically, 1 to 5, 1 to 3, 1 or 2, or 1. In this specification, data collected about the prediction results may be referred to as mutation prediction data. The mutation prediction data may include the protein clone number (JW ID), the amino acid before the mutation (Before), the position of the amino acid at which the mutation is introduced (Position), the amino acid after the mutation (After), the predicted value of the solubility degree (Predict_sol), and the change in the solubility degree (dif) (attribute related to the solubility degree). The change in the solubility degree (attribute related to the solubility degree) may be calculated as the difference between the solubility degree of the protein predicted before the mutation and the solubility degree of the protein predicted after the mutation.

[0044] Attributes used in formal concept analysis The attributes set in the step of setting the protein attributes in the identification method of the present invention are not particularly limited as long as they can be used for input and output of a machine learning machine (protein solubility prediction model), and for example, the attributes may be extracted and set by the method described above. Specifically, (A) protein primary structure (amino acid sequence) data, structural information data, and biochemical attributes of amino acids, etc. are collected in a database. By formalizing them by formal concept analysis, attributes that are likely to affect solubility are organized as a logical formula. In this way, attributes that are factors in solubilization are extracted. Specific examples of the attributes include the amount of change in solubility degree, amino acids before introduction of mutation, amino acids after introduction of mutation, properties of amino acids before introduction of mutation, properties of amino acids after introduction of mutation, classification of amino acids before introduction of mutation, classification of amino acids after introduction of mutation, position of amino acid mutation, and secondary structure of position of amino acid mutation. In order to create a context table to be used in formal concept analysis, attributes need to be given as binary values. Below, a method of binarizing each attribute will be explained.

[0045] Change in solubility The change in the solubility level is classified into one of two attributes: an increase in the solubility level (Pre-sol-up) or a decrease in the solubility level (Pre-sol-down). If the value of the change in the solubility level of the mutation prediction data is, for example, greater than 0.055, it is classified as an increase in the solubility level, and if the value is less than -0.055, it is classified as a decrease in the solubility level. The value of 0.055 is the minimum value that a computer can tolerate in performing formal concept analysis, so it can be used preferably, but is not limited to this value and can be set appropriately between values ​​of -1 and 1, for example. If an attempt is made to analyze all mutations using formal concept analysis, a huge load is placed on the computer, and the computer may not be able to perform the formal concept analysis. Therefore, a minimum threshold value may be set for the change in the solubility level to reduce the number of protein mutations to a range that can be processed by the computer.

[0046] Amino acid before mutation The attribute is determined based on the amino acid before the mutation. For example, if the amino acid before the mutation is A (alanine), the attribute is "Amino acid before the mutation is A (Before-A)".

[0047] Amino acids after mutation The attribute is determined by the amino acid after mutation. For example, if the amino acid after mutation is A, the attribute is "Amino acid after mutation is A (After-A)".

[0048] -Properties of amino acids before mutation For example, amino acids before the introduction of mutation are classified into either hydrophilic or hydrophobic attributes according to Table 1. For example, if an amino acid before the introduction of mutation is hydrophilic, the attribute is "amino acid before the introduction of mutation is hydrophilic (Before-hydrophilic)".

[0049] -Properties of amino acids after mutation For example, the amino acid after the introduction of the mutation is classified into either hydrophilic or hydrophobic attribute according to Table 1. For example, if the amino acid after the introduction of the mutation is hydrophilic, the attribute is "amino acid after the mutation is hydrophilic (After-hydrophilic)".

[0050] Classification of amino acids before mutation For example, according to Table 2, the amino acid before the introduction of the mutation is classified into one of the five types: aliphatic, negatively charged, positively charged, aromatic, and uncharged. For example, if the amino acid before the introduction of the mutation is hydrophilic, the attribute is "amino acid before the introduction of the mutation is hydrophilic (Before-hydrophilic)".

[0051] Classification of amino acids after mutation For example, the amino acid after the mutation is classified into one of the five types, aliphatic, negatively charged, positively charged, aromatic, and uncharged, according to Table 2. For example, if the amino acid after the mutation is hydrophilic, the attribute is "After-hydrophilic amino acid."

[0052] -Amino acid mutation position For example, the position of an amino acid mutation is calculated as a relative mutation position p for the amino acid sequence using the following formula 1, and classified into each attribute depending on which of the following four ranges of p is satisfied: 0 or more and less than 0.25 (position-1), 0.25 or more and less than 0.5 (position-2), 0.5 or more and less than 0.75 (position-3), or 0.75 or more and less than 1 (position-4). In the following formula 1, the "maximum length of the amino acid sequence of the protein" may refer to the full-length amino acid sequence of the protein for which it is desired to calculate the mutation introduction position p. In this specification, the region from position-1 to position-2 is also referred to as the "first half of the amino acid sequence," and the region from position-3 to position-4 is also referred to as the "second half of the amino acid sequence."

[0053]

number

[0054] -Secondary structure of amino acid mutation site For example, the secondary structure at the mutation introduction position is classified into one of the three types: α-helix, β-sheet, or coil. For example, if the secondary structure at the mutation introduction position is an α-helix, the attribute is "The secondary structure at the mutation introduction position is an α-helix (Helix)."

[0055] · Creating a context table Based on the attributes described above, a context table is created to be given as an input for formal concept analysis. Formal concept analysis is performed based on the created context table.

[0056] Association Rules In the identification in the present invention, in order to analyze whether the degree of solubility has changed due to the introduction of mutation, among the obtained rules, rules having an attribute in the conclusion part that increases the degree of solubility and rules having an attribute that decreases the degree of solubility are extracted. In addition, as a rule extraction condition, for example, only rules having a reliability of 60% or more are extracted. The reliability can be appropriately set between 0 and 100, and may be set to, for example, 70% or more, 75% or more, 80% or more, 85% or more, 90% or more, or 95% or more. Based on the rules extracted in this way, the relationship between each attribute and the change in the degree of solubility is analyzed as described below.

[0057] Analysis of the relationship between amino acid mutations and changes in solubility The relationship between mutations from amino acids to (other) amino acids and changes in solubility is analyzed separately for cases where solubility increases and cases where solubility decreases.

[0058] Analysis of the relationship between the secondary structure and solubility changes at the amino acid mutation sites The relationship between mutations from amino acids to (other) amino acids and changes in solubility is analyzed separately for cases where solubility increases and cases where solubility decreases.

[0059] Relationship between amino acid mutation position and change in solubility The relationship between the position of the amino acid mutation and the change in solubility is analyzed separately for cases in which the solubility increases and cases in which the solubility decreases.

[0060] Relationship between multiple attributes and changes in solubility In the identification method of the present invention, factors that affect the degree of solubility may be searched for by focusing on amino acid mutations that may affect the change in the degree of solubility when an amino acid is mutated into (another) amino acid, and analyzing the relationship between multiple attributes and the change in the degree of solubility. In this case, all of the attributes set above may be used as the multiple attributes. Alternatively, the method may focus only on mutations that increase the degree of solubility, and search for only the tendency of mutations that tend to increase the degree of solubility of a protein.

[0061] Visualization of classification models Regarding the specific procedure of the identification method of the present invention, in one embodiment, a data collection (Enzyme Chemical Engineering Laboratory (Asano Laboratory) at the Biotechnology Research Center, Faculty of Engineering, Toyama Prefectural University) (sometimes referred to as "Thermus data" in this specification) summarizing the soluble expression of proteins derived from 414 types of extremely thermophilic bacteria Thermus thermophilus, is used to predict the soluble expression probability of a protein in which a mutation has been introduced comprehensively into a protein described in the data collection (sometimes referred to as "Thermus data" in this specification) using a trained protein solubility prediction model (classification model), and the predicted result is subjected to formal concept analysis to identify the attribute of the protein involved in the solubility of the protein. In the present invention, a mutation is introduced comprehensively into a portion of the amino acid sequence of the protein, and all amino acid sequences are predicted using the above-mentioned classification model. As a rule for mutation, 19 patterns of mutant amino acid sequences in which a portion of the base amino acid sequence is mutated to 19 types excluding the original amino acid are created as mutant amino acid sequences, and the soluble expression probability of the mutant amino acid sequence is predicted using the classification model. However, since the first character of the amino acid sequence is basically the amino acid M, typically, the first character is not mutated. The protein into which mutations are introduced comprehensively may be a wild-type protein. The number of mutations to be introduced is one to several (2, 3, 4, 5, 6, 7, 8, 9, etc.), more specifically, 1 to 5, 1 to 3, 1 or 2, or 1. In this specification, data summarizing the prediction results may be referred to as mutation prediction data. The mutation prediction data may include the name of the protein (name), the amino acid before the introduction of the mutation (Before), the position of the amino acid mutation (Position), the amino acid after the introduction of the mutation (After), the predicted value of the soluble expression probability (Predict_sol), and the change in the soluble expression probability (dif) (in the present invention, the change is also included in the attribute related to the solubility degree). The change in the soluble expression probability (attribute related to the solubility degree) may be calculated as the difference between the soluble expression probability of the protein predicted before the introduction of the mutation and the soluble expression probability of the protein predicted after the introduction of the mutation.

[0062] Attributes used in formal concept analysis The attributes set in the step of setting the protein attributes in the identification method of the present invention are not particularly limited as long as they can be used for input and output of a machine learning machine (protein solubility prediction model), and for example, the attributes may be extracted and set by the method described above. Specifically, (A) protein primary structure (amino acid sequence) data, structural information data, and biochemical attributes of amino acids are collected in a database. By formalizing them by formal concept analysis, attributes that are likely to affect solubility (expression) are organized as a logical formula. In this way, attributes that are factors of solubility are extracted. Specific examples of the attributes include the amount of change in solubility expression probability and amino acids before and after the introduction of mutations, the properties of amino acids before and after the introduction of mutations, the properties of amino acids after the introduction of mutations, the classification of amino acids before and after the introduction of mutations, the position of amino acid mutations, and the disorder region of the position of amino acid mutations. In this specification, the disorder region means a region in the three-dimensional structure of a protein where the structure has not been determined due to large structural fluctuations. The disorder region takes a value from 0 to 1, and the lower the value, the more stable the structure. The disordered region can be predicted by a method known per se or by using a site capable of predicting disordered regions in proteins (e.g., P. Radivojac, et al., Biophys J. 2007 Mar 1;92(5):1439-56. doi: 10.1529 / biophysj.106.094045, Jpred4 (http: / / www.compbio.dundee.ac.uk / jpred / ), PSIPHRED (http: / / bioinf.cs.ucl.ac.uk / psipred / ), NetSurfP2.0 (https: / / services.healthtech.dtu.dk / services / NetSurfP-2.0 / ), DISOPRED (http: / / bioinf.cs.ucl.ac.uk / disopred / ), Predictor of Natural Disordered Regions (PONDR) (http: / / www.pondr.com), etc.).In addition, the actual measured value of the disordered region may be, for example, a region with a large B-factor (a region with large structural fluctuation) in structural information obtained by X-ray crystal structure analysis. In one embodiment, the disordered region in the present invention is predicted by NetSurfP2.0 (https: / / services.healthtech.dtu.dk / services / NetSurfP-2.0 / ). In order to create a context table to be used in formal concept analysis, attributes must be given as binary values. The method of binarizing each attribute will be described below.

[0063] Change in probability of soluble expression The change in the soluble expression probability is classified into one of two attributes: the soluble expression probability has increased (Pre-sol-up) or the soluble expression probability has decreased (Pre-sol-down). If the value of the change in the soluble expression probability of the mutation prediction data is, for example, greater than 0.002, it is classified as the soluble expression probability has increased, and if the value is less than -0.002, it is classified as the soluble expression probability has decreased. The value of 0.002 is preferably used because it is the minimum value that a computer can tolerate when performing formal concept analysis, but is not limited to this value and can be appropriately set, for example, between values ​​of -1 and 1. If an attempt is made to analyze all mutations by formal concept analysis, a huge load is placed on the computer, and the computer may not be able to perform formal concept analysis. Therefore, a minimum threshold value may be set for the change in the soluble expression probability to reduce the number of protein mutations to a range that can be processed by the computer.

[0064] Amino acid before mutation The attribute is determined based on the amino acid before the mutation. For example, if the amino acid before the mutation is A (alanine), the attribute is "Amino acid before the mutation is A (Before-A)".

[0065] Amino acids after mutation The attribute is determined by the amino acid after mutation. For example, if the amino acid after mutation is A, the attribute is "Amino acid after mutation is A (After-A)".

[0066] -Properties of amino acids before mutation For example, amino acids before the introduction of mutation are classified into either hydrophilic or hydrophobic attributes according to Table 1. For example, if an amino acid before the introduction of mutation is hydrophilic, the attribute is "amino acid before the introduction of mutation is hydrophilic (Before-hydrophilic)".

[0067] -Properties of amino acids after mutation For example, the amino acid after the introduction of the mutation is classified into either hydrophilic or hydrophobic attribute according to Table 1. For example, if the amino acid after the introduction of the mutation is hydrophilic, the attribute is "amino acid after the mutation is hydrophilic (After-hydrophilic)".

[0068] Classification of amino acids before mutation For example, according to Table 2, the amino acid before the introduction of the mutation is classified into one of the five types: aliphatic, negatively charged, positively charged, aromatic, and uncharged. For example, if the amino acid before the introduction of the mutation is hydrophilic, the attribute is "amino acid before the introduction of the mutation is hydrophilic (Before-hydrophilic)".

[0069] Classification of amino acids after mutation For example, the amino acid after the mutation is classified into one of the five types, aliphatic, negatively charged, positively charged, aromatic, and uncharged, according to Table 2. For example, if the amino acid after the mutation is hydrophilic, the attribute is "After-hydrophilic amino acid."

[0070] -Amino acid mutation position For example, the position at which an amino acid is mutated is calculated using the above-mentioned formula 1 to determine the relative position p of the mutation in the amino acid sequence, and the attribute is classified according to whether the value of p is in the range of 0 or more and less than 0.25 (position-1), 0.25 or more and less than 0.5 (position-2), 0.5 or more and less than 0.75 (position-3), or 0.75 or more and less than 1 (position-4).

[0071] -Secondary structure of amino acid mutation site For example, the secondary structure at the mutation introduction position is classified into one of the three types: α-helix, β-sheet, or coil. For example, if the secondary structure at the mutation introduction position is an α-helix, the attribute is "The secondary structure at the mutation introduction position is an α-helix (Helix)."

[0072] Disorder region of amino acid mutation position For example, each attribute is classified according to whether the value of the Disorder region at the position where the amino acid mutation is introduced is less than 0.01 (Disorder-1), 0.01 or more but less than 0.1 (Disorder-2), 0.1 or more but less than 0.3 (Disorder-3), 0.3 or more but less than 0.8 (Disorder-4), or 0.8 or more (Disorder-5). For example, if the value of the Disorder region at the position where the mutation is introduced is 0.2, the attribute will be "the value of the Disorder region at the position where the mutation is introduced is 0.1 or more but less than 0.3 (Disorder-3)."

[0073] · Creating a context table Based on the attributes described above, a context table is created to be given as an input for formal concept analysis. Formal concept analysis is performed based on the created context table.

[0074] Association Rules In the identification in the present invention, in order to analyze whether the soluble expression probability has changed due to the introduction of mutation, among the obtained rules, rules having an attribute in the conclusion part that increases the soluble expression probability and rules having an attribute that decreases the soluble expression probability are extracted. In addition, as a rule extraction condition, for example, only rules with a reliability of 60% or more are extracted. Note that the reliability can be appropriately set between a value of 0 to 100, and may be set to, for example, 70% or more, 75% or more, 80% or more, 85% or more, 90% or more, or 95% or more. Based on the rules extracted in this way, the relationship between each attribute and the change in the soluble expression probability is analyzed as described below.

[0075] Analysis of the relationship between amino acid mutations and changes in the probability of soluble expression The relationship between mutations from an amino acid to (another) amino acid and changes in the probability of soluble expression is analyzed separately for cases where the probability of soluble expression increases and cases where the probability of soluble expression decreases.

[0076] Analysis of the relationship between disordered regions of amino acid mutation introduction positions and changes in soluble expression probability The relationship between mutations from one amino acid to another and the change in the probability of solubility expression is analyzed separately for cases where the degree of solubility increases and cases where the degree of solubility decreases.

[0077] Relationship between amino acid mutation position and change in solubility expression probability The relationship between the position of amino acid mutation and the change in the probability of solubility expression is analyzed separately for cases where the probability of solubility expression increases and cases where the degree of solubility decreases.

[0078] Relationship between multiple attributes and changes in solubility In the identification method of the present invention, factors that affect the solubility expression probability may be searched for by focusing on amino acid mutations that may affect the change in solubility expression probability when an amino acid is mutated to (another) amino acid, and analyzing the relationship between multiple attributes and the change in solubility expression probability. In this case, all of the attributes set above may be used for the multiple attributes. For the disorder region at the mutation introduction position, only the attributes of Disorder-1 and Disorder-5, which are separated and show the difference in the values ​​of the disorder region, may be used as attributes. In addition, focusing only on mutations that increase the solubility degree, only the tendency of mutations that tend to increase the solubility degree of a protein may be searched for.

[0079] ·Protein properties that contribute to the solubility of the identified protein Examples of the attributes of proteins that increase the solubility of proteins obtained by the identification method of the present invention include the following: (i) Substitution of at least one amino acid selected from the group consisting of arginine, tyrosine, and glutamine among amino acids constituting a protein with an amino acid other than arginine, tyrosine, and glutamine; (ii) Substitution of amino acids other than glycine, valine, threonine, cysteine, and lysine constituting a protein with glycine, valine, threonine, cysteine, or lysine; and (iii) Substitution of aromatic amino acids in the disordered region of the protein with glycine, valine, or threonine etc.

[0080] The amino acids to be substituted in the above (i) to (iii) are preferably amino acids in the latter half of the amino acid sequence of the protein. It is also preferable that aromatic amino acids in the disordered region of the protein are hydrophilic.

[0081] 2. The manufacturing method of the present invention The present invention provides a method for producing a protein with improved solubility, which comprises a step of modifying the protein so that it has an attribute identified by the identification method of the present invention (herein also referred to as production method (1) of the present invention). The production method (1) of the present invention can modify a protein to be produced so that the protein has an attribute that improves the solubility of the protein identified by the identification method of the present invention (an attribute that increases solubility). Specific production and modification (introduction of mutations, etc.) of a protein can be performed by appropriate design, setting, etc., based on a method known per se.

[0082] The present invention also provides a method for producing a protein having improved solubility, comprising the steps of: A step of substituting at least one amino acid selected from the group consisting of arginine, tyrosine, and glutamine among amino acids constituting a protein with an amino acid other than arginine, tyrosine, and glutamine; A step of substituting amino acids other than glycine, valine, threonine, cysteine, and lysine constituting a protein with glycine, valine, threonine, cysteine, or lysine; and Substituting aromatic amino acids in a disordered region of a protein with glycine, valine, or threonine; The present invention also provides a method (herein also referred to as the production method (2) of the present invention) comprising one or more steps selected from the group consisting of:

[0083] In the production method (2) of the present invention, the amino acids to be substituted are preferably those in the latter half of the amino acid sequence of the protein. It is also preferable that aromatic amino acids in the disordered region of the protein are hydrophilic.

[0084] The production method (2) of the present invention can modify a protein to be produced so that the protein has an attribute that improves the degree of solubility of the protein identified by the identification method of the present invention (an attribute that increases solubility). Specific production and modification (introduction of mutations, etc.) of a protein can be performed by appropriate design and setting based on a method known per se. For example, the materials and methods used in the production method of the present invention are as follows.

[0085] In the case of E. coli, the E. coli carrying a gene encoding an amino acid sequence constituting a target protein to be mutated may be cultured by batch culture, continuous culture, or the like. Static culture or shaking culture may also be used, but shaking culture is preferred. The medium may be LB medium (yeast extract 5.0 g / l, NaCl 10.0 g / l, tryptone 10.0 g / l), or the like. Examples of the culture method include culturing at 37°C until the bacterial turbidity reaches about 0.5, adding IPTG, and culturing at a temperature of 16 to 37°C for 16 to 24 hours, but the culture method is not particularly limited as long as the recombinant E. coli can grow. In the case of a heterologous expression host other than E. coli, a culture method known per se for each host may be used.

[0086] Protein extraction After the main culture, the host such as E. coli can be disrupted to prepare a crude enzyme solution containing the target protein into which a mutation has been introduced. In the present invention, the host such as E. coli can be collected, the cells can be disrupted by an ultrasonic disrupter or the like, and the supernatant and the precipitate containing the inclusion bodies can be separated by centrifugation to obtain a supernatant (soluble fraction) that can be used as the crude enzyme solution. The target protein into which a mutation has been introduced obtained by the production method of the present invention is usually soluble, so that the crude protein suspension contains the target protein into which a mutation has been introduced and has a predetermined activity or function. Therefore, the obtained crude protein suspension can be used as it is as a liquid containing the target protein into which a mutation has been introduced. The target protein into which a mutation has been introduced can also be isolated and purified from the obtained crude protein suspension and used. In this case, general biochemical methods (e.g., ammonium sulfate precipitation, gel chromatography, ion exchange chromatography, affinity chromatography, etc.) can be used alone or in appropriate combination to isolate and purify the protein into which a mutation has been introduced. The isolated and purified target protein into which a mutation has been introduced can be used, for example, in a state suspended in a buffer solution of a predetermined pH.

[0087] Proteins (active mutant enzymes) having the same type of enzymatic activity as that obtained in the natural form may be selected from the mutant proteins expressed in E. coli or the like. Selection of active mutant enzymes can be carried out, for example, as follows. In a system capable of measuring the enzymatic activity of each protein in the natural form, the enzyme activity of the obtained recombinant protein can be measured. As the enzyme activity measurement system, a method known for each enzyme can be appropriately used.

[0088] For necessary matters other than those mentioned above regarding the production methods (1) and (2) of the present invention, the contents of "1. Identification method of the present invention" are entirely incorporated by reference.

[0089] 3. Method for improving the present invention The present invention relates to a method for improving protein solubility, the method comprising the steps of: A step of substituting at least one amino acid selected from the group consisting of arginine, tyrosine, and glutamine among amino acids constituting a protein with an amino acid other than arginine, tyrosine, and glutamine; A step of substituting amino acids other than glycine, valine, threonine, cysteine, and lysine constituting a protein with glycine, valine, threonine, cysteine, or lysine; and Substituting aromatic amino acids in a disordered region of a protein with glycine, valine, or threonine; The method includes one or more steps selected from the group consisting of:

[0090] In the method for improving the solubility of a protein of the present invention, the amino acids to be substituted are preferably those in the latter half of the amino acid sequence of the protein. It is also preferable that aromatic amino acids in the disordered region of the protein are hydrophilic.

[0091] The method for improving the solubility of a protein of the present invention can substitute amino acids so that a desired protein has an attribute that improves the degree of solubility of the protein identified by the identification method of the present invention (an attribute that increases solubility). A specific method for substituting amino acids (method for introducing mutations) can be appropriately designed and set based on a method known per se.

[0092] For other necessary matters concerning the improvement method of the present invention other than those mentioned above, the contents of "1. Identification method of the present invention" and "2. Production method of the present invention" are all incorporated by reference.

[0093] The present invention will be described in more detail below with reference to examples, but the present invention is not limited to these in any way. EXAMPLES

[0094] Example 1: Solubility prediction using a recurrent neural network (creation of a trained protein solubility prediction model) Using LSTM, a type of RNN, we constructed a regression model that predicts and outputs the degree of solubility from protein structural information, and a classification model that predicts whether a protein is "soluble" or "insoluble" based on its structural information, as described below.

[0095] About regression models (1) Data used The "Solubility database of all Escherichia coli proteins (eSOL)" was used as the protein data to be trained (eSOL database developed in the Targeted Proteins Research Project, http: / / www.tanpaku.org / tp-esol / index.php?lang=en (accessed May. 26, 2022)). eSOL is a data collection that covers the experimental results of the degree of aggregation (solubility rate) and synthesis amount when all proteins of Escherichia coli (E. coli) are expressed using the PURE system, a reconstructed in vitro protein synthesis system, and protein data is summarized. The data summarizes data on proteins such as the clone ID (JW_ID), solubility degree (Solubility), and molecular weight (Calculated MW) of each protein. For network training, 80% of eSOL was used as training data for learning, and the remaining 20% ​​was used as test data for verifying the prediction accuracy of the regression model. The secondary structure of proteins is not stored in eSOL. Therefore, information on the secondary structure must be obtained from elsewhere. In Example 1, PSIPRED was used as a means for acquiring secondary structure information. PSIPRED is one of the secondary structure prediction methods incorporating a forward propagation neural network that analyzes the output of PSI-BLAST.

[0096] (2) Construction of neural network In Example 1, one node was placed in the output layer to create a regression model that predicts the degree of solubility from the amino acid sequence and secondary structure information of a protein. An identity function was used as the activation function for the output layer. This identity function is a function that outputs the value received as an argument as is, as shown in the following equation 2.

[0097]

number

[0098] (3) Various settings during study Learning Settings The mean square error was used as the loss function of the regression model created in Example 1. If the loss is l, the mean square error can be calculated by the following formula 3.

[0099]

number

[0100] In the above formula, xN indicates the Nth data input to the network. dN indicates the correct answer for input xN. f(xN) indicates the output when xN is input to the network. The network was trained to minimize the loss of this function.

[0101] Hyperparameters The neural network created in Example 1 has four hyperparameters. Table 5 shows the hyperparameters in the regression model.

[0102] [Table 5]

[0103] Input data settings The amino acid sequence and secondary structure information were divided into amino acids and secondary structure characters, respectively, and one character was used as input at each time point. In order to provide the regression model with information on 20 types of amino acids, three types of secondary structures, and dummy characters, the characters were converted into 61-dimensional one-hot vectors. A one-hot vector is a vector in which only one component is 1 and the remaining components are all 0. For example, when amino acid A is converted, the vector becomes [1,0,0,0,...,0], and when another amino acid C is converted, the component that becomes 1 changes depending on the type of character converted, such as [0,1,0,0,...,0].

[0104] Furthermore, in order to perform mini-batch learning with LSTM, it is necessary to standardize the length of the data, that is, the length of the amino acid sequence. Therefore, in Example 1, the length of the amino acid sequence is limited to 600 characters. Some proteins have amino acid sequences that are less than 600 characters long. In that case, the length of all data is standardized by padding with dummy characters X so that the characters are 600 characters. With the above settings, the dimension of the input data given to LSTM is B x 61 x n. B represents the batch size. n was calculated based on the following formula 4, where N is the length of the amino acid sequence.

[0105]

number

[0106] (4) Learning outcomes Figure 2 shows the results of predicting the degree of solubility of the test data using a regression model (trained protein solubility prediction model) trained on training data. The horizontal axis of the figure shows the predicted value output by the regression model, and the vertical axis shows the correct value. In order to evaluate the accuracy of the created regression model, a rule was established to calculate the accuracy rate. A sufficient prediction was considered to be made if the difference between the predicted value and the correct value was 0.2 (20%) or less. In the figure, the area surrounded by the dashed line indicates that range. The percentage of the test data that fell within the dashed area was regarded as the accuracy rate. The prediction results of the regression model created by training on the amino acid sequence and secondary structure of the protein showed a coefficient of determination (R2) of 0.38, a slope of the regression line of the scatter plot of 0.77, and a correct answer rate of 65%. The coefficient of determination, 0.38, is the same level of predictive accuracy as a regression model predicting the degree of protein solubility described in eSOL created elsewhere (Xi Han, et al., Bioinformatics, Volume 35, Issue 22, November 2019, Pages 4640-4646, https: / / doi.org / 10.1093 / bioinformatics / btz294), and it can be said that the regression model we created is a model with sufficient accuracy.

[0107] About classification models (1) Data used Experimental data from the Enzyme Chemical Engineering Laboratory (Asano Laboratory), Biotechnology Research Center, Faculty of Engineering, Toyama Prefectural University, was used as protein data for training the classification model. The data is a collection of data on the soluble expression of 414 types of proteins derived from the extreme thermophile Thermus thermophilus, which were examined using an E. coli expression system. In the data collection, the solubility (sol) of the protein is evaluated in four classes: -, +, ++, and +++. In the examples, a binary classification model was created in which Sol class - was considered insoluble and Sol class other than - was considered soluble. In addition, 80% of the data was used as learning data, and the remaining 20% ​​was used as test data for verifying the prediction accuracy of the classification model.

[0108] (2) Neural network configuration In this embodiment, to create a classification model that classifies proteins into two classes, "soluble" and "insoluble," two nodes corresponding to each class were placed in the output layer. A softmax function was used as the activation function for the output layer. If the total input to the k-th node in the output layer is uk, the softmax function is the function shown in the following formula 5.

[0109]

number

[0110] The above function has the characteristic shown in the following equation 6, so the output of each node can be regarded as a probability.

[0111]

number

[0112] Therefore, it can be said that the neural network outputs the probability that the protein inputted into the neural network is soluble or insoluble. The classification model classifies the inputted protein into the class for which this probability is the maximum value. In this specification, the probability that the protein is soluble may be referred to as the "solubility expression probability."

[0113] (3) Various settings during study Learning Settings In the embodiment, the loss function of the neural network is the cross entropy error. If the loss is l, the cross entropy error can be calculated by the following formula 7.

[0114]

number

[0115] In the above formula, xn indicates the nth data input to the network. dn indicates the correct answer for input xn. y(xn) indicates the output when the input is input to the network. The network was trained to minimize the loss of this function.

[0116] Hyperparameters The neural network created in the example has four hyperparameters. Table 6 shows the hyperparameters in the classification model.

[0117] [Table 6]

[0118] Input data settings 155 amino acids were extracted from the position of a certain amino acid in the amino acid sequence, and the block of 155 amino acid sequences was used as the input at each time. In addition, as the time progressed, the position of the amino acid was shifted back by one. In order to provide the classification model with 20 types of amino acids and dummy characters as information, the characters were converted into 21-dimensional one-hot vectors. Since the classification model does not perform mini-batch learning, it is not necessary to standardize the length of the data, i.e., the length of the amino acid sequence. Therefore, there is no restriction on the length of the amino acid sequence, but in order to input an amino acid sequence of 155 length at each time of LSTM, the length of the amino acid sequence shorter than 155 was adjusted using the dummy character X. With the above settings, the dimension of the input data provided to LSTM is 1 × 3255 (21 × 155) × W. W was calculated based on the following formula 8, where N is the length of the amino acid sequence.

[0119]

number

[0120] Learning outcomes The confusion matrix created from the results of solubility prediction of the test data by the classification model trained on the training data (trained protein solubility prediction model) is shown in Table 7 below. For model evaluation, the accuracy rate was calculated from the confusion matrix based on the following equation 9. TP, TN, FP, and FN in equation 9 are abbreviations for true positive, true negative, false positive, and false negative, respectively, and correspond to the respective positions in Table 8 below.

[0121] [Table 7]

[0122]

number

[0123] [Table 8]

[0124] The accuracy rate was calculated to be 79.3%. Therefore, it can be said that the classification model created in the embodiment has sufficient prediction accuracy.

[0125] Example 2: Visualization of solubility predictors using formal concept analysis (regression models) Neural network models have a black box problem, and it is difficult to explain how the models have learned. Therefore, by applying formal concept analysis to the regression model and classification model created in Example 1, the learning trends of these models were analyzed, and attempts were made to visualize the models so that explanations could be made. In this way, mutations that are likely to improve the degree of solubility and the probability of solubility expression were searched for.

[0126] Visualization of regression models In the working examples, the degree of solubility of proteins into which comprehensive mutations have been introduced into proteins listed in eSOL was predicted using a trained protein solubility prediction model (regression model), and the predicted results were subjected to formal concept analysis to identify the protein attributes involved in the solubility of the protein.

[0127] (1) Mutation prediction data In the embodiment, a mutation was introduced comprehensively into a portion of the amino acid sequence of the protein, and all amino acid sequences were predicted by the regression model created in the embodiment 1. As a rule of mutation, 19 patterns of mutant amino acid sequences were created by mutating a portion of the base amino acid sequence to 19 types excluding the original amino acid, and the solubility of the mutant amino acid sequence was predicted by the regression model. However, since the first character of the amino acid sequence is basically the amino acid M, the first character was not mutated. The data summarizing the above prediction results may be referred to as "mutation prediction data" hereinafter. The mutation prediction data includes the protein clone number (JW_ID), the amino acid before the mutation (Before), the position of the amino acid mutation (Position), the amino acid after the mutation (After), the predicted value of the solubility degree (Predict_sol), the change in the solubility degree (dif) (attributes related to the solubility degree), and the like. The change in solubility (attribute related to solubility) was calculated as the difference between the solubility of the protein predicted before the introduction of the mutation and the solubility of the protein predicted after the introduction of the mutation.

[0128] (2) Attributes used in formal concept analysis In the embodiment, the change in solubility, amino acid before mutation, amino acid after mutation, property of amino acid before mutation, property of amino acid after mutation, classification of amino acid before mutation, classification of amino acid after mutation, position of amino acid mutation, and secondary structure of position of amino acid mutation were used as attributes of formal concept analysis. In order to create a context table to be used in formal concept analysis, attributes must be given as binary values. The method of binarizing each attribute is explained below.

[0129] Change in solubility The change in solubility was classified into one of two attributes: increased solubility (Pre-sol-up) or decreased solubility (Pre-sol-down). If the value of the change in solubility in the mutation prediction data was greater than 0.055, it was classified as increased solubility, and if the value was less than -0.055, it was classified as decreased solubility. The value of 0.055 was determined because it is the minimum value that a computer can tolerate when performing formal concept analysis. If an attempt is made to analyze all mutations using formal concept analysis, a huge load is placed on the computer, and the computer may not be able to perform formal concept analysis. Therefore, a minimum threshold was set for the change in solubility to reduce the number of protein mutations to a range that can be processed by a computer.

[0130] Amino acid before mutation The attribute was determined based on the amino acid before the mutation. For example, if the amino acid before the mutation was A (alanine), the attribute was "Amino acid before the mutation was A (Before-A)."

[0131] Amino acids after mutation The attribute was classified according to the amino acid after mutation. For example, if the amino acid after mutation is A, the attribute was "Amino acid after mutation is A (After-A)".

[0132] -Properties of amino acids before mutation For example, amino acids before the introduction of mutations were classified into either hydrophilic or hydrophobic attributes according to Table 1. For example, if an amino acid before the introduction of mutations is hydrophilic, the attribute was set to "amino acid before the introduction of mutations is hydrophilic (Before-hydrophilic)."

[0133] -Properties of amino acids after mutation For example, the amino acids after the introduction of the mutation were classified into either hydrophilic or hydrophobic attributes according to Table 1. For example, if the amino acid after the introduction of the mutation is hydrophilic, the attribute was given as "amino acid after the mutation is hydrophilic (After-hydrophilic)."

[0134] Classification of amino acids before mutation For example, the amino acid before the mutation was classified into five attributes: aliphatic, negatively charged, positively charged, aromatic, and uncharged, according to Table 2. For example, if the amino acid before the mutation was hydrophilic, the attribute was "amino acid before the mutation is hydrophilic (Before-hydrophilic)."

[0135] Classification of amino acids after mutation For example, the amino acid after the mutation was classified into five attributes: aliphatic, negatively charged, positively charged, aromatic, and uncharged, according to Table 2. For example, if the amino acid after the mutation is hydrophilic, the attribute was "amino acid after the mutation is hydrophilic (After-hydrophilic)."

[0136] -Amino acid mutation position The position for introducing an amino acid mutation was determined by calculating the relative mutation position p for the amino acid sequence from the above formula 1, and classifying the attributes according to which of the following four ranges for p: 0 or more and less than 0.25 (position-1), 0.25 or more and less than 0.5 (position-2), 0.5 or more and less than 0.75 (position-3), or 0.75 or more and within 1 (position-4). In the above formula 1, the "maximum length of the amino acid sequence of the protein" may mean the full-length amino acid sequence of the protein for which it is desired to calculate the mutation introduction position p.

[0137] -Secondary structure of amino acid mutation site For example, the secondary structure at the mutation introduction position was classified into three types: α-helix, β-sheet, and coil. For example, if the secondary structure at the mutation introduction position is an α-helix, the attribute was set as "The secondary structure at the mutation introduction position is an α-helix (Helix)."

[0138] (3) Creating a context table Based on the attributes described above, a context table was created to be used as an input for formal concept analysis. Formal concept analysis was performed based on the created context table.

[0139] (4) Association Rules Table 9 shows some of the association rules obtained by formal concept analysis. To analyze whether the degree of solubility changed due to the introduction of mutations, from the obtained rules, rules whose conclusions have attributes that increase the degree of solubility and rules whose conclusions have attributes that decrease the degree of solubility were extracted. In addition, as a rule extraction condition, for example, only rules with a reliability of 60% or more were extracted. Table 10 shows some of the extracted rules with attributes that increase the degree of solubility, and Table 11 shows some of the rules with attributes that decrease the degree of solubility. Based on the rules extracted in this way, the relationship between each attribute and the change in the degree of solubility was analyzed.

[0140] [Table 9]

[0141] [Table 10]

[0142] [Table 11]

[0143] (5) Analysis of the relationship between amino acid mutations and changes in solubility The relationship between mutations from amino acids to (other) amino acids and changes in solubility was analyzed separately for cases where solubility increased and cases where solubility decreased.

[0144] -Rules for increasing solubility We analyzed the relationship between mutations from a specific amino acid to another amino acid and changes in solubility when the degree of solubility increases. First, we extracted rules from the rules in Table 9 that have attributes of the amino acid before and after the mutation in the premise, and an attribute that increases the degree of solubility in the conclusion. We then created a table with the amino acids before the mutation as rows and the amino acids after the mutation as columns, and analyzed which amino acids were mutated to which amino acids based on the extracted rules, and summarized the mutations from amino acids to amino acids (Figure 3). We also created a heat map (Figure 3) based on the reliability of the rules, and analyzed which mutations are likely to affect the degree of solubility.

[0145] As shown in Figure 3, the amino acids Y (tyrosine), W (tryptophan), F (phenylalanine), and R (arginine) appeared frequently before the introduction of mutations, and the amino acids D (aspartic acid), E (glutamic acid), and K (lysine) appeared frequently after the introduction of mutations, as mutations that increased the degree of solubility.

[0146] When looking at the color distribution of the heat map in Figure 3, the parts where the amino acids before the mutation were D, E, and K, and the parts where the amino acids before the mutation were Y, W, R, and F, are colored overall. Among these, the parts where the amino acids before the mutation were Y, W, F, and R and the amino acids after the mutation were D, E, and K were colored darker. From the above, it was thought that mutating either the amino acid Y or W, or mutating to either the amino acids D, E, or K would likely increase the degree of solubility. In addition, it was thought that mutating either the amino acid Y, W, F, or R to either the amino acid D, E, or K would also likely increase the degree of solubility.

[0147] -Rules for decreasing solubility We analyzed the relationship between mutations from a specific amino acid to another amino acid and changes in solubility when the degree of solubility decreases. First, we extracted rules from the rules in Table 11 that have the attributes of the amino acid before and after the introduction of the mutation in the premise, the attribute of decreasing the degree of solubility in the conclusion, and the attribute of decreasing the degree of solubility in the conclusion. We then created a table with the amino acids before the introduction of the mutation as rows and the amino acids after the introduction of the mutation as columns, and analyzed which amino acids were mutated to which amino acids based on the extracted rules, and summarized the mutations from amino acids to amino acids. We also created a heat map based on the reliability of the rules (Figure 4) and analyzed which mutations are likely to affect the degree of solubility.

[0148] As shown in Figure 4, the amino acids D, E, and K that appeared frequently before the mutation reduced the degree of solubility, and the amino acids Y, W, R, F, and C (cysteine) that appeared frequently after the mutation.

[0149] From the color distribution of the heat map in Figure 4, it was found that the parts where the amino acids before the mutation were D, E, and K, and the parts where the amino acids after the mutation were Y and W, were generally colored. Among them, the parts where the amino acids before the mutation were D, E, and K were generally darker in color. From the above, it was thought that mutating any of the amino acids D, E, and K, or mutating to any of the amino acids Y and W, would likely reduce the degree of solubility. In addition, it was thought that mutations from amino acids D, E, and K to amino acids R, F, and C would also likely reduce the degree of solubility.

[0150] (6) Analysis of the relationship between the secondary structure and the degree of solubility at the amino acid mutation site The relationship between the introduction of amino acid mutations and the change in solubility was analyzed in two cases: when the solubility increased and when it decreased.

[0151] -Rules for increasing solubility The relationship between the change in solubility and the mutation of amino acids to (other) amino acids for each secondary structure at the mutation introduction position of an amino acid in the case where the degree of solubility increases was analyzed. First, from the rules in Table 10, rules were extracted that have the attributes of the amino acid before the mutation introduction, the amino acid after the mutation introduction, and the secondary structure at the mutation introduction position of the amino acid in the premise part, and the attribute that increases the degree of solubility in the conclusion part. Then, a table was created with the amino acids before the mutation introduction as rows and the amino acids after the mutation introduction as columns, and based on the extracted rules, it was analyzed which amino acid was mutated to which amino acid on which secondary structure. The introduction of mutations from amino acids to (other) amino acids when the secondary structure at the mutation introduction position is a coil is summarized in Figure 5, the introduction of mutations from amino acids to (other) amino acids when the secondary structure at the mutation introduction position is a β-sheet is summarized in Figure 6, and the introduction of mutations from amino acids to (other) amino acids when the secondary structure at the mutation introduction position is an α-helix is ​​summarized in Figure 7. In addition, heat maps were created based on the reliability of the rules (Figures 5 to 7), and which mutations are likely to affect the degree of solubility were analyzed.

[0152] From each of Figures 5 to 7, when the secondary structure of the mutation introduction position is coil, the amino acids R, Y, and W appeared frequently as amino acids before the mutation introduction, and the amino acids G (glycine), P (proline), A, D, and E appeared frequently as amino acids after the mutation introduction. When the secondary structure of the mutation introduction position is β-sheet, the amino acids Y, W, and F appeared frequently as amino acids before the mutation introduction, and the amino acids D, E, K, and T (threonine) appeared frequently as amino acids after the mutation introduction. In addition, although the number of appearances is small, the amino acids Q (glutamine) and I (isoleucine) appeared frequently as amino acids after the mutation introduction. When the secondary structure of the mutation introduction position is α-helix, the amino acids P, I, C, Y, and F appeared frequently as amino acids before the mutation introduction, and the amino acids D, E, K, A, V (valine), and M (methionine) appeared frequently as amino acids after the mutation introduction. It was found that for each secondary structure at the mutation introduction site, there were common mutations that increased the degree of solubility and other mutations that differed, and that the mutations that tended to increase the degree of solubility differed for each secondary structure.

[0153] From the color distribution of the heat maps in Figures 5 to 7, it was found that when the secondary structure of the mutation introduction position was a coil, the parts where the amino acids R, Y, and W were mutated to the amino acids G, P, A, D, and E were darkened. When the secondary structure of the mutation introduction position was a β-sheet, the parts where the amino acids Y, W, and F were mutated to the amino acids I, D, E, K, and T were darkened. When the secondary structure of the mutation introduction position was an α-helix, the parts where the amino acids P, I, C, Y, and F were mutated to the amino acids A, V, D, E, and K were darkened.

[0154] From the above, it was thought that in any secondary structure, mutation of amino acid Y or mutation of an amino acid to any of the amino acids D, E, or K would likely increase the degree of solubility. For each secondary structure, when the secondary structure at the mutation introduction position is a coil, mutation of amino acid W or mutation of amino acid R to any of the amino acids G, P, or A would likely increase the degree of solubility. When the secondary structure at the mutation introduction position is a β-sheet, mutation of either amino acid F or W to amino acid T would likely increase the degree of solubility. When the secondary structure at the mutation introduction position is an α-helix, mutation of either amino acid P or I to any of the amino acids A or V would likely increase the degree of solubility.

[0155] -Rules for decreasing solubility The relationship between the change in solubility and the mutation of amino acids to (other) amino acids for each secondary structure at the position of the amino acid mutation introduction in the case where the degree of solubility is decreased was analyzed. First, from the rules in Table 11, rules with the attributes of the amino acid before the mutation introduction, the amino acid after the mutation introduction, and the secondary structure at the position of the amino acid mutation introduction in the premise part, and the attribute of the degree of solubility decrease in the conclusion part were extracted. Then, a table was created with the amino acids before the mutation introduction as rows and the amino acids after the mutation introduction as columns, and based on the extracted rules, it was analyzed which amino acids were mutated to which amino acids on which secondary structure. The mutations from amino acids to amino acids when the secondary structure at the position of the mutation introduction is a coil are summarized in Figure 8, the mutations from amino acids to (other) amino acids when the secondary structure at the position of the mutation introduction is a β-sheet are summarized in Figure 9, and the mutations from amino acids to (other) amino acids when the secondary structure at the position of the mutation introduction is an α-helix are summarized in Figure 10. In addition, heat maps were created based on the reliability of the rules (Figures 8 to 10), and which mutations are likely to affect the degree of solubility were analyzed.

[0156] From each of Figures 8 to 10, when the secondary structure at the mutation introduction position is a coil, it was found that the amino acids G, P, A, D, and E appeared frequently as amino acids before the mutation introduction, and the amino acids R, Y, and W appeared frequently as amino acids after the mutation introduction. When the secondary structure at the mutation introduction position is a β-sheet, the amino acids D, E, K, and T appeared frequently as amino acids before the mutation introduction, and the amino acids Y, W, and F appeared frequently as amino acids after the mutation introduction. When the secondary structure at the mutation introduction position is an α-helix, the amino acids A, V, D, E, and K appeared frequently as amino acids before the mutation introduction, and the amino acids P, I, C, Y, F, and W appeared frequently as amino acids after the mutation introduction. It was found that there were parts that appeared in common as mutations that increased the degree of solubility for each secondary structure at the mutation introduction position and parts that differed, and that there were differences in the introduction of mutations that tended to increase the degree of solubility for each secondary structure.

[0157] From the color distribution of the heat maps in Figures 8 to 10, it was found that when the secondary structure of the mutation introduction position was a coil, the parts where the amino acids G, P, A, D, and E were mutated to the amino acids R, Y, and W were darkened. When the secondary structure of the mutation introduction position was a β-sheet, the parts where the amino acids I, D, E, K, and T were mutated to the amino acids Y, W, and F were darkened. When the secondary structure of the mutation introduction position was an α-helix, the parts where the amino acids P, I, C, Y, and F were mutated to the amino acids A, V, D, E, and K were darkened.

[0158] From the above, it is thought that in any secondary structure, mutation of any of the amino acids D, E, or K, or mutation of an amino acid to another amino acid is likely to reduce the degree of solubility. For each secondary structure, if the secondary structure at the mutation introduction position is a coil, mutation of any of the amino acids G, P, or A to any of the amino acids R or W is likely to reduce the degree of solubility. If the secondary structure at the mutation introduction position is a β-sheet, mutation of the amino acid T to the amino acid F or W is likely to reduce the degree of solubility. If the secondary structure at the mutation introduction position is an α-helix, mutation of any of the amino acids P or I to any of the amino acids A or V is likely to reduce the degree of solubility.

[0159] (7) Relationship between the position of amino acid mutation and the change in solubility The relationship between the position of the amino acid mutation and the change in solubility was analyzed for two cases: increased solubility and decreased solubility.

[0160] -Rules for increasing solubility The relationship between the change in solubility and the mutation of an amino acid to (another) amino acid for each mutation position of an amino acid in the case where the degree of solubility increases was analyzed. First, from the rules in Table 10, rules were extracted that have the attributes of the amino acid before the mutation introduction, the amino acid after the mutation introduction, and the mutation position of the amino acid in the premise part, and the attribute of the degree of solubility increase in the conclusion part. Then, a table was created with the amino acid before the mutation introduction as rows and the amino acid after the mutation introduction as columns, and based on the extracted rules, it was analyzed which amino acid was mutated to which (other) amino acid at which position based on the extracted rules. The mutation of an amino acid to (another) amino acid when the mutation introduction position is position-1 is summarized in FIG. 11, the mutation of an amino acid to (another) amino acid when the mutation introduction position is position-2 is summarized in FIG. 12, the mutation of an amino acid to (another) amino acid when the mutation introduction position is position-3 is summarized in FIG. 13, and the mutation of an amino acid to (another) amino acid when the mutation introduction position is position-4 is summarized in FIG. 14. In addition, heat maps were created based on the reliability of the rules (Figures 11 to 14) to analyze which mutations are likely to affect the degree of solubility.

[0161] 11 to 14, it appears that the mutations that increase the degree of solubility at each amino acid mutation position are common. At each mutation position, the amino acids Y, W, F, and R that increase the degree of solubility were most common before mutation, and the amino acids D, E, and K that increase the degree of solubility were most common after mutation. This is similar to the tendency obtained in (5) of Example 2, and it was considered that there was no difference in the mutations that easily increase the degree of solubility at each mutation position.

[0162] From the color distribution of the heat maps in Figures 11 to 14, it was found that the parts where the amino acids before the mutation were D, E, and K, and the parts where the amino acids after the mutation were Y and W were generally colored. Among them, the parts where the amino acids before the mutation were D, E, and K were generally darker in color. This tendency was also similar to the tendency obtained from (5) of Example 2.

[0163] From the above, it was considered that there was no difference in the tendency of the mutation to easily increase the degree of solubility regardless of the position at which the amino acid was mutated.

[0164] -Rules for decreasing solubility The relationship between the change in solubility and the mutation of an amino acid to (another) amino acid for each mutation position of an amino acid in cases where the degree of solubility decreases was analyzed. First, from the rules in Table 11, rules with the attributes of the amino acid before the mutation, the amino acid after the mutation, and the mutation position of the amino acid in the premise and the attribute of the decrease in solubility in the conclusion were extracted. Then, a table was created with the amino acids before the mutation as rows and the amino acids after the mutation as columns, and based on the extracted rules, it was analyzed which amino acid was mutated to which (other) amino acid at which position based on the extracted rules. The mutation of an amino acid to (another) amino acid when the mutation position is position-1 is summarized in Figure 15, the mutation of an amino acid to (another) amino acid when the mutation position is position-2 is summarized in Figure 16, the mutation of an amino acid to (another) amino acid when the mutation position is position-3 is summarized in Figure 17, and the mutation of an amino acid to (another) amino acid when the mutation position is position-4 is summarized in Figure 18. In addition, heat maps were created based on the reliability of the rules (FIGS. 15 to 18), and analysis was performed to determine which mutations were most likely to affect the degree of solubility.

[0165] From each figure, it was found that the amino acids D, E, and K appeared frequently before the introduction of mutations that reduced the degree of solubility, and the amino acids Y, W, R, F, and C (cysteine) appeared frequently after the introduction of mutations. This is similar to the tendency obtained from (5) of Example 2, and it was considered that there was no difference in the mutations that tended to increase the degree of solubility depending on the position of the mutation.

[0166] From the color distribution of the heat maps in Figures 15 to 18, it was found that the amino acids D, E, and K before the mutation and the amino acids Y and W after the mutation were generally colored. Among them, the amino acids D, E, and K before the mutation were generally darker in color.

[0167] From the above, it is believed that there is no difference in the degree of solubility reduction that occurs regardless of the amino acid mutation position.

[0168] (8) Relationship between multiple attributes and changes in solubility Focusing on amino acid mutations that may affect the change in solubility in the mutation from an amino acid to (another) amino acid, factors that affect the solubility were explored by analyzing the relationship between multiple attributes and the change in solubility. All attributes set in (2) of Example 2 were used as the multiple attributes. In addition, since the results of the analysis of the relationship between each attribute and the change in solubility were considered to be symmetrical between mutations that increase the solubility and mutations that decrease the solubility, attention was focused only on mutations that increase the solubility, and only the tendency of mutations that tend to increase the solubility of a protein was explored.

[0169] From (5) of Example 2, as an amino acid mutation that is considered to affect the change in solubility degree and to easily increase the solubility degree, attention was paid to the attributes of the amino acid mutations in which the amino acid before the mutation is Y, the amino acid before the mutation is W, the amino acid before the mutation is F, the amino acid after the mutation is D, the amino acid after the mutation is E, and the amino acid after the mutation is K. The amino acid before or after the mutation is fixed as an attribute, and a rule was extracted in which the premise part contains a combination of the fixed attribute and a specific attribute, and the conclusion part has an attribute that increases the solubility degree. A graph was created from the extracted rules with the reliability on the vertical axis and the combination of each attribute on the horizontal axis, and a mutation introduction that is likely to change the solubility degree in the fixed attribute was searched for.

[0170] Figure 19 shows the relationship between multiple attributes and change in solubility degree when the amino acid before mutation is Y, Figure 35 shows the relationship between multiple attributes and change in solubility degree when the amino acid before mutation is W, Figure 36 shows the relationship between multiple attributes and change in solubility degree when the amino acid before mutation is F, Figure 37 shows the relationship between multiple attributes and change in solubility degree when the amino acid after mutation is D, Figure 38 shows the relationship between multiple attributes and change in solubility degree when the amino acid after mutation is E, and Figure 39 shows the relationship between multiple attributes and change in solubility degree when the amino acid after mutation is K.

[0171] From each of Figures 19 and 35 to 39, some differences can be seen depending on the mutation position, but in most multi-attribute combinations, there is little difference between each mutation position category, so it is thought that there is no difference in the ease with which the degree of solubility changes depending on the mutation position.

[0172] Focus on the amino acids Y, W, and F before the mutation. When looking at the diagram of amino acids before the mutation, it was found that the following mutations tend to increase the degree of solubility: mutation of amino acid Y to a negatively charged amino acid (D, E), mutation of amino acid Y to a hydrophilic, aliphatic amino acid (G, P) when the secondary structure at the mutation site is a coil, and mutation of amino acid Y to a hydrophilic, uncharged amino acid (S, T, Q, N) when the secondary structure at the mutation site is a β-sheet.

[0173] For amino acid W, it was found that the mutations most likely to increase solubility were: mutating amino acid W to a negatively charged amino acid when the secondary structure at the mutation site is a coil, mutating amino acid W to a hydrophilic, aliphatic amino acid when the secondary structure at the mutation site is a coil, and mutating amino acid W to a hydrophilic, uncharged amino acid when the secondary structure at the mutation site is a beta-sheet. For amino acid F, it was found that the mutations most likely to increase solubility were: mutating amino acid F to a negatively charged amino acid, mutating amino acid F to a hydrophilic, aliphatic amino acid when the secondary structure at the mutation site is a coil, and mutating amino acid F to a hydrophilic, positively charged amino acid (K, R, H) when the secondary structure at the mutation site is an alpha-helix. When the amino acids Y, W, and F are considered together as aromatic amino acids (Y, W, F), they have a certain degree of common tendency to mutate. For aromatic amino acids, if the secondary structure at the mutation site is a coil, the amino acid is mutated to a negatively charged amino acid, and if the secondary structure at the mutation site is a coil, the amino acid is mutated to a hydrophilic, aliphatic amino acid, which is thought to be a mutation that is likely to increase the degree of solubility.

[0174] We focused on the amino acids D, E, and K after the mutation. From Figures 19, 35 to 39 on the amino acids after the mutation, it was found that for amino acid D, the mutation of an aromatic amino acid to amino acid D, the mutation of a positive charge to amino acid D when the secondary structure at the mutation introduction position is a coil, and the mutation of a hydrophobic uncharged amino acid (C, M) to amino acid D when the secondary structure at the mutation introduction position is a coil are mutations that tend to increase the degree of solubility. For amino acid E, it was found that the mutation of an aromatic amino acid to amino acid E, the mutation of a positive charge to amino acid E when the secondary structure at the mutation introduction position is a coil, and the mutation of a hydrophobic uncharged amino acid E when the secondary structure at the mutation introduction position is a coil are mutations that tend to increase the degree of solubility. For amino acid K, it was found that the mutation of an aromatic amino acid to amino acid K, the mutation of a positive charge to amino acid K when the secondary structure at the mutation introduction position is an α-helical structure, and the mutation of a hydrophobic uncharged amino acid (C, M) to amino acid D when the secondary structure at the mutation introduction position is a coil are mutations that tend to increase the degree of solubility. Looking at the amino acids D and E as negatively charged, they share some common mutation tendencies, and negative charges are thought to be mutations that tend to increase solubility such as mutating aromatic to negative, mutating positive to negative when the secondary structure at the mutation site is a coil, and mutating hydrophobic uncharged to negative when the secondary structure at the mutation site is a coil. The amino acid K does not share some tendencies with the negatively charged amino acids. Since the amino acid K is classified as a positively charged amino acid, and the amino acids Y, W, and F share some of the same tendencies as aromatic amino acids, it was thought that there would be some differences in tendencies for each amino acid classification.

[0175] (9) From (5) of Example 2, it is considered that the degree of solubility tends to increase when the amino acids R, Y, W, and F are mutated, or when an amino acid is mutated to the amino acids D, E, and K, and tends to decrease when the amino acids D, E, and K are mutated, or when an amino acid is mutated to the amino acids R, Y, W, and F. From this, in the regression model created in the examples, an increase or decrease in the amino acids D, E, K, R, Y, W, and F may affect the degree of solubility, and it is considered that the number of amino acids R, Y, W, F, D, E, and K contained in the amino acid sequence may contribute to the ease with which the degree of solubility changes.

[0176] From (6) of Example 2, the amino acids thought to affect the degree of solubility changed depending on the secondary structure at the mutation introduction position. It is thought that the amino acids G, P, and A affect the degree of solubility when the secondary structure at the mutation introduction position is a coil, the amino acid T affects the degree of solubility when the secondary structure at the mutation introduction position is a β-sheet, and the amino acids P and I affect the degree of solubility when the secondary structure at the mutation introduction position is an α-helix. From this, it is thought that in the regression model created in the examples, a specific combination of secondary structure and amino acids may affect the degree of solubility.

[0177] From (7) of Example 2, it was considered that the tendency of the mutation affecting the degree of solubility may not change depending on the position of the mutation, and the tendency was similar to that obtained in (5) of Example 2, regardless of the position of the mutation. From this, it was considered that the regression model created in the examples does not contribute to the ease of change in the degree of solubility by the position of the amino acid mutation.

[0178] From (8) of Example 2, it is considered that the degree of solubility tends to increase when the amino acid before the mutation is an aromatic amino acid (Y, W, F), the aromatic amino acid is mutated to a negative charge (D, E) when the secondary structure at the mutation site is a coil, or the aromatic amino acid is mutated to a hydrophilic aliphatic amino acid (G, P) when the secondary structure at the mutation site is a coil, and the protein may be easily solubilized. In addition, when the amino acid after the mutation is a negatively charged amino acid and the amino acid K, the aromatic amino acid may be mutated to a negatively charged amino acid or the amino acid K, and the protein may be easily solubilized. When only the negative charge is taken into consideration, it is considered that the protein may be easily solubilized when the positive charge is mutated to a negative charge when the secondary structure at the mutation site is a coil, or when the hydrophobic uncharged (C, M) positive charge is mutated to a negative charge when the secondary structure at the mutation site is a coil. These trends are considered to be similar to the combined trends obtained from (5) to (7) in Example 2, and it is considered that the analysis using a combination of multiple attributes conducted in the example enabled the identification of trends that could not be seen from the relationship of only a single attribute. From the above, it appears that it is possible to visualize the characteristics of the complex solubility mechanisms obtained by the regression model using formal concept analysis.

[0179] Example 3: Visualization of solubility predictors using formal concept analysis (classification models) We used a classification model to predict the probability of soluble expression of proteins into which comprehensive mutations had been introduced into proteins listed in the Thermus data, and visualized the classification model by performing formal concept analysis on the predicted results.

[0180] (1) Mutation introduction prediction data In the embodiment, a portion of the amino acid sequence of a protein is comprehensively mutated, and all amino acid sequences are predicted by a classification model. As a rule of mutation, 19 patterns of mutated amino acid sequences in which a portion of the base amino acid sequence is mutated to 19 types excluding the original amino acid are created as mutated amino acid sequences, and the soluble expression probability of the mutated amino acid sequence is predicted by a classification model. However, since the first character of the amino acid sequence is basically fixed to the amino acid M, the first character of the amino acid sequence was not mutated. In the embodiment, data summarizing the prediction results may be referred to as "mutation introduction prediction data". The mutation introduction prediction data includes the name of the protein (name), the amino acid before the mutation (Before), the position of the amino acid mutation (Position), the amino acid after the mutation (After), the predicted value of the soluble expression probability (Predict sol), the change in the soluble expression probability (dif), and the like. The change in soluble expression probability was calculated as the difference between the soluble expression probability of the protein predicted before the introduction of the mutation and the soluble expression probability of the protein predicted after the introduction of the mutation.

[0181] (2) Attributes used in formal concept analysis The attributes used in formal concept analysis were the change in the probability of solubility expression, the amino acid before the mutation, the amino acid after the mutation, the properties of the amino acid before the mutation, the properties of the amino acid after the mutation, the classification of the amino acid before the mutation, the classification of the amino acid after the mutation, the position of the amino acid mutation, and the disorder region of the amino acid mutation position. The attributes of the disorder region of the amino acid mutation position have been suggested to be related to solubility by biochemistry experts. In order to create a context table to be used in formal concept analysis, the attributes must be given as binary values. The method for binarizing each attribute is explained below.

[0182] Change in probability of soluble expression The change in the probability of soluble expression was classified into one of two attributes: an increase in the probability of soluble expression (Pre-sol-up) or a decrease in the probability of soluble expression (Pre-sol-down). If the value of the change in the probability of soluble expression in the mutation introduction prediction data was greater than 0.002, it was classified as an increase in the probability of soluble expression, and if the value was less than -0.002, it was classified as a decrease in the probability of soluble expression. The value of 0.002 was decided because it is the minimum value that a computer can tolerate when performing formal concept analysis. If an attempt is made to analyze all mutations using formal concept analysis, it may place an enormous load on the computer, and the computer may not be able to perform formal concept analysis. Therefore, a minimum threshold was set for the change in the probability of soluble expression to reduce the number of protein mutations to a range that can be processed by a computer.

[0183] Amino acid before mutation The attributes are classified according to the amino acid before the mutation. For example, if the amino acid before the mutation is A, the attribute is "Amino acid before the mutation is A (Before-A)".

[0184] Amino acids after mutation The attribute is determined by the amino acid after mutation. For example, if the amino acid after mutation is A, the attribute is "Amino acid after mutation is A (After-A)".

[0185] -Properties of amino acids before mutation The amino acids before the mutation were classified into either hydrophilic or hydrophobic attributes according to Table 1. For example, if the amino acid before the mutation is hydrophilic, the attribute is "amino acid before the mutation is hydrophilic (Before-hydrophilic)."

[0186] -Properties of amino acids after mutation The amino acid after the mutation was classified into either hydrophilic or hydrophobic attributes according to Table 1. For example, if the amino acid after the mutation is hydrophilic, the attribute is "After-hydrophilic".

[0187] Classification of amino acids before mutation According to Table 2, the amino acid before the mutation was classified into five attributes: aliphatic, negatively charged, positively charged, aromatic, and uncharged. For example, if the amino acid before the mutation is hydrophilic, the attribute is "amino acid before the mutation is hydrophilic (Before-hydrophilic)".

[0188] Classification of amino acids after mutation According to Table 2, the amino acid after the mutation was classified into five attributes: aliphatic, negatively charged, positively charged, aromatic, and uncharged. For example, if the amino acid after the mutation is hydrophilic, the attribute is "After-hydrophilic".

[0189] -Amino acid mutation position The position for introducing an amino acid mutation is determined by calculating the relative mutation position p for the amino acid sequence using the above-mentioned formula 1, and classifying the attribute according to whether the value of p is in the range of 0 or more and less than 0.25 (position-1), 0.25 or more and less than 0.5 (position-2), 0.5 or more and less than 0.75 (position-3), or 0.75 or more and less than 1 (position-4).

[0190] Disorder region of amino acid mutation introduction position A disordered region is a region in the three-dimensional structure of a protein where the structure has not been determined due to large structural fluctuations. The value ranges from 0 to 1, with a lower value indicating a more stable structure.

[0191] Each attribute was classified according to whether the value of the Disorder region at the position where the amino acid mutation was introduced was less than 0.01 (Disorder-1), 0.01 or more but less than 0.1 (Disorder-2), 0.1 or more but less than 0.3 (Disorder-3), 0.3 or more but less than 0.8 (Disorder-4), or 0.8 or more (Disorder-5). For example, if the value of the Disorder region at the position where the mutation was introduced was 0.2, the attribute would be "the value of the Disorder region at the position where the mutation was introduced was 0.1 or more but less than 0.3 (Disorder-3)."

[0192] (3) Creating a context table Based on the above attributes, a context table was created to be used as an input for formal concept analysis. Formal concept analysis was performed based on the created context table.

[0193] (4) Association Rules Table 12 shows some of the association rules obtained by the formal concept analysis. In the embodiment, in order to analyze whether the soluble expression probability has changed due to the mutation, from the obtained rules, rules with attributes in the conclusion part that increase the soluble expression probability and rules with attributes that decrease the soluble expression probability were extracted. In addition, as a rule extraction condition, only rules with a reliability of 60% or more were extracted. Among the extracted rules, some of the rules with attributes that increase the soluble expression probability are shown in Table 13, and some of the rules with attributes that decrease the soluble expression probability are shown in Table 14. Based on these extracted rules, the relationship between each attribute and the change in the soluble expression probability was analyzed.

[0194] [Table 12]

[0195] [Table 13]

[0196] [Table 14]

[0197] (5) Analysis of the relationship between amino acid mutations to (other) amino acids and changes in the probability of soluble expression The relationship between mutations from one amino acid to another and the change in the probability of soluble expression was analyzed separately for cases where the probability of soluble expression increases and cases where the probability of soluble expression decreases.

[0198] -About the rules that increase the probability of solubility We analyzed the relationship between the introduction of a mutation from a specific amino acid to (another) amino acid and the change in the probability of soluble expression when the probability of soluble expression increases. First, from the rules in Table 13, we extracted rules whose premise has the attributes of the amino acid before and after the mutation, and whose conclusion has the attribute of increasing the probability of soluble expression. We then created a table with the amino acids before the mutation as rows and the amino acids after the mutation as columns, and analyzed which amino acids were mutated to which amino acids based on the extracted rules. The mutations from amino acids to amino acids are summarized in Figure 20, and a heat map is created based on the reliability of the rules (Figure 20), to analyze which mutations are likely to affect the probability of soluble expression.

[0199] From Figure 20, it was found that the amino acids R, Y, W, and Q (glutamine) appeared frequently before the introduction of mutations, and the amino acids G, V, T, and C appeared frequently after the introduction of mutations, as mutations that increased the probability of soluble expression. The color distribution of the heat map in the figure shows that the areas where the amino acids Y, W, and Q have been mutated to amino acids G, V, K, and T are darker in color. From the above, it was thought that the probability of soluble expression would be likely to increase if any of the amino acids Y, W, or Q were mutated to any of the amino acids G, V, K, or T.

[0200] -About the rules that reduce the probability of soluble manifestation The relationship between the introduction of a mutation from a specific amino acid to (another) amino acid and the change in the soluble expression probability in the case where the soluble expression probability is decreased was analyzed. First, from the rules in Table 14, rules were extracted that have the attributes of the amino acid before the mutation and the amino acid after the mutation in the premise, and the attribute that the soluble expression probability is decreased in the conclusion. Then, a table was created in which the amino acids before the mutation are rows and the amino acids after the mutation are columns, and which amino acids were mutated to which amino acids based on the extracted rules was analyzed. Based on the analysis, Figure 21 was created, and a heat map was created based on the reliability of the rules (Figure 21), and which mutations are likely to affect the soluble expression probability was analyzed.

[0201] From FIG. 21, it was found that the amino acids G, K, and T appeared frequently before the introduction of mutations, and the amino acids L, Y, and Q appeared frequently after the introduction of mutations, which reduced the probability of soluble expression.

[0202] The color distribution of the heat map in the figure shows that the areas where the amino acids G, V, K, and T have been mutated to amino acids Y, W, and Q are darker in color. From the above, it was considered that the probability of solubility expression is likely to decrease when any of the amino acids G, V, K, and T is mutated to any of the amino acids Y, W, and Q.

[0203] (6) Analysis of the relationship between the disorder region of the amino acid mutation introduction site and the change in the probability of soluble expression The relationship between mutations from one amino acid to another and the change in the probability of soluble expression was analyzed separately for cases where the probability of soluble expression increases and cases where the probability of soluble expression decreases.

[0204] -About the rules that increase the probability of solubility We analyzed the relationship between the mutation of a specific amino acid to another amino acid for each disorder region of the amino acid mutation introduction position and the change in the probability of soluble expression when the probability of soluble expression increases. First, from the rules in Table 13, we extracted rules that have the amino acid before and after the mutation introduction, and the attribute of the disorder region of the amino acid mutation introduction position in the premise, and the attribute that increases the probability of soluble expression in the conclusion. We then created a table with the amino acid before the mutation introduction as rows and the amino acid after the mutation introduction as columns, and analyzed which amino acid was mutated to which amino acid in which category of the disorder region of the amino acid mutation introduction position based on the extracted rules. In addition, we focused on the attributes of Disorder-1 and Disorder-5, which are in separate categories and have a clear difference in the values ​​of the disorder region. Figure 22 summarizes the amino acid mutations from Disorder-1 to other amino acids in the disorder region where the mutations were introduced, and Figure 23 summarizes the amino acid mutations from Disorder-5 to (other) amino acids in the disorder region where the mutations were introduced. We then created a heat map based on the reliability of the rules and analyzed which mutations were most likely to affect the probability of soluble expression.

[0205] From Figures 22 and 23, it was found that when the Disorder region at the mutation introduction position is Disorder-1, the amino acids Y, W, Q, and M appeared frequently as amino acids before the mutation introduction, and the amino acids G, V, and T appeared frequently as amino acids after the mutation introduction. When the Disorder region at the mutation introduction position is Disorder-5, the amino acids L, I, R, S, and Q appeared frequently as amino acids before the mutation introduction, and the amino acids V, F, T, and C appeared frequently as amino acids after the mutation introduction. It was found that there were common parts and different parts that appeared as mutations that increased the probability of soluble expression for each Disorder region at the mutation introduction position, and that there were differences in the mutations that are likely to increase the probability of soluble expression for each Disorder region.

[0206] From the color distribution of the heat maps in Figures 22 and 23, when the disorder region at the mutation introduction position is Disorder-1, it was found that the parts where the amino acids G, V, I, F, T, and C are mutated to amino acids A, L, E, Y, W, Q, and M are darker. When the disorder region at the mutation introduction position is Disorder-5, it was found that the parts where the amino acids L, R, and Q are mutated are generally colored, and among them, the parts where the amino acids L, R, and Q are mutated to amino acids G, V, K, F, and T are darker. From the above, it was thought that in both disorder regions, mutating the amino acid Q and mutating the amino acid to any of the amino acids G, V, or T would likely increase the probability of soluble expression. For each disorder region, when the disorder region at the mutation introduction position is Disorder-1, it was thought that mutating any of the amino acids Y or W to any of the amino acids G, V, or T would likely increase the probability of soluble expression. When the disorder region at the mutation introduction position is Disorder-5, it was thought that mutating any of the amino acids R or L would likely increase the probability of soluble expression.

[0207] -About the rules that reduce the probability of soluble manifestation The relationship between the introduction of a specific amino acid to an amino acid for each Disorder region of the amino acid mutation introduction position and the change in the soluble expression probability in the case where the soluble expression probability decreases was analyzed. First, from the rules in Table 13, rules were extracted that have the amino acid before the mutation introduction and the amino acid after the mutation introduction, and the attribute of the Disorder region of the amino acid mutation introduction position in the premise part, and the attribute of the soluble expression probability decrease in the conclusion part. Then, a table was created with the amino acid before the mutation introduction as rows and the amino acid after the mutation introduction as columns, and based on the extracted rules, it was analyzed which amino acid was mutated to which amino acid in which category of the Disorder region of the amino acid mutation introduction position. In addition, for the attributes of the Disorder region of the mutation introduction position, the attributes of Disorder-1 and Disorder-5, which are separated in category and the difference in the value of the Disorder region was clearly seen. Figure 24 summarizes the introduction of amino acid to amino acid mutation in the Disorder region of the mutation introduction position of Disorder-1, and Figure 25 summarizes the introduction of amino acid to amino acid mutation in the Disorder region of the mutation introduction position of Disorder-5. In addition, heat maps were created based on the reliability of the rules (FIGS. 24 and 25) to analyze which mutations are likely to affect the probability of soluble expression.

[0208] From Figures 24 and 25, when the Disorder region at the mutation introduction position is Disorder-1, the amino acids G, V, I, F, and C appeared frequently as amino acids before the mutation introduction, and the amino acids E, Y, W, Q, and M appeared frequently as amino acids after the mutation introduction. When the Disorder region at the mutation introduction position is Disorder-5, the amino acids G, V, K, F, and T appeared frequently as amino acids before the mutation introduction, and the amino acids A, L, R, Y, and Q appeared frequently as amino acids after the mutation introduction. It was found that there are parts that commonly appear as mutations that increase the probability of soluble expression and parts that differ for each Disorder region at the mutation introduction position, and that there are differences in the mutations that are likely to increase the probability of soluble expression for each Disorder region.

[0209] From the color distribution of the heat maps in Figures 24 and 25, when the disorder region at the mutation introduction position is Disorder-1, the parts where the amino acids G, V, and T have been mutated to amino acids L, Y, W, and Q are darker. When the disorder region at the mutation introduction position is Disorder-5, the parts where the amino acids G, V, and T have been mutated are generally colored, and among them, the parts where the amino acids G, V, K, and T have been mutated to amino acids A, L, R, Y, and Q are darker.

[0210] From the above, it was thought that in both disorder regions, mutations from amino acids G, V, or T to amino acids L, Y, or Q were likely to decrease the probability of soluble expression. For each disorder region, when the disorder region at the mutation site was Disorder-1, it was thought that the introduction of a mutation from amino acid G, V, or T to amino acid W was likely to decrease the probability of soluble expression. When the disorder region at the mutation site was Disorder-5, it was thought that the mutation from amino acid G, V, or T to amino acid R was likely to decrease the probability of soluble expression.

[0211] (7) Relationship between the position of amino acid mutation and the change in the probability of solubility expression The relationship between the position of amino acid mutation and the change in the probability of soluble expression was analyzed for two cases: when the probability of soluble expression increased and when the probability of soluble expression decreased.

[0212] -About the rules that increase the probability of solubility The relationship between the mutation of a specific amino acid to an amino acid for each amino acid mutation introduction position and the change in the soluble expression probability in the case where the soluble expression probability increases was analyzed. First, from the rules in Table 12, rules were extracted that have the attributes of the amino acid before the mutation introduction, the amino acid after the mutation introduction, and the amino acid mutation introduction position in the premise, and the attribute that increases the soluble expression probability in the conclusion. Then, a table was created with the amino acids before the mutation introduction as rows and the amino acids after the mutation introduction as columns, and an analysis was performed on which amino acid was mutated and from which amino acid based on the extracted rules. The mutation of an amino acid to (another) amino acid at the mutation introduction position-1 is summarized in FIG. 26, the mutation of an amino acid to (another) amino acid at the mutation introduction position-2 is summarized in FIG. 27, the mutation of an amino acid to (another) amino acid at the mutation introduction position-1 is summarized in FIG. 28, and the mutation of an amino acid to (another) amino acid at the mutation introduction position-1 is summarized in FIG. 29. In addition, heat maps were created based on the reliability of the rules (FIGS. 26 to 29) to analyze which mutations are likely to affect the probability of soluble expression.

[0213] From each of Figures 26 to 29, it was found that when the mutation introduction position was position-1, the amino acids E, Y, W, Q, and N appeared frequently as amino acids before the mutation introduction, and the amino acids A, K, S, and T appeared frequently as amino acids after the mutation introduction. When the mutation introduction position was position-2, the amino acids Y, W, Q, and N appeared frequently as amino acids before the mutation introduction, and the amino acids K and T appeared frequently as amino acids after the mutation introduction. When the mutation introduction position was position-3, the amino acid W appeared frequently as amino acid before the mutation introduction, and the amino acids G, V, I, and T appeared frequently as amino acids after the mutation introduction. When the mutation introduction position was position-4, the amino acids L, R, Y, and Q appeared frequently as amino acids before the mutation introduction, and the amino acids G, V, and T appeared frequently as amino acids after the mutation introduction.

[0214] From the color distribution of the heat maps in Figures 26 to 29, when the position of mutation introduction was position-1, the color of the parts where amino acids Y, W, and Q were mutated was dark overall. When the position of mutation introduction was position-2, the color of the parts where amino acids Y and W were mutated and the color of the parts where amino acids K and T were mutated was dark. When the position of mutation introduction was position-3, the color of the parts where amino acids Y and W were mutated to amino acids G, V, I, and T was dark. When the position of mutation introduction was position-4, the color of the parts where amino acids A, L, Y, and Q were mutated to amino acids G, V, and T was dark.

[0215] From the above, it was considered that the probability of solubility expression is likely to increase when the amino acid is mutated to the amino acid T at any mutation introduction position. For each mutation introduction position, there is a rough difference between the mutation introduction positions of position-1 and position-2, i.e., the mutation introduction positions are in the first half of the amino acid sequence, and the mutation introduction positions of position-3 and position-4, i.e., the mutation introduction positions are in the second half of the amino acid sequence. In the first half of the amino acid sequence, it was considered that the probability of solubility expression is likely to increase when the amino acid is mutated to either amino acid Y or W and the amino acid is mutated to amino acid K. In the second half of the amino acid sequence, it was considered that the probability of solubility expression is likely to increase when the amino acid is mutated to either amino acid G or V. In addition, it was considered that the probability of solubility expression is likely to increase when the mutation introduction position is position-1, i.e., near the N-terminus of the amino acid sequence, and when the mutation introduction position is position-4, i.e., the mutation of amino acid Q near the C-terminus of the amino acid sequence, the mutation of the amino acid near the N-terminus to either amino acid A or S, and the mutation of amino acid L or R near the C-terminus.

[0216] -About the rules that reduce the probability of soluble manifestation The relationship between the amino acid-to-amino acid mutation for each amino acid mutation position and the change in the soluble expression probability in the case where the soluble expression probability is decreased was analyzed. First, from the rules in Table 14, rules were extracted that have the attributes of the amino acid before the mutation introduction, the amino acid after the mutation introduction, and the amino acid mutation introduction position in the premise, and the attribute of the soluble expression probability being decreased in the conclusion. Then, a table was created with the amino acids before the mutation introduction as rows and the amino acids after the mutation introduction as columns, and it was analyzed which amino acids were mutated to which amino acids at which positions based on the extracted rules. The amino acid-to-amino acid mutations when the mutation introduction position is position-1 are summarized in FIG. 30, the amino acid-to-amino acid mutations when the mutation introduction position is position-2 are summarized in FIG. 31, the amino acid-to-amino acid mutations when the mutation introduction position is position-3 are summarized in FIG. 32, and the amino acid-to-amino acid mutations when the mutation introduction position is position-4 are summarized in FIG. 33. In addition, heat maps were created based on the reliability of the rules (FIGS. 30 to 33), and it was analyzed which mutations are likely to affect the soluble expression probability.

[0217] From each of Figures 30 to 33, it was found that when the position of mutation introduction was position-1, the amino acids K, S, A, and T appeared frequently as amino acids before the mutation introduction, and the amino acids E, Y, W, Q, and M appeared frequently as amino acids after the mutation introduction. When the position of mutation introduction was position-2, the amino acids K, T, and C appeared frequently as amino acids before the mutation introduction, and the amino acids E, Y, W, N, and M appeared frequently as amino acids after the mutation introduction. When the position of mutation introduction was position-3, the amino acids G, K, and T appeared frequently as amino acids before the mutation introduction, and the amino acids Y and W appeared frequently as amino acids after the mutation introduction. When the position of mutation introduction was position-4, the amino acids G, V, and T appeared frequently as amino acids before the mutation introduction, and the amino acids A, L, R, Y, and Q appeared frequently as amino acids after the mutation introduction.

[0218] From the color distribution of the heat maps in Figures 30 to 33, when the position of mutation introduction was position-1, the color of the parts where amino acids S and A were mutated and the amino acids E, Y, W, and Q were mutated was dark overall. When the position of mutation introduction was position-2, the color of the parts where amino acids K, T, and C were mutated and the amino acids E, Y, W, and M were mutated was dark. When the position of mutation introduction was position-3, the color of the parts where amino acids G, V, K, and T were mutated to amino acids Y and W was dark. When the position of mutation introduction was position-4, the color of the parts where amino acids G, V, and T were mutated to amino acids L, R, Y, and Q was dark.

[0219] From the above, it was considered that the probability of solubility expression is likely to decrease when an amino acid is mutated to amino acid Y at any mutation introduction position. For each mutation introduction position, there is a rough difference between the first half of the amino acid sequence and the second half of the amino acid sequence. When it is in the first half of the amino acid sequence, it was considered that the probability of solubility expression is likely to decrease when an amino acid K is mutated and when an amino acid is mutated to amino acid E, Y, W, or M. When it is in the second half of the amino acid sequence, it was considered that the probability of solubility expression is likely to increase when an amino acid G is mutated. In addition, it was considered that the probability of solubility expression is likely to decrease when an amino acid near the N-terminus and near the C-terminus of the amino acid sequence is mutated to amino acid Q, when an amino acid near the N-terminus is mutated to amino acid A or S, and when an amino acid near the C-terminus is mutated to amino acid L or R.

[0220] (8) Relationship between multiple attributes and change in probability of soluble expression Focusing on amino acid mutations that may affect the change in soluble expression probability in amino acid-to-amino acid mutations, factors that affect the soluble expression probability were explored by analyzing the relationship between multiple attributes and changes in soluble expression probability. All attributes set in (2) of Example 3 were used as multiple attributes. However, for the Disorder region at the mutation introduction position, only the attributes of Disorder-1 and Disorder-5, which are separated and show differences in the values ​​of the Disorder region, were used as attributes. In addition, since the results so far show that there is a symmetry between mutations that increase the soluble expression probability and mutations that decrease it, we focused only on mutations that increase the soluble expression probability and explored the tendency of mutations that tend to increase the soluble expression probability of proteins.

[0221] From (5) of Example 3, as an amino acid mutation that is thought to affect the change in the soluble expression probability and to easily increase the soluble expression probability, attention was paid to the amino acid mutations in which the amino acid before the mutation is Y, the amino acid before the mutation is W, the amino acid before the mutation is F, the amino acid after the mutation is D, and the amino acid after the mutation is K. The amino acid before or after the mutation is fixed as an attribute, and a rule was extracted in which the premise part contains a pair of the fixed attribute and a specific attribute, and the conclusion part has an attribute that increases the soluble expression probability. A graph was created from the extracted rules with the reliability on the vertical axis and the combination of each attribute on the horizontal axis, and mutations that are likely to change the soluble expression probability in the fixed attribute were searched for.

[0222] Figure 34 shows the relationship between multiple attributes and the change in soluble expression probability when the amino acid before mutation is Y, Figure 40 shows the relationship between multiple attributes and the change in soluble expression probability when the amino acid before mutation is Q, Figure 41 shows the relationship between multiple attributes and the change in soluble expression probability when the amino acid after mutation is G, Figure 42 shows the relationship between multiple attributes and the change in soluble expression probability when the amino acid after mutation is V, and Figure 43 shows the relationship between multiple attributes and the change in soluble expression probability when the amino acid after mutation is T.

[0223] From each figure, it was found that in most of the combinations of multiple attributes, there were differences in each classification of the mutation introduction position, and therefore there were differences in the ease with which the probability of solubility expression changed depending on the mutation introduction position.

[0224] We focused on amino acids Y and Q before the mutation. From Figures 40 to 43 regarding the amino acids before the mutation, it was found that, for amino acid Y, when the mutation position is position-2 and the Disorder region at the mutation position is Disorder-5, the amino acid Y is mutated to a hydrophobic, aliphatic amino acid (A, V, L, I), when the mutation position is position-4 and the Disorder region at the mutation position is Disorder-5, the amino acid Y is mutated to a positively charged amino acid (K, R, H), and when the mutation position is position-4 and the Disorder region at the mutation position is Disorder-5, the amino acid Y is mutated to a hydrophobic, uncharged amino acid (C, M), etc., are mutations that are likely to increase the probability of soluble expression. For amino acid Q, it was found that the probability of solubility expression is likely to increase when the amino acid Q is mutated to a hydrophilic aliphatic amino acid (G, P) when the mutation introduction position is position-4, when the mutation introduction position is position-1 or position-4 and the Disorder region of the mutation introduction position is Disorder-5, the amino acid Y is mutated to a positively charged amino acid, and when the mutation introduction position is position-1, the amino acid Q is mutated to a hydrophobic aliphatic amino acid. It was also found that the probability of solubility expression is likely to increase when the amino acid Q at the mutation introduction position is position-1 or position-4 is mutated to another amino acid. This tendency was similar to that obtained in (7) of Example 3.

[0225] We will focus on the amino acids G, V, and T after mutation. From Figures 40 to 43 regarding the amino acids after mutation, it was found that for amino acid G, mutations that tend to increase the probability of soluble expression include a mutation from a hydrophilic and aromatic amino acid (Y, W) to amino acid G when the Disorder region at the mutation introduction position is Disorder-1, a mutation from an uncharged amino acid (S, T, Q, N, C, M) to amino acid G when the mutation introduction position is position-2, and a mutation from an aliphatic amino acid (G, P, A, V, L, I) to amino acid G when the mutation introduction position is position-4 and the Disorder region at the mutation introduction position is Disorder-5. For amino acid V, mutations that are likely to increase the probability of solubility include a hydrophilic, aromatic amino acid to amino acid V when the disorder region at the mutation introduction site is Disorder-1, a hydrophilic, uncharged amino acid (S, T, Q, N) to amino acid V when the mutation introduction site is position-1 and the disorder region at the mutation introduction site is Disorder-5, and a hydrophobic, aliphatic amino acid when the mutation introduction site is position-4. For amino acid T, mutations that are likely to increase the probability of solubility include a hydrophilic, aromatic amino acid to amino acid V when the disorder region at the mutation introduction site is Disorder-1, and a hydrophobic, aliphatic amino acid when the mutation introduction site is position-4. Looking at amino acids G, V, and T together, there is a common tendency for mutations, and it is thought that a mutation from a hydrophilic, aromatic amino acid to any of the amino acids G, V, and T when the disorder region at the mutation introduction site is Disorder-1 is likely to increase the probability of solubility. In addition, when the disorder region at the mutation introduction site was Disorder-5, there was no apparent tendency for the probability of soluble expression to increase when a hydrophilic, aromatic amino acid was mutated to any of the amino acids G, V, or T.

[0226] From the above, we believe that the protein is more easily soluble when a hydrophilic, aromatic amino acid with a small value in the disorder region is mutated to any of the amino acids G, V, and T. Furthermore, we believe that the tendency of this mutation does not change depending on the position of the mutation.

[0227] (9) From (5) of Example 3, it is considered that the probability of solubility expression is likely to increase when any of the amino acids Y, W, and Q is mutated to any of the amino acids G, V, K, and T, and the probability of solubility expression is likely to decrease when any of the amino acids G, V, K, and T is mutated to any of the amino acids Y, W, and Q. From this, it is considered that the protein is easily soluble when any of the amino acids Y, W, and Q is mutated to any of the amino acids G, V, K, and T, and in the classification model created in the examples, it is considered that the amino acids Y, W, and Q may contribute to the insolubilization of the protein, and the amino acids G, V, K, and T may contribute to the solubility of the protein.

[0228] From (6) of Example 3, it is considered that the amino acid that is thought to affect the soluble expression probability may change depending on the Disorder region at the mutation introduction position. When the Disorder region at the mutation introduction position is Disorder-1, a mutation from either amino acid Y or W to either amino acid G, V, or T, and when the Disorder region at the mutation introduction position is Disorder-5, a mutation from either amino acid R or L is likely to increase the soluble expression probability. From this, in the classification model created in the examples, it is possible that a specific amino acid affects the structure formation of the protein, which in turn affects the soluble expression probability. In addition, since information about the Disorder region is not provided for learning, the model may have learned features related to the structure formation of the protein.

[0229] From (7) of Example 3, it is considered that the tendency of mutations affecting the soluble expression probability depending on the mutation introduction position may change depending on the mutation introduction position. There is a difference in tendency between the mutation introduction position in the first half of the amino acid sequence and the mutation introduction position in the second half of the amino acid sequence, and it is considered that the amino acids Y, W, and K affect the soluble expression probability when the mutation introduction position is in the first half of the amino acid sequence, and the amino acids G and V affect the soluble expression probability when the mutation introduction position is in the second half of the amino acid sequence. In addition, it is considered that the amino acid Q near the N-terminus of the amino acid sequence and the amino acid A and S near the N-terminus, and the amino acids A, L, and R near the C-terminus affect the soluble expression probability. From this, it is considered that in the classification model created in the examples, a specific amino acid contributes to the ease of change in the soluble expression probability at a specific mutation introduction position.

[0230] From (8) of Example 3, it is considered that the tendency for the probability of solubility expression to increase is that when the Disorder region at the mutation introduction position is Disorder-1, the protein is easily soluble when a hydrophilic and aromatic amino acid is mutated to any of the amino acids G, V, and T. Furthermore, there is no difference in this tendency depending on the mutation introduction position. From this, it is considered that this tendency is similar to the tendency obtained by combining the trends obtained from (5) to (7) of Example 3, and it is considered that the analysis of the combination of multiple attributes performed this time was able to confirm a tendency that could not be seen from the relationship of only a single attribute.

[0231] From the above, it appears that it is possible to visualize the characteristics of the complex soluble mechanisms obtained by the classification model using formal concept analysis.

[0232] In the examples, machine learning was performed based on protein structural information, and a regression model for predicting the degree of protein solubility and a classification model for predicting protein solubility were constructed. The regression model and classification model were then used to virtually and comprehensively predict protein mutation introduction, and mutation introduction prediction data was generated. Formal concept analysis was applied to the data to visualize the trends obtained by the regression model and classification model through machine learning, and the solubility mechanism of the protein was elucidated.

[0233] The tendency obtained by analyzing the regression model using formal concept analysis is that proteins are more easily soluble when aromatic amino acids (Y, W, F) are mutated to other amino acids or when amino acids are mutated to negatively charged amino acids (D, E). In addition, since the tendency of mutation introduction does not change depending on the position of the amino acid mutation, it is considered that there is no difference in the tendency of proteins to be more soluble depending on the position of the mutation introduction. From this, it is considered that the regression model is influenced by the number of aromatic amino acids and negatively charged amino acids in the solubility. On the other hand, the tendency obtained by analyzing the classification model using formal concept analysis is that proteins are more easily soluble when the value of the disorder region is low and hydrophilic and aromatic amino acids are mutated to one of the amino acids G, V, or T. Since the classification model created did not receive any information about the disorder region for learning, it is considered that the classification model has acquired some characteristics regarding the structure formation of proteins.

[0234] Example 4: Location of mutation sites predicted by the learning model The amino acid sequence of the protein derived from the extreme thermophile Thermus thermophilus used in this example was input into a computer, and the solubility probability was calculated by replacing amino acids one by one, and the increase was plotted. As a result, it was found that there was a tendency for the solubility probability to be located relatively late in the sequence.

[0235] Figure 44 shows where the sites affected by amino acid substitutions are located on the sequence, with the horizontal axis showing the number and the vertical axis showing the position, with the N-terminus at the bottom and the C-terminus at the top. From Figure 44, it can be seen that the amino acids involved in soluble expression are located relatively late on the sequence.

[0236] Example 5: Amino acid sites involved in solubility probability We investigated the types of amino acid residues in sequences with increased solubility probability using machine learning. Substitution with Arg, Tyr, and Gln tended to increase the solubility probability. On the other hand, substitution with Gly, Val, Lys, Thr, and Cys tended to increase the solubility probability.

[0237] In Figure 45, the top shows the amino acids before substitution that increase solubility, and the bottom shows the amino acids after substitution that decrease solubility. In Figure 45, R and Y are only seen in the upper graph, and the bars get higher toward the end, so it is thought that substituting these amino acids is involved in increasing solubility. If there are amino acids that are rarely seen in the upper graph but are seen in the lower graph, it is effective to return to those amino acids after substitution, as calculated by the constructed trained protein solubility prediction model. G, V, T, C, and K surrounded by solid lines correspond to the effective ones mentioned above, and it is suggested that any amino acid can be effectively substituted with these amino acids. In addition, it is suggested that R, Y, and Q surrounded by dashed lines will increase solubility expression by substituting them with other amino acids.

[0238] Example 6: Examination of the effect of substitutions on GVTC in actual mutagenesis Eight proteins (enzymes whose activity can be easily measured by colorimetry or absorbance) derived from the extreme thermophile Thermus thermophilus (Table 15) were randomly selected, and mutations were introduced into the top 20 amino acid residues involved in solubility, a total of 160, using the amino acid sequences that make up each enzyme, to confirm whether activity increased or not. In Table 15, the solubility probability of introducing the wild-type enzyme sequence of each protein is shown in % at the end of the parentheses.

[0239] [Table 15]

[0240] Of the top 20 mutation sites found to be effective from the trained protein solubility prediction model constructed in Example 1, only 52 types (approximately 33%) showed increased activity, and the rest were below the detection limit or showed no increase in activity. However, when focusing on the mutant enzymes mutated to the amino acid GVTC presented above, approximately 73% (27 types) showed increased activity (Table 16).

[0241] [Table 16]

[0242] As can be seen from the above results, with regard to elements (amino acids) such as GVTC found by visualizing the trained protein prediction model constructed in Example 1 as shown in Examples 2 and 3, it can be said that the model correctly identifies residues that are biochemically effective.

[0243] Example 7: Relationships with multiple attributes We investigated the properties and positions of amino acids in sequences with increased solubility probability by machine learning. As a result, we found that the solubility probability tends to increase by replacing aromatic amino acids with Gly, Val, and Thr. They also tend to be found frequently in disordered regions of the protein structure. As shown in Figures 46 and 47, it was found that replacement of aromatic amino acids in the disorder region with GVT tended to increase soluble expression.

[0244] The results of Examples 1 to 3 and 4 to 7 are as described above. The attributes of the protein that increases the solubility of the protein obtained by the identification method of the present invention, which can be understood from the results of Examples 1 to 7, are particularly as follows: (i) Substitution of at least one amino acid selected from the group consisting of arginine, tyrosine, and glutamine among amino acids constituting a protein with an amino acid other than arginine, tyrosine, and glutamine; (ii) Substitution of amino acids other than glycine, valine, threonine, cysteine, and lysine constituting a protein with glycine, valine, threonine, cysteine, or lysine; and (iii) Substitution of aromatic amino acids in the disordered region of the protein with glycine, valine, or threonine etc.

[0245] Moreover, the amino acids to be substituted in the above (i) to (iii) are preferably amino acids in the latter half of the amino acid sequence of the protein, as will be understood from Examples 4 to 7. Moreover, it is also preferable that aromatic amino acids in the disordered region of the protein are hydrophilic. [Industrial Applicability]

[0246] According to the present invention, since the attributes of a protein involved in the solubility of the protein can be identified, it is useful because it is possible to design insoluble or soluble recombinant proteins based on the identified attributes. In particular, for proteins that form precipitates, it is useful because it is possible to synthesize them as soluble proteins without forming precipitates based on the identified attributes, which can improve the production amount of the protein. Furthermore, it is useful because it is possible to use enzymes derived from animals and plants, which are considered to be difficult to put into practical use because they are difficult to produce in recombinant microorganisms, by utilizing the identified attributes. Furthermore, it is useful because it can be used to produce biopharmaceuticals, antibodies, etc. other than enzymes (development into new pharmaceuticals, biosimilars, etc.).

Claims

1. 1. A method for identifying a protein attribute involved in protein solubility, comprising: setting attributes of the protein; Calculating attributes related to the degree of solubility based on output data obtained from the input data by the trained protein solubility prediction model; Preamble: an attribute of the protein; Confidence level: 60% or higher, and Conclusion: Attributes related to the degree of solubility Extracting rules that fit the A method comprising:

2. The method according to claim 1, wherein the input data is a data set of an amino acid sequence of a wild-type protein and a sequence of the wild-type protein in which one to several amino acids have been mutated.

3. 3. The method according to claim 1 or 2, wherein the attribute of the protein is one or more selected from the group consisting of the amino acid before the mutation, the properties of the amino acid before the mutation, the classification of the amino acid before the mutation, the amino acid after the mutation, the properties of the amino acid after the mutation, the classification of the amino acid after the mutation, the position of the amino acid mutation, and the secondary structure of the position of the amino acid mutation.

4. The method of claim 3 , wherein the amino acid has a hydrophilic or hydrophobic nature.

5. 5. The method of claim 3 or 4, wherein the amino acid classes are aliphatic, negatively charged, positively charged, aromatic, or uncharged.

6. The method according to any one of claims 3 to 5, wherein the secondary structure at the position where the amino acid is mutated is an alpha helix, a beta sheet, or a coil.

7. The method according to any one of claims 1 to 6, wherein the attribute relating to the degree of solubility is based on the degree of solubility of a wild-type protein and a change in the degree of solubility of a protein in which one to several amino acids of the wild-type protein have been mutated.

8. A method for producing a protein with improved solubility, comprising modifying the protein so that it has an attribute identified by the method of any one of claims 1 to 7.

9. 1. A method for producing a protein having improved solubility, comprising: A step of substituting at least one amino acid selected from the group consisting of arginine, tyrosine, and glutamine among amino acids constituting a protein with an amino acid other than arginine, tyrosine, and glutamine; A step of substituting amino acids other than glycine, valine, threonine, cysteine, and lysine constituting a protein with glycine, valine, threonine, cysteine, or lysine; and Substituting aromatic amino acids in a disordered region of a protein with glycine, valine, or threonine; The method comprises one or more steps selected from the group consisting of:

10. The method according to claim 9 , wherein the amino acids constituting the protein are amino acids in the latter half of the amino acid sequence of the protein.

11. The method of claim 9 or 10, wherein the aromatic amino acids in the disorder region of the protein are hydrophilic.

12. A method for improving the solubility of a protein, comprising: A step of substituting at least one amino acid selected from the group consisting of arginine, tyrosine, and glutamine among amino acids constituting a protein with an amino acid other than arginine, tyrosine, and glutamine; A step of substituting amino acids other than glycine, valine, threonine, cysteine, and lysine constituting a protein with glycine, valine, threonine, cysteine, or lysine; and Substituting aromatic amino acids in a disordered region of a protein with glycine, valine, or threonine; The method comprises one or more steps selected from the group consisting of:

Citation Information

Patent Citations

  • Active-form mutant enzyme production method, new active-form mutant enzyme, and solubilized mutant protein production method

    WO2016199898A1