KR20220151388A - A system for searching the new peptide - Google Patents

Abstract

The present invention relates to a peptide new material search method and system that can precisely predict the immunoactivity of a target peptide sequence by learning the binding possibility of MHC 1 and an epitope peptide by an artificial intelligence deep learning method. The peptide new material search method according to the present invention comprises the steps of: 1) collecting training data using known epitope peptides of MHC 1 as a positive training set and a human-reference protein sequence having immunotolerance as a negative training set; 2) using the training data collected in step 1) to learn whether the training data and MHC 1 are combined using an artificial intelligence deep learning technique, thereby establishing an epitope peptide prediction model; and 3) inputting an epitope candidate peptide into the epitope peptide prediction model established in step 2) to predict whether the epitope candidate peptide and MHC 1 are combined.

Description

Peptide new material search method and system {A SYSTEM FOR SEARCHING THE NEW PEPTIDE}

The present invention relates to a method and system for searching for a new peptide material, and more particularly, to search for a new peptide material that can precisely predict the immunoactivity of a target peptide sequence by learning the possibility of binding MHC 1 and an epitope peptide with an artificial intelligence deep learning learning method It relates to methods and systems.

MHC 1 is present in the cell membrane of cells with nuclei. Their purpose is to distinguish infected cells from healthy own cells. When a pathogenic peptide binds to MHC and is presented on the cell surface of an Antigen Presenting Cell (APC), T cells recognize it, activate it, and initiate an immune response.

At this time, a specific portion (mainly 8 to 10 mer) of the antigen that is presented in a state bound to MHC-1 and identified by B cells or T cells is called an epitope.

Here, it is known that the more stable the binding between the MHC and the peptide, the stronger the immune response to efficiently remove the pathogen. Therefore, a technique for determining peptides of pathogens that can stably bind to a specific MHC can be usefully utilized in the development of vaccines for diseases caused by the pathogens.

However, the MHC gene is a gene that shows the most severe polymorphism among genes possessed by humans, and a number of alleles exist to induce immune responses against various pathogens. Due to such polymorphism of MHC, it is very inefficient and time-consuming to directly test and determine a peptide that can be a T cell epitope among various peptides.

The technical problem to be solved by the present invention is to learn the binding possibility of MHC 1 and epitope peptides with an artificial intelligence deep learning learning method to precisely and quickly predict the immune activity for a target peptide sequence. To provide a new peptide discovery method and system will be.

The peptide new material search method according to the present invention to solve the above-mentioned technical problem is: 1) known epitope peptides of MHC 1 are used as a positive training set, and immunotolerant human reference (Hunan-Reference) protein sequences are used as negative collecting training data as a training set; 2) establishing an epitope peptide prediction model by learning whether the training data and MHC 1 are combined with artificial intelligence deep learning using the training data collected in step 1); 3) inputting the epitope candidate peptide into the epitope peptide prediction model established in step 2) to predict whether the epitope candidate peptide binds to MHC 1;

And, in the present invention, the positive training set is preferably obtained from IEDB (IMMUNE EPITOPE DATABASE AND ANALYSYS RESOURCE).

In the present invention, the negative training set is preferably obtained by randomly extracting sequences from human reference protein sequences and then selecting sequences having little homology with the positive training set.

In addition, in the peptide new material search method according to the present invention, it is preferable to analyze the homology between the extracted human reference protein sequence and the positive training set using blastp.

In addition, in the present invention, the epitope peptide prediction model is a long short-term memory neural network (LTSM), which is a type of convolution neural network (CNN) and recurrent neural network (RNN). It is desirable to use a mixture of memory models).

In addition, in the present invention, it is preferable to apply the average value of probabilities of the epitope peptide predicted by each model by making 6 prediction models.

In addition, in the present invention, the six prediction models are preferably two CNN models, three LSTM models, and one CNN+LSTM mixed model.

On the other hand, the present invention also provides a peptide new material search system characterized by predicting whether or not the binding of an epitope candidate peptide and MHC 1 is performed by the above-described peptide new material search method.

According to the peptide new material search method of the present invention, the possibility of binding MHC 1 and epitope peptides is learned by artificial intelligence deep learning learning method, and the immune activity to the target peptide sequence is much faster and more sophisticated than the existing mass/binding-peptide antigen search method. It has the advantage of being predictable.

On the other hand, the peptide new material search system of the present invention is universal and has the advantage of being applicable to neoantigens or various viruses. As an example of this, when the epitope candidate peptide is extracted from blood malignancy-related proteins or the Covid-19 virus, blood malignancy-derived antigens or Covid-19 virus-derived antigens can be obtained.

1 is a process chart of a new peptide discovery method according to an embodiment of the present invention.
2 is a process diagram of a training data collection step according to an embodiment of the present invention.
3 is a diagram summarizing the progress of the training data collection step according to an embodiment of the present invention.
4 is a diagram summarizing the progress of the step of predicting the binding of MHC 1 with an epitope candidate peptide according to an embodiment of the present invention.
Figure 5 is a diagram showing the immuno-anticancer agent development process by the peptide new material search method according to an embodiment of the present invention.

Hereinafter, specific embodiments of the present invention will be described in detail with reference to the accompanying drawings.

As shown in FIG. 1, the peptide new material search method according to this embodiment starts with the training data collection step (S100). In this step (S100), training data used for learning the epitope peptide prediction model is collected, and a positive training set and a negative training set are prepared respectively.

In this embodiment, this step (S100) is preferably divided into sub-steps as shown in FIG. 2 in detail. First, a step of preparing a positive training set (S110) is performed. Here, the 'positive training set' refers to an epitope peptide that binds to MHC 1, and in this embodiment, the positive training set is known as an MHC class 1 epitope in IEDB (IMMUNE EPITOPE DATABASE AND ANALYSYS RESOURCE) take the data

Next, sub-steps for collecting a negative training set are performed. First, as shown in FIG. 2, a step of extracting a human reference protein (S120) is performed. Here, the 'negative training set' refers to a protein sequence that does not bind to MHC 1 and does not function as an MHC class 1 epitope, and in this step (S120), as shown in FIG. 3, a human reference with immune tolerance (Human Reference) is to randomly extract protein sequences.

Next, as shown in FIG. 2 , homology analysis with the positive training set for the human reference protein extracted in the previous step (S120) is performed (S130). In this step (S130), homology with data selected as the positive training set for the human reference protein sequences randomly extracted in the previous step (S120) is analyzed. In this case, the homology analysis between the human reference protein and the positive training set is preferably performed using blastp, as shown in FIG. 3 .

Next, as shown in FIG. 2, a step of selecting a negative training set (S140) proceeds. In this step (S140), a negative training set is determined by selecting sequences having little homology with the positive training set among the human reference protein sequences for which the homology analysis in the previous step (S130) has been completed.

Using the human reference protein in this way has the advantage of being able to secure a very large number of negative training sets, thereby establishing an accurate epitope peptide prediction model.

Next, as shown in FIG. 1, a step of establishing an epitope peptide prediction model (S200) proceeds. That is, using the training data including the positive training set and the negative training set collected in the previous step (S100), the AI deep learning technique learns whether the training data and MHC 1 are combined to establish an epitope peptide prediction model. .

In this embodiment, it is preferable to use a mixture of a Recurrent Neural Network (RNN) and a Convolution Neural Network (CNN) as the epitope peptide prediction model. At this time, it is preferable to make the epitope peptide prediction model under six conditions and average the probabilities of the epitope predicted by each model in order to obtain improved binary classification performance.

Here, the six prediction models are preferably two CNN models, three LSTM models, and one CNN+LSTM mixed model.

Next, as shown in FIG. 1, a step (S300) of predicting binding between an actual epitope candidate peptide and MHC 1 using the epitope peptide prediction model established in the previous step (S200) is performed. That is, if the binding between the epitope candidate peptide and MHC 1 is predicted through this step (S300), the corresponding epitope candidate peptide becomes a candidate antigen peptide.

Therefore, in this step (S300), as shown in FIG. 4, the epitope candidate peptide is input into the epitope peptide prediction model established in the previous step (S200) to predict whether the epitope candidate peptide binds to MHC 1. As an example of this, when an epitope candidate peptide is extracted from a blood malignancy-related protein or a Covid-19 virus, a blood malignancy-derived antigen or a Covid-19 virus-derived antigen can be obtained.

As shown in FIG. 5, using the peptide new material search system according to this example, 52 candidate antigenic peptides could be selected within about 2 weeks. The candidate antigen peptides selected in this way actually confirmed the binding efficacy with MHC 1 over about 2 months, and as a result, as shown in FIG. 5, 37 of the 52 candidate antigen peptides showed binding to MHC 1 You can confirm that it works.

Among the candidate antigen peptides whose binding efficacy has been confirmed, antigen peptides with high binding activity are selected through a procedure for confirming the T-cell activation efficacy for about 3 months, as shown in FIG. 5 , to select the final peptide.

Claims (8)

1) collecting training data using known epitope peptides of MHC 1 as a positive training set and immune-tolerant human reference protein sequences as a negative training set;
2) establishing an epitope peptide prediction model by learning whether the training data and MHC 1 are combined with artificial intelligence deep learning using the training data collected in step 1);
3) inputting the epitope candidate peptide into the epitope peptide prediction model established in step 2) to predict whether the epitope candidate peptide binds to MHC 1; The method of claim 1, wherein the positive training set,
Peptide new material search method, characterized in that obtained from IEDB (IMMUNE EPITOPE DATABASE AND ANALYSYS RESOURCE). The method of claim 2, wherein the negative training set,
A peptide new material search method, characterized in that obtained by randomly extracting sequences from human reference protein sequences and selecting sequences having little homology with the positive training set. According to claim 3,
A peptide new material search method, characterized in that the homology between the extracted human reference protein sequence and the positive training set is analyzed using blastp. The method of claim 1, wherein the epitope peptide prediction model,
Peptide new material search method characterized by using a mixture of long short-term memory models (LTSM), which is a type of convolution neural network (CNN) and recurrent neural network (RNN). The method of claim 5, wherein the epitope peptide prediction model,
A peptide new material search method characterized by applying the average value of the probabilities of epitopes predicted by each model by creating six prediction models. The method of claim 6, wherein the six predictive models,
Peptide new material search method, characterized in that 2 CNN models, 3 LSTM models, 1 CNN + LSTM mixed model. A peptide new material search system characterized by predicting whether or not the binding of an epitope candidate peptide and MHC 1 is performed by the method according to any one of claims 1 to 7.

Images