KR102645477B1 - Prediction method of immunogenic determinant and immunogen binding site - Google Patents

Abstract

The present invention relates to a method for predicting an immunogen-determining portion and a method for predicting an immunogen-binding portion that can more reliably predict an immunogen-determining portion and/or an immunogen-binding portion that specifically binds thereto in a protein or cell that causes an immune response.

Description

Method for predicting immunogen determinant and immunogen binding site {PREDICTION METHOD OF IMMUNOGENIC DETERMINANT AND IMMUNOGEN BINDING SITE}

The present invention relates to a method for predicting an immunogen-determining portion and a method for predicting an immunogen-binding portion that can more reliably predict an immunogen-determining portion and/or an immunogen-binding portion that specifically binds thereto in a protein or cell that causes an immune response.

The immune system refers to a defense system that defends and neutralizes infectious pathogens or harmful proteins in the body, and the immune response is an immune response in which immune proteins such as antibodies or immune cells such as T cells and B cells attack immunogens such as harmful proteins. It can refer to any reaction related to the immune system that specifically recognizes or binds.

In this immune response, within an immunogen (e.g., antigen) such as an infectious pathogen or harmful protein, an amino acid residue (or sequence) specifically recognized by the immune protein is represented by an epitope. It can be defined as an immunogenic determinant. In addition, amino acid residues in immune proteins, such as antibodies that specifically recognize and bind to the immunogen determinant within the immunogen, can be defined as an immunogen binding site, represented by an antigen binding site (paratope).

In addition, the antigenic determinants can be divided into linear epitopes and structural epitopes, depending on their shape and mode of action with the antigen-binding portion. Among them, the linear antigenic determinant is composed of a continuous linear amino acid sequence and can bind to the one-dimensional structure of the antigen binding site, while the three-dimensional antigenic determinant is composed of a discontinuous amino acid sequence reflecting the three-dimensional protein structure and can bind to the antigen. It can be combined with the three-dimensional structure of wealth.

Recently, interest in immune-based treatments or treatment methods based on immune responses, etc., has greatly increased, and research and development on these has also been very actively conducted. Since these immune-based therapeutics are developed based on the specific binding and immune response between the immunogenic determinant, such as the antigenic determinant, and the immunogen binding portion, such as the antigen binding portion, etc., the immunogenic determinant within the immunogen is predicted or confirmed, and the immune protein Predicting or confirming immunogen binding sites within the immune system has emerged as a very important issue in the development of effective immune-based therapeutics.

However, because it is very difficult to experimentally confirm the immunogen determinant or immunogen binding site based on the in vivo immune response, several methods have previously been proposed to predict such immunogen determinant or immunogen binding site or their candidate groups. And research on this is continuing.

However, it is true that the methods for predicting the immunogen determinant and immunogen binding site known to date have not been sufficiently reliable. This is mainly due to insufficient available data on the sequence or structure of immunogens and immune proteins. In particular, in order to reliably predict the structural epitope and the three-dimensional antigen binding site that binds to it, a deep understanding of the three-dimensional folding structure of target proteins such as immunogens and immune proteins and a lot of data were required. However, securing such data is insufficient, which makes prediction of the three-dimensional antigenic determinant and three-dimensional antigen binding site more difficult.

Due to these problems, the development of methods or related technologies that can more reliably predict the immunogen determinant and/or immunogen binding site continues, even if it is based on insufficient data on proteins, especially the three-dimensional folding structure of the protein. is being requested.

Accordingly, one embodiment of the invention provides a method for predicting an immunogen prediction portion that can more reliably predict the presence, location, or sequence of an immunogen determinant within an immunogen that is specifically recognized or bound to by an immune protein or immune cell and causes an immune response. It is done.

In addition, another embodiment of the invention provides a method for predicting an immunogen binding site that can more reliably predict the presence, location, or sequence of an immunogen binding site that specifically binds to an immunogen within an immune protein or immune cell and causes an immune response. will be.

Accordingly, according to one embodiment of the invention, based on the first database on proteins, a two-way artificial neural network model is learned while masking each tokenized unit sequence included in each protein, and the unit of the protein is Pre-learning the function and arrangement of each sequence;

Based on a second database of immunogenic and non-immunogenic polypeptides classified according to whether immune proteins or immune cells specifically recognize or bind to the immune response, the amount is classified by the tokenized unit sequence contained in each polypeptide. Proceeding with directional artificial neural network learning to learn information on immunogenic determinants that cause an immune response for each unit sequence of the polypeptide; and

A method for predicting immunogen determinants is provided, including the step of predicting immunogen determinants of a target immunogen or target protein using a learned artificial neural network.

In this prediction method of one embodiment, unit sequences of immunogenic and non-immunogenic polypeptides are tokenized, and bidirectional artificial neural network learning is performed on the presence, location, or arrangement of immunogenic determinants for each tokenized unit sequence. By using the artificial neural network learned in this way, even when there is insufficient information about the three-dimensional folding structure of the protein, an antigenic determinant (epitope) that specifically binds to an antibody, etc. within the target immunogen or target protein (e.g., antigen protein) ), etc., can be predicted reliably.

In the prediction method of this embodiment, the pre-learning step may include dividing the amino acid sequence included in the protein into a plurality of unit sequences with a certain length and tokenizing them.

In a more specific example, the pre-learning step includes masking some of the plurality of tokenized unit sequences; And it may further include the step of pre-learning the function, arrangement, or order of each unit sequence by learning an artificial neural network model along both directions around the masked unit sequence. At this time, the artificial neural network model learning step includes predicting the masked unit sequence; and verifying the accuracy of the prediction result and providing feedback. Through this process, the artificial neural network can predict the function and arrangement of each unit sequence of the protein with high reliability.

Additionally, in the prediction method of one embodiment, the immunogen determinant learning step may include dividing the amino acid sequence included in the polypeptide into a plurality of unit sequences with a certain length and tokenizing them.

In a more specific example, the step of learning the immunogen determinant includes: predicting and classifying in both directions whether an immunogen determinant exists for each of the plurality of tokenized unit sequences; And it may include a step of verifying the accuracy of the prediction result and feeding it back. Through this process, the presence or absence of an immunogenic determinant for each unit sequence is predicted and classified, output and verified as data, and based on the output data. As a result, a prediction model for the immunogenic determinant can be derived.

In the method of one embodiment, the sequence or structure information of the target immunogen or target protein is input based on an artificial neural network learned by the above-described method or a prediction model of the immunogen determinant, and the presence, location, or The sequence can be predicted reliably.

Meanwhile, according to another embodiment of the invention, based on the first database on proteins, bidirectional artificial neural network learning is performed while masking each tokenized unit sequence included in each protein, and the protein unit Pre-learning the function and arrangement of each sequence;

Based on a third database on complexes of mutually bound immunogen polypeptides and immune proteins, bidirectional artificial neural network learning is performed while classifying each tokenized unit sequence included in each complex according to whether it is an immunogen binding site, and the immune Learning information about the immunogen binding site that causes an immune response with the immunogen polypeptide for each protein unit sequence; and

A method for predicting an immunogen binding site is provided, which includes predicting an immunogen binding site in a target immune protein or target immune cell using a learned artificial neural network.

In the prediction method of this other embodiment, the same pre-learning step as in one embodiment is performed, and a learning step of the immunogen binding portion based on a third database on complexes of mutually bound immunogen polypeptides and immune proteins is additionally performed. At this time, the unit sequences included in the complex are tokenized, and bidirectional artificial neural network learning is performed on the presence, location, or arrangement of the immunogen binding portion for each tokenized unit sequence. By using the artificial neural network learned in this way, even when there is insufficient information about the three-dimensional folding structure of the protein, etc., it is possible to specifically bind to the immunogenic determinant such as the antigenic determinant within the target immune protein (e.g., antibody protein). Immunogen binding sites, such as antigen binding sites (paratopes), can be predicted reliably.

In addition, in the prediction method of the other embodiment, by utilizing a third database on the complex in which immunogen polypeptides and immune proteins are interconnected, not only the immunogen binding portion but also the immunogen determining portion within the target immunogen, such as an antigen, can be reliably predicted. there is.

In the prediction method of another embodiment, the immunogen binding site learning step includes dividing the amino acid sequences included in the mutually bound immunogen polypeptide and immune protein into a plurality of unit sequences with a certain length and tokenizing them. can do.

In a more specific example, the immunogen binding site learning step includes predicting and classifying in both directions whether an immunogen binding site exists for each of the plurality of tokenized unit sequences; And it may include verifying the accuracy of the prediction result and providing feedback.

At this time, the tokenized unit sequences corresponding to the positive edges of the immunogen polypeptide and immune protein are given a unique classification identifier and can be calculated as each embedding vector, and these embedding vectors are The artificial neural network learning can proceed by being input in a concatenated state at the edge. Through the assignment of such unique classification identifiers, immunogens and immune proteins can be distinguished, artificial neural network learning can proceed, and the immunogen binding portion and immunogen determining portion can be predicted reliably.

According to the method of the above-described embodiment, even when there is insufficient information about the protein, especially data about the three-dimensional folding structure of the protein, etc., the immune protein or immune cell specifically recognizes or binds within the immunogen to initiate an immune response. The presence, location, or sequence of the immunogen-determining part that causes the immunogen can be predicted more reliably.

In addition, according to another embodiment of the invention, even if the protein-related data is insufficient, the presence, location, or sequence of an immunogen binding portion that specifically binds to the immunogen and causes an immune response in an immune protein or immune cell is determined. It can be predicted reliably. Furthermore, according to the method of the other embodiment, not only the immunogen binding portion but also the immunogen determining portion within the immunogen can be predicted reliably.

Accordingly, the prediction method of one embodiment or another embodiment of the invention can predict the immunogen determinant and/or immunogen binding portion more reliably, thereby greatly contributing to the economical development of a more effective immune-based therapeutic agent or treatment method in a short period of time.

Figure 1 is a schematic diagram schematically showing a method for predicting an immunogen determinant according to an embodiment of the invention.
Figure 2 is a schematic diagram schematically showing the immunogen binding site learning step, which is distinct from one embodiment, in the immunogen binding site prediction method according to another embodiment of the invention.

Hereinafter, with reference to the attached drawings, a method for predicting an immunogen determining portion and an immunogen binding portion according to embodiments of the invention will be described. For reference, Figure 1 schematically shows an example of the immunogen determinant prediction method according to an embodiment of the invention, specifically the dictionary learning step and immunogen determinant learning step included therein.

As shown in Figure 1, in the prediction method according to one embodiment of the invention, a dictionary learning step based on a first database on proteins and a second database on immunogenic and non-immunogenic polypeptides Through the immunogen determinant learning step, a prediction model is derived that predicts the presence, location, or sequence of immunogen determinants, such as epitopes, within the immunogen.

In this pre-learning and immunogen determination part learning step, the unit sequences of each protein and immunogen or non-immunogenic polypeptide included in the first and second databases are tokenized, and the tokenized An artificial neural network model is learned in both directions while masking or classifying part of the unit sequence.

Through this two-way artificial neural network learning process, the order, arrangement, and function of each unit sequence are predicted, and prediction is made to predict whether each unit sequence corresponds to an immune protein such as an antibody or an immunogenic determinant specifically recognized by an immune cell. A model can be derived.

By using this prediction model, it was confirmed that the immunogenic determinant within the target immunogen, such as the target antigen, can be predicted reliably. In particular, in the prediction method of this embodiment, the immunogenic determinant can be reliably predicted based on sequence information of the protein and the immunogenic/non-immunogenic polypeptide. Therefore, according to the method of one embodiment, even when there is insufficient data on the three-dimensional folding structure of proteins such as immunogens, it is possible to reliably predict the immunogen determinant that causes an immune response within the immunogen, and as a result, the prediction is not possible using existing methods. Structural epitopes, which were not easy to determine, can also be predicted more reliably.

Meanwhile, in the prediction method of one embodiment, the first database used in the pre-learning step may include information about one or more of sequence, function, structure, and known mutations for a plurality of proteins. More specifically, the first database may be a protein database containing information on function, sequence, domain structure, and identified mutations for multiple proteins, and examples of such first databases include “Bairoch, A., and R. Apweiler. 1996. “The SWISS-PROT Protein Sequence Data Bank and Its New Supplement TREMBL.” Nucleic Acids Research 24 (1): 21-25” or “Boutet, Emmanuel, Damien Lieberherr, Michael Tognolli, Michel Schneider, and Amos Bairoch. 2007. “UniProtKB/Swiss-Prot.” Known protein databases include “Methods in Molecular Biology 406: 89-112”.

In the pre-learning step, information about a number of proteins included in the first database can be used as learning data without additional processing. However, in order to further increase the efficiency of artificial neural network learning, the information about the plurality of proteins included in the first database can be used as learning data. It can be pre-processed and utilized as information about proteins having amino acid sequences of less than, or less than, 3000, or less than, 2500, or more specifically, amino acid sequences of a certain length that are equivalent to each other. In addition, in the immunogen determination part learning step described later, information about a plurality of immunogenic and non-immunogenic polypeptides included in the second database is converted into information about polypeptides having amino acid sequences of a certain length or less or equal to each other. By pre-processing, the efficiency of artificial neural network learning for deriving a prediction model can be further increased.

Additionally, in the pre-learning step, as shown in FIG. 1, the amino acid sequence included in each protein in the first database can be divided into a plurality of unit sequences with a certain length and tokenized. For example, such tokenization can be performed for each amino acid unit sequence of a certain length, and through this, a plurality of tokens can be generated from the amino acid sequence of the protein.

In the pre-learning step, a portion of the plurality of tokenized unit sequences, for example, 5 to 25%, or 10 to 20%, is masked, while an artificial neural network is used along both directions of the protein around the masked unit sequence. Bidirectional learning can be achieved through the model's attention mechanism. At this time, masking refers to the operation of covering part of the tokenized unit sequence, and the masked token can be distinguished as a 'MASK' token.

In addition, in the artificial neural network learning process, the unit sequence corresponding to the masked token is predicted, and the accuracy of the unit sequence prediction result is confirmed and verified based on a verification dataset and a test dataset separated from the first database. You can give feedback. In this way, by masking some of the tokenized unit sequences in various combinations along both directions of the protein, predicting the masked unit sequences, and confirming, verifying, and feeding back the accuracy of the prediction results, an artificial neural network The function, arrangement, or order of each unit sequence included in the protein can be pre-trained in the model.

As a result of this prior learning, the artificial neural network model can predict the order and arrangement of unit sequences included in the protein under minimal consideration of biological and/or chemical characteristics. In the prediction method of one embodiment, by performing the pre-learning step first, the reliability of the prediction model of the immunogen determination part derived through the steps described later can be further improved.

Meanwhile, for learning a bidirectional artificial neural network in the above-described pre-learning step, a tokenization module that tokenizes the unit sequences, a masking module that masks some of the tokenized unit sequences, and learning while converting and predicting the masked unit sequences A prediction model generation device is used, including a conversion module and a learning module that perform the process, a prediction accuracy calculation module that checks and calculates the accuracy of the unit sequence prediction results, and a prediction model generation module that generates a prediction model for the arrangement of the unit sequences, etc. You can.

The above-described bidirectional artificial neural network learning and prediction model generation device for its progress may have a similar form to the bidirectional artificial neural network learning method and prediction model generation device previously applied for language learning, an example of which is registered in Korea. Patent Publication No. 2426508 or “Devlin, Jacob and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina. 2019. “BERT: Pre-Training of Deep Bidirectional Transformers for Language Understanding.” In Devlin-Etal-2019-Bert, 4171-86. Association for Computational Linguistics”, etc.

Meanwhile, in the prediction method of one embodiment, after performing the above-described prior learning step, immunogenic and non-immunogenic polypeptides classified according to whether immune proteins such as antibodies or immune cells specifically recognize or bind to the immune response are identified. Based on the second database, a two-way artificial neural network learning process is performed to learn a prediction model of the immunogen determination part.

In this immunogenic determinant learning step, the second database is divided into immunogen polypeptides containing immunogen determinants that cause an immune response and non-immunogenic polypeptides that do not contain immunogen determinants, and includes information on their sequences or structures. can do. More specifically, the second database may include information on immunogenic and non-immunogenic polypeptides that have been confirmed to cause or not to cause an immune response with two or more immune proteins or immune cells through multiple experiments, and these verified By utilizing a second database on immunogenic/non-immunogenic polypeptides, the reliability of the prediction method of one embodiment can be further increased.

In a more specific embodiment, the second database may include one or more types of information selected from the group consisting of continuous amino acid sequence information, discontinuous amino acid sequence information, and three-dimensional structure information of the immunogenic and non-immunogenic polypeptides. For example, according to the prediction method of one embodiment, immunogenic determinants in the form of linear epitopes as well as structural epitopes can be predicted. Among these, in the method for predicting the linear antigenic determinant, a second database containing continuous amino acid sequence information of the immunogenic and non-immunogenic polypeptides can be used as learning data, and when trying to predict the three-dimensional antigenic determinant, A second database containing non-contiguous amino acid sequence information or three-dimensional structure information may be used. An example of such a secondary database is “Vita, Randi, Swapnil Mahajan, James A. Overton, Sandeep Kumar Dhanda, Sheridan Martini, Jason R. Cantrell, Daniel K. Wheeler, Alessandro Sette, and Bjoern Peters. 2019. “The Immune Epitope Database (IEDB): 2018 Update.” Nucleic Acids Research 47 (D1): D339-43” or a known database through Protein Data Bank (PDB).

In the immunogen determinant learning step, the information on a number of immunogenic and non-immunogenic polypeptides included in the second database is pre-processed into information on polypeptides having an amino acid sequence of a certain length, thereby increasing the reliability of the prediction method. The fact that it can be increased is as already described above.

Meanwhile, in the immunogen determinant learning step, as shown in FIG. 1, the amino acid sequence included in each polypeptide of the second database can be divided into a plurality of unit sequences with a certain length and tokenized. For example, such tokenization can be performed for each amino acid unit sequence of a certain length corresponding to an immunogenic determinant such as an antigenic determinant, and through this, a plurality of tokens can be generated from the amino acid sequence of an immunogenic or non-immunogenic polypeptide.

In the immunogen determinant learning step, artificial neural network model learning can be performed along both directions of the polypeptide around the plurality of tokenized unit sequences, as in the pre-learning step. In this two-way artificial neural network learning process, it is possible to predict and classify whether each unit sequence corresponds to an immunogen-determining part or whether such an immunogen-determining part exists and output it as data (for example, the immunogen-determining part in “B” in Figure 1 After predicting the decision part, the predicted data is classified as “0” or “1” and output).

The accuracy of this immunogen determinant prediction result can be confirmed and verified based on a test dataset separated from the second database and fed back. For example, parameters can be adjusted through loss functions such as weighted cross entropy. In this way, through the process of predicting whether a tokenized unit sequence is an immunogen determinant or not, and confirming, verifying, and feeding back the accuracy of the prediction result, information on the presence, location, and arrangement of the immunogen determinant within the immunogen can be reliably obtained. A predictive model that makes predictions can be derived.

Based on the artificial neural network model learned in this way or the immunogen determinant prediction model, the sequence or structure information of the unknown target immunogen or target protein is input, and the presence, location, or sequence of the immunogen determinant within this immunogen can be reliably determined. It can be predicted.

In the above-mentioned immunogen decision part learning step, for learning a bidirectional artificial neural network, a tokenization module for tokenizing the unit sequences, a classification module for classifying the tokenized unit sequences, and learning while predicting the unit sequence A prediction model generating device may be used, including a learning module that progresses, a prediction accuracy calculation module that checks and calculates the accuracy of the unit sequence prediction results, and a prediction model generation module that generates a prediction model for the arrangement of the unit sequences.

Meanwhile, according to the prediction method of the above-described embodiment, it is possible to reliably predict a target immunogen represented by an antigen protein or an immunogen determinant of the target protein, and this immunogen determinant specifically binds to an antibody in the antigen protein to induce an immune response. It can be represented by the epitope that causes it. In addition, it was confirmed that in the prediction method of one embodiment, not only linear epitopes but also structural epitopes can be reliably predicted even when information on the three-dimensional structure of a protein is lacking.

Accordingly, the prediction method of one embodiment can more reliably predict immunogenic determinants, such as antigenic determinants, in unknown antigen proteins, etc., using only basic sequence information, thereby contributing to the more effective development of immune-based therapeutics or treatment methods.

Meanwhile, according to another embodiment of the invention, in addition to the above-described immunogen-determining portion, an immunogen-binding portion (e.g., antigen-binding portion; paratope) that causes an immune response with an immunogen, such as the antigen, within an immune protein such as an antibody or an immune cell. ) is provided. The prediction method of this other embodiment is based on a first database on proteins, and proceeds with learning a two-way artificial neural network model while masking each tokenized unit sequence contained in each protein, to determine the unit sequence of the protein. Pre-learning star features and arrangements;

Based on a third database on complexes of mutually bound immunogen polypeptides and immune proteins, bidirectional artificial neural network learning is performed while classifying each tokenized unit sequence included in each complex according to whether it is an immunogen binding site, and the immune Learning information about the immunogen binding site that causes an immune response with the immunogen polypeptide for each protein unit sequence; and

It may include predicting an immunogen binding site in a target immune protein or target immune cell using a learned artificial neural network.

For reference, a schematic diagram of the immunogen binding site learning step in the immunogen binding site prediction method according to this other embodiment, which is distinct from one embodiment, is shown in FIG. 2. In the prediction method of the other embodiment, the dictionary learning step based on the first database may be performed in the same manner as the prediction method of the above-described embodiment, and therefore, further description thereof will be omitted.

Meanwhile, in the prediction method of another embodiment, for the learning step of the immunogen binding portion, a third database on the complex of mutually bound immunogen polypeptides and immune proteins is used instead of the above-described second database. More specifically, the third database may be a database about the sequence or structure of immunogen polypeptides such as the antigen, immune proteins such as the antibody, and complexes in which they are bound to each other, and the immunogen determinant contained in the immunogen polypeptide. Information about the immune protein and information about the immunogen binding part contained in the immune protein may be included together.

In a more specific example, the third database relates to the immunogenic polypeptide, the immune protein, or a complex thereof, their respective sequence or structure, the sequence or binding site of the immunogenic determinant included in the immunogenic polypeptide, and the immune protein. It may include information about the sequence or binding site of the immunogen binding portion included in, or it may include information about all of them. An example of such a third database is EpiPred (Krawczyk K, Liu 94), Docking Benchmarking Dataset (DBD) v5 (Vreven T, Moal IH, Vangone A, Pierce BG, Kastritis PL, Torchala M, et al. Updates to the Integrated Protein-Protein Interaction Benchmarks: Docking Benchmark Version 5 and Affinity Benchmark Version 2. J Mol Biol (2015) 427:3031-41) or “Daberdaku S, Ferrari C (2019) Antibody interface prediction with 3D Zernike descriptors and SVM. Known databases such as “Bioinformatics 35:1870-1876”, etc.

Meanwhile, in the immunogen binding site learning step, as shown in FIG. 2, the amino acid sequences included in the immunogen polypeptide and immune protein of the third database can be divided into a plurality of unit sequences with a certain length and tokenized. there is. For example, this tokenization can be done for each amino acid unit sequence of a certain length corresponding to an immunogen determinant, such as an antigenic determinant of an immunogen polypeptide, and an immunogen binding portion, such as an antigen binding portion of an immune protein. Multiple tokens can be generated from each amino acid sequence of the peptide and immune protein.

At this time, in the immunogenic polypeptide and/or immune protein, the tokenized unit sequence corresponding to each positive edge has a unique classification identifier (e.g., “CLS”, “SEP” in FIG. 2). ”, etc.) can be assigned and calculated as each embedding vector, and the embedding vectors of the immunogen polypeptide and immune protein are input in a concatenated state at the edge, so that subsequent artificial neural network learning can proceed.

At this time, the order in which the immunogen polypeptide and immune protein are connected in the embedding vector is not particularly limited, but for effective artificial neural network learning, an embedding vector in which the immunogen polypeptide and immune protein are connected in a certain order is input and learning is performed. It is appropriate.

In this way, as the data regarding the immunogenic polypeptide and/or immune protein is input in a state separated by a unique classification identifier and artificial neural network learning progresses, in another embodiment of the prediction method, not only the immunogen binding portion of the immune protein but also the immunogen. The immunogenic determinants can be predicted together.

In the immunogen binding site learning step, artificial neural network learning may be performed along both directions around the plurality of tokenized unit sequences. In this two-way artificial neural network learning process, it is possible to predict and classify whether each unit sequence corresponds to an immunogen-binding portion (or an immunogen-determining portion bound thereto) or whether such an immunogen-binding portion, etc. exists, and output it as data (e.g. For example, after predicting the immunogen binding site in Figure 2, the predicted data is classified as “0” or “1” and output). At this time, the immunogen determination part that is combined with the immunogen binding part may also be classified by the unique classification identifier, etc., and the prediction result may be output.

The accuracy of this immunogen binding site (and/or immunogen determining portion) prediction result can be confirmed and verified based on a test dataset separated from a third database and fed back. In this way, through the process of predicting whether the tokenized unit sequence contains an immunogen binding site and confirming, verifying, and feeding back the accuracy of the prediction result, information such as the presence, location, and arrangement of the immunogen binding site within immune proteins such as antibodies A prediction model that reliably predicts can be derived.

Based on the artificial neural network learned in this way or the immunogen binding site prediction model, sequence information of the unknown target immune protein or target immune cell is input, and the presence, location or sequence of the immunogen binding site within the immune protein such as an antibody is determined. etc. can be predicted reliably. Furthermore, in the prediction method of the other embodiment, artificial neural network learning is performed in the third database with data on immunogen polypeptides as well as immune proteins classified by unique classification identifiers, so that the immunogen in the immune protein Not only the binding site, but also the immunogen determining region within the immunogen that binds to it can be further predicted.

In the above-described immunogen binding site learning step, similar to the immunogen determination portion learning step in the method of one embodiment already described above, a prediction model including a tokenization module, a classification module, a learning module, a prediction accuracy calculation module, and a prediction model generation module. A generation device may be used to conduct artificial neural network learning.

Meanwhile, according to the prediction method of another embodiment described above, the immunogen binding portion of the target immune protein represented by an antibody protein and the immunogen determining portion such as the target immunogen represented by an additional antigen protein can be reliably predicted. At this time, the immunogen-binding portion may be represented by an antigen-binding portion (paratope) that specifically binds to the epitope of the antigen in the antibody protein and causes an immune response.

Accordingly, the prediction method of another embodiment more reliably predicts the immunogen-binding portion, such as the antigen-binding portion, and the immunogen-determining portion, such as the antigenic determinant, using only basic sequence information in an unknown antibody and/or antigen protein, etc., thereby providing a more reliable method of predicting an immune-based therapeutic agent or treatment method. It can contribute to effective development.

Hereinafter, preferred embodiments of the invention and comparative examples are described. However, the following examples are only preferred examples of the invention and the invention is not limited thereto.

Example 1: Evaluation of prediction reliability of immunogen determinant

First, the first database for the pre-learning step is “Bairoch, A., and R. Apweiler. 1996. “The SWISS-PROT Protein Sequence Data Bank and Its New Supplement TREMBL.” The Swiss-Prot database known through “Nucleic Acids Research 24 (1): 21-25” was used. This database contains information on the function, domain structure, mutation, and sequence of more than 560,000 proteins. Data on protein sequences, etc. are extracted from this first database, and these data are divided into a learning dataset, a validation dataset, and a test dataset at a ratio of 6:2:2, and based on this, “A” in Figure 1 The described pre-learning steps were performed.

In addition, as the second database for the immunogen determinant learning step, those related to linear antigenic determinants or those related to three-dimensional structural antigenic determinants were used. First, the second database on linear epitopes is “Vita, Randi, Swapnil Mahajan, James A. Overton, Sandeep Kumar Dhanda, Sheridan Martini, Jason R. Cantrell, Daniel K. Wheeler, Alessandro Sette, and Bjoern Peters. 2019. “The Immune Epitope Database (IEDB): 2018 Update.” A known database in “Nucleic Acids Research 47 (D1): D339-43” was used. Additionally, as a second database on three-dimensional structural epitopes, data on some antigens were extracted from the Protein Data Bank (PDB) and used.

The data extracted from this database was divided into a training dataset and a test dataset at a ratio of 8:2, and cross-validation using the cross validation technique was performed on the training dataset. Based on the above classified data, the immunogen determination part learning step shown in “B” in Figure 1 was performed. Through the above process, after performing the pre-learning step and immunogen determinant learning step, the antigenic determinant (linear antigenic determinant: Table 1 or three-dimensional antigenic determinant: Table 2) of the target immunogen is predicted, and the reliability of the prediction result is verified. Statistical evaluation was performed and is shown in Tables 1 and 2 below, respectively. At this time, the statistical parameters that evaluated reliability are “https://jennainsight.tistory.com/entry/F1-Score-Roc%EA%B3%A1%EC%84%A0-Auc-%EA%B3%84%EC %82%B0%EB%B0%A9%EB%B2%95-scikit-learn-%EC%BD%94%EB%93%9C%EB%A1%9C-%EA%B5%AC%ED%98 %84%ED%95%98%EA%B8%B0”, etc., was calculated using a known method, and the higher each calculated statistical parameter, especially the AUC, the higher the reliability of the prediction results.

The reliability evaluation results of these prediction results were shown by comparing Example 1, Comparative Example 1 in which the pre-learning step was omitted in Example 1, and Additional Comparative Examples 2 to 15 in which the existing prediction method was applied. For reference, the existing prediction methods of Comparative Examples 2 to 15 predicted antigenic determinants according to the methods described in the literature summarized at the bottom of Tables 1 and 2, and then evaluated the reliability of the prediction results.

For reference, Comparative Examples 2 to 15 are learning and prediction models known to be applicable to existing antigenic determinants, etc., and the representative prediction models of Comparative Examples 3 and 5 are as follows:

- Random Forest Classifier (Comparative Example 3): A tree-based model that makes the optimal choice at each step. A typical classification-based prediction model that creates many models through data sampling and ultimately selects the one with the most votes through voting;

- Gradient Boosting classifier (Comparative Example 5): A method that adds a boosting function to the tree-based model above and reflects data incorrectly predicted by the model into the next model to make the correct prediction again.

Reliability assessment of linear epitope prediction results statistical parameters AUC Macro F-1 Micro F-1 Precision Recall Example 1 0.922 0.853 0.860 0.855 0.852 Comparative Example 1 (omitting the pre-learning step in Example 1) 0.557 0.509 0.509 0.538 0.537 Comparative Example 2 (Bagging Classifier) 0.861 0.752 0.771 0.747 0.767 Comparative Example 3 (RandomForest Classifier) 0.857 0.762 0.780 0.756 0.780 Comparative Example 4 (ExtraTrees Classifier) 0.857 0.763 0.781 0.756 0.780 Comparative Example 5 (GradientBoosting Classifier) 0.736 0.661 0.699 0.659 0.695 Comparative Example 6 (AdaBoost Classifier) 0.590 0.489 0.611 0.535 0.586 Comparative Example 7 (LogisticRegression Classifier) 0.590 0.588 0.611 0.535 0.568 Comparative Example 8 (Linear Discriminant Analysis) 0.590 0.490 0.612 0.536 0.586 Comparative Example 9 (Bernoulli NB) 0.588 0.518 0.609 0.544 0.576 Comparative Example 10 (Gaussian NB) 0.583 0.536 0.536 0.558 0.558 Comparative Example 11 (Quadratic Discriminant Analysis) 0.515 0.506 0.549 0.511 0.512 Comparative Example 12 (SupportVector Classifier) 0.510 0.384 0.601 0.502 0.570

* Prediction method for Comparative Examples 2 to 12 (source):

- Comparative Example 2:

https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.BaggingClassifier.html

- Comparative Example 3:

https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.RandomForestClassifier.html

- Comparative Example 4:

https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.ExtraTreesClassifier.html

- Comparative Example 5:

https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.GradientBoostingClassifier.html

- Comparative Example 6:

https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.AdaBoostClassifier.html

- Comparative Example 7:

https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LogisticRegression.html

- Comparative Example 8:

https://scikit-learn.org/stable/modules/generated/sklearn.discriminant_analysis.LinearDiscriminantAnalysis.html

- Comparative Example 9:

https://scikit-learn.org/stable/modules/generated/sklearn.naive_bayes.BernoulliNB.html

- Comparative Example 10:

http://scikit-learn.org/stable/modules/generated/sklearn.naive_bayes.GaussianNB.html

- Comparative Example 11:

https://scikit-learn.org/stable/modules/generated/sklearn.discriminant_analysis.QuadraticDiscriminantAnalysis.html

- Comparative Example 12:

https://scikit-learn.org/stable/modules/generated/sklearn.svm.SVC.html

Reliability evaluation of three-dimensional antigenic determinant prediction results statistical parameters AUC Macro F-1 Micro F-1 Precision Recall Example 1 0.678±0.021 0.478±0.014 0.625±0.037 0.535±0.007 0.615±0.027 Comparative Example 1 (omitting the pre-learning step in Example 1) 0.571±0.019 0.253±0.164 0.331±0.288 0.426±0.194 0.523±0.023 Comparative Example 13 (Bepipred-2.0) 0.62 - - - - Comparative Example 14 (LBtope) 0.54 - - - - Comparative Example 15 (NetSurfP) 0.60 - - - -

* Prediction method of Comparative Examples 13 to 15 (source):

- Comparative Example 13: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5570230/

- Comparative Example 14: https://pubmed.ncbi.nlm.nih.gov/23667458/

- Comparative Example 15: https://pubmed.ncbi.nlm.nih.gov/19646261/

Referring to Tables 1 and 2 above, the antigenic determinant prediction method of Example 1 is not only Comparative Example 1, which omits the prior learning step, but also any of the previously known prediction methods of Comparative Examples 2 to 15, for example, using existing antigens. It was confirmed that linear and three-dimensional antigenic determinants within immunogens such as antigens can be predicted with high reliability compared to prediction methods such as Comparative Examples 3 and 5, which are tree-based models used to predict determinants.

Unlike existing tree-based models such as Comparative Examples 3 and 5, which require considerable data for reliable prediction through step-by-step classification and selection, the prediction method in the embodiment uses dictionary learning based on bidirectional artificial neural network learning and immunogen This appears to be because prediction reliability can be improved even when insufficient data is used through the decision learning stage.

Example 2: Evaluation of prediction reliability of immunogen binding site

First, the pre-learning step was performed in the same way as in Example 1.

Additionally, as the main third database for the immunogen binding site learning step, “Daberdaku S, Ferrari C (2019) Antibody interface prediction with 3D Zernike descriptors and SVM. Bioinformatics 35:1870-1876 Using the complex sequence of the database on complexes of antigens and antibodies known in “ope dataset source: Processed by Daberdaku and Ferrari (2019)” as input data, the antigen-binding portion was identified by the method described below. The prediction was made.

Meanwhile, as an additional third database in this example, EpiPred (Krawczyk K, Liu :2288-94) and Docking Benchmarking Dataset (DBD) v5 (Vreven T, Moal IH, Vangone A, Pierce BG, Kastritis PL, Torchala M, et al. Updates to the Integrated Protein-Protein Interaction Benchmarks: Docking Benchmark Version 5 and The complex sequence from the database on complexes of antigens and antibodies (Affinity Benchmark Version 2. J Mol Biol (2015) 427:3031-41) was used as input data. This was used to predict not only the immunogen binding site, but also immunogen determinants such as antigenic determinants according to the method of this example.

Meanwhile, the data extracted from the third database was divided into a learning dataset and a test dataset at a ratio of 8:2, and cross-validation using the cross validation technique was performed on the learning dataset as in Example 1. . Based on the above classified data, the immunogen binding site learning step shown in Figure 2 was performed. Through the above process, after performing the pre-learning step and the immunogen binding site learning step, the antigen-binding site of the target antibody was predicted, and the reliability of the prediction result was statistically evaluated, which is shown in Table 3 below. At this time, the statistical parameters that evaluated reliability are “https://jennainsight.tistory.com/entry/F1-Score-Roc%EA%B3%A1%EC%84%A0-Auc-%EA%B3%84%EC %82%B0%EB%B0%A9%EB%B2%95-scikit-learn-%EC%BD%94%EB%93%9C%EB%A1%9C-%EA%B5%AC%ED%98 %84%ED%95%98%EA%B8%B0”, etc., was calculated using a known method, and the higher each calculated statistical parameter, especially AUC-ROC, the higher the reliability of the prediction results.

The reliability evaluation results of these prediction results were shown by comparing Example 2 and Additional Comparative Examples 16 to 17 to which the existing prediction method was applied. For reference, the existing prediction method of Comparative Examples 16 to 17 predicted the antigen binding site according to the method described in the literature summarized at the bottom of Table 3, and then evaluated the reliability of the prediction result.

Reliability evaluation of antigen binding site prediction results statistical parameters AUC ROC AUC PR Example 2 0.969 0.633 Comparative Example 16 (ANTIBODY I-PATCH) 0.840 0.376 Comparative Example 17 (DABERDAKU ET AL.) 0.950 0.658

* Prediction method of Comparative Examples 16 to 17 (source):

- Comparative Example 16: https://pubmed.ncbi.nlm.nih.gov/24006373/

- Comparative Example 17: https://pubmed.ncbi.nlm.nih.gov/30395191/

Referring to Table 3 above, it was confirmed that the antigen-binding site prediction method of Example 3 can predict antigen-binding sites within immune proteins such as antibodies with greater reliability than the previously known prediction methods of Comparative Examples 16 to 17.

Claims (21)

Based on the first database on proteins, a two-way artificial neural network model is learned while masking each tokenized unit sequence contained in each protein, and the function and arrangement of each unit sequence of the protein are pre-learned. step;
Based on a second database of immunogenic and non-immunogenic polypeptides classified according to whether immune proteins or immune cells specifically recognize or bind to the immune response, the amount is classified by the tokenized unit sequence contained in each polypeptide. Performing directional artificial neural network learning to learn information on immunogenic determinants that cause an immune response for each unit sequence of the polypeptide; and
Comprising the step of predicting an immunogenic determinant of a target immunogen or target protein using a learned artificial neural network,
A method for predicting immunogen determinants using the Swiss-Prot database as the first database.
delete The method of claim 1, wherein the pre-learning step includes dividing the amino acid sequence included in the protein into a plurality of unit sequences with a certain length and tokenizing them.
The method of claim 3, wherein the pre-learning step includes: masking some of the plurality of tokenized unit sequences; and
A method for predicting an immunogen determinant further comprising the step of pre-learning the function, arrangement, or order of each unit sequence by learning an artificial neural network model along both directions around the masked unit sequence.
The method of claim 4, wherein the artificial neural network model training step includes: predicting the masked unit sequence; and
A method for predicting an immunogen determinant comprising the step of verifying the accuracy of the prediction result and providing feedback.
The method of claim 1, wherein the second database includes information on immunogenic and non-immunogenic polypeptides that have been confirmed to cause or not cause an immune response with two or more immune proteins or immune cells through multiple experiments. method.
The method of claim 6, wherein the second database includes at least one type of information selected from the group consisting of continuous amino acid sequence information, non-contiguous amino acid sequence information, and three-dimensional structure information of the immunogenic and non-immunogenic polypeptides. method.
The method of claim 1, wherein the immunogen determinant learning step includes dividing the amino acid sequence included in the polypeptide into a plurality of unit sequences with a certain length and tokenizing them.
The method of claim 8, wherein the immunogen determining part learning step includes predicting and classifying in both directions whether an immunogen determining part exists for each of the plurality of tokenized unit sequences; and
A method for predicting an immunogen determinant comprising the step of verifying the accuracy of the prediction result and providing feedback.
The method of claim 9, wherein, from the immunogen determining portion learning step, the presence or absence of an immunogen determining portion for each unit sequence is predicted and classified, output and verified as data, and a prediction model of the immunogen determining portion is derived based on the output data. Immunogen crystal region prediction method.
The method of claim 1, wherein the first database or the second database relates to a protein or polypeptide having an amino acid sequence of a certain length or less and is pre-processed.
The method of claim 1 or 10, wherein the immunogen determinant prediction step is performed by inputting sequence or structure information of the target immunogen or target protein based on the learned artificial neural network or the immunogen determinant prediction model, A method for predicting immunogen determinants that predicts the presence, location, or sequence of immunogen determinants.
The method of claim 1, wherein the target immunogen or target protein is an antigen protein, and in the step of predicting the immunogen determinant, an immunogen determinant predicts an epitope in the antigen protein that specifically binds to an antibody and causes an immune response. Prediction method.
The method of claim 13, wherein in the step of predicting the immunogenic determinant, a linear epitope or a structural epitope is predicted.
Based on the first database on proteins, a two-way artificial neural network model is learned while masking each tokenized unit sequence contained in each protein, and the function and arrangement of each unit sequence of the protein are pre-learned. step;
Based on a third database on complexes of mutually bound immunogen polypeptides and immune proteins, bidirectional artificial neural network learning is performed while classifying each tokenized unit sequence included in each complex according to whether it is an immunogen binding site, and the immune Learning information about the immunogen binding site that causes an immune response with the immunogen polypeptide for each protein unit sequence; and
Predicting an immunogen binding site in a target immune protein or target immune cell using a learned artificial neural network,
In the immunogen binding site learning step, the tokenized unit sequences corresponding to both edges of the immunogen polypeptide and immune protein are given a unique classification identifier and calculated as each embedding vector,
Embedding vectors of the immunogenic polypeptide and immune protein are input in a concatenated state at the edge, and artificial neural network learning is performed,
In the immunogen binding site prediction step, the immunogen binding site prediction method further predicts the immunogen determining portion of the target immunogen.
The method of claim 15, wherein the third database is related to the immunogen polypeptide, the immune protein, or a complex thereof, the respective sequence or structure, the sequence or binding site of the immunogenic determinant included in the immunogen polypeptide, and the immune protein. A method for predicting an immunogen binding site including information on the sequence or binding location of the immunogen binding site included in a protein.
The method of claim 15, wherein the immunogen binding site learning step includes dividing the amino acid sequences included in the immunogen polypeptide and the immune protein into a plurality of unit sequences with a certain length and tokenizing them. method.
The method of claim 17, wherein the immunogen binding site learning step includes predicting and classifying in both directions whether an immunogen binding site exists for each of the plurality of tokenized unit sequences; and
A method for predicting an immunogen binding site, comprising the step of verifying the accuracy of the prediction result and providing feedback.
delete delete The method of claim 15, wherein the target immune protein is an antibody protein, and in the immunogen binding part prediction step, an antigen binding part (paratope) that specifically binds to an epitope in the antibody protein and causes an immune response is predicted. Immunogen binding site prediction method.