KR102184720B1 - Prediction method for binding preference between mhc and peptide on cancer cell and analysis apparatus - Google Patents

Abstract

The present invention is to provide a technique for predicting the binding potential of major histocompatibility complex (MHC)-peptide expressed in cancer cells. According to the present invention, a method for predicting the binding potential of MHC-peptide binding on a surface of cancer cells comprises the steps of: receiving genome data of a sample by an analysis device; generating, by the analysis device, a matrix representing the binding preference of amino acid pairs using a first amino acid sequence of an MHC and a second amino acid sequence of an antigen generated by the cancer cell; and predicting, by the analysis device, the binding potential of the MHC with the antigen by inputting the matrix into a neural network model learned in advance. The first amino acid sequence is a sequence commonly found in a cohort to which the sample belongs, or a sequence detected by the analysis device from the genome data.

Description

A method and analysis device for predicting MHC-peptide binding degree on the surface of cancer cells {PREDICTION METHOD FOR BINDING PREFERENCE BETWEEN MHC AND PEPTIDE ON CANCER CELL AND ANALYSIS APPARATUS}

The technique described below relates to a technique for predicting the degree of binding between MHC and a peptide.

Cancer cells produce new antigens. Epitopes of new antigens are expressed in major histocompatibility complexes (MHCs) located on the surface of cancer cells. T cells recognize MHC-epitopes and trigger an immune response.

Cancer cells evade immunity by using the immune checkpoint of immune cells. Immune checkpoint inhibitors suppress immune checkpoints and kill cancer cells by activating immune cells in the body. It is necessary to predict MHC-peptide binding to identify new antigens produced by cancer cells.

Nielsen, M. et al. NetMHCpan, a method for quantitative predictions of peptide binding to any HLA-A and -B locus protein of known sequence. PLoS One 2, 2007.

The technique described below is intended to provide a technique for predicting the binding potential of MHC-peptide expressed in cancer cells.

In the method for predicting MHC-peptide binding on the surface of cancer cells, an analysis device receives genome data of a sample, and the analysis device receives a first amino acid sequence of a major histocompatibility complex (MHC) and a second amino acid of an antigen generated by cancer cells. Generating a matrix representing the interaction of the amino acid pair using the sequence, and predicting a degree of binding between the MHC and the antigen by inputting the matrix into a previously learned neural network model by the analysis device. The first amino acid sequence is a sequence commonly found in a cohort to which the sample belongs, or a sequence detected by the analysis device in the genome data.

The analysis device for predicting the degree of MHC-peptide binding is an input device that receives genomic data of a sample, and the interaction of amino acid pairs with the first amino acid sequence constituting the major histocompatibility complex (MHC) and the second amino acid sequence constituting the antigen. A storage device for storing a neural network model that predicts the degree of binding between MHC and antigen based on a matrix image representing, and detecting the first amino acid sequence and the second amino acid sequence from the genome data, and the amino acid of the first amino acid sequence And a computing device that generates a matrix image indicating the degree of interaction between the amino acid pair of the second amino acid sequence and the second amino acid sequence, and inputs the generated matrix image to the neural network model to predict the degree of binding of the MHC and the antigen to the sample. .

The technique described below can accurately predict the degree or possibility of binding MHC and peptide using a neural network model.

1 is an example of a system for predicting the degree of MHC-peptide binding.
2 is an example of a process of predicting the degree of MHC-peptide binding.
3 is an example of an interaction map.
4 is an example of a conventional CNN model.
5 is an example of a CNN model for predicting the degree of MHC-peptide binding.
6 is an example of an analysis device for predicting MHC-peptide binding.
7 is a result of evaluating a CNN model for predicting MHC-peptide binding.

The technology to be described below may be modified in various ways and may have various embodiments, and specific embodiments will be illustrated in the drawings and described in detail. However, this is not intended to limit the technology to be described below with respect to a specific embodiment, and it should be understood to include all changes, equivalents, and substitutes included in the spirit and scope of the technology described below.

Terms such as 1st, 2nd, A, B, etc. may be used to describe various components, but the components are not limited by the above terms, only for the purpose of distinguishing one component from other components. Is only used. For example, without departing from the scope of the rights of the technology described below, a first component may be referred to as a second component, and similarly, a second component may be referred to as a first component. The term and/or includes a combination of a plurality of related listed items or any of a plurality of related listed items.

In terms of the terms used in the present specification, expressions in the singular should be understood as including plural expressions unless clearly interpreted differently in context, and terms such as "includes" are specified features, numbers, steps, actions, and components. It is to be understood that the presence or addition of one or more other features or numbers, step-acting components, parts or combinations thereof is not meant to imply the presence of, parts, or combinations thereof.

In addition, in performing the method or operation method, each of the processes constituting the method may occur differently from the specified order unless a specific order is clearly stated in the context. That is, each process may occur in the same order as the specified order, may be performed substantially simultaneously, or may be performed in the reverse order.

Hereinafter, terms used in the description will be described.

Antigens are substances that induce an immune response.

Neoantigens are antigens with alterations that occur through mutations in tumor cells or post-translational modifications specific to tumor cells. The neoantigen may comprise a polypeptide sequence or a nucleotide sequence. Mutations can be frame shifted or non-lattice shifted indels, missense or nonsense substitutions, splice site alterations, genomic rearrangements or gene fusions, or any genomic or expression alterations that result in a newborn ORF. It may include. Mutations can also include splice variants. Post-translational modifications specific to tumor cells can include abnormal phosphorylation. Post-translational modifications specific to tumor cells can also include proteasome-generated spliced antigens.

An epitope may refer to a specific portion of an antigen to which an antibody or T-cell receptor usually binds.

MHC is a peptide structure that acts as a mediator to recognize the target substance of the immune response as an antigen.

Peptide refers to a polymer of amino acids. For convenience of explanation, hereinafter "peptide" refers to an amino acid polymer or an amino acid sequence expressed on the surface of cancer cells.

The MHC-peptide complex is expressed in cancer cells and is a complex structure of MHC and peptide. T cells recognize the MHC-peptide complex and perform an immune response.

The degree of binding refers to the degree of binding between the MHC and the peptide. Binding affinity to binding affinity refers to the binding affinity between the MHC molecule and the peptide.

A sample or sample means a single cell or multiple cells, cell fragments, body fluids, etc. collected from an individual to be analyzed.

Subjects include cells, tissues or organisms. The entity is primarily intended for humans, but is not limited thereto.

Exomes are a subset of the genome that encodes proteins. An exome may refer to a cell, a group of cells, or a collection of exons present in an individual.

Genomic data or genome information refers to genetic information that is calculated by analyzing a sample. For example, genomic data is a base sequence obtained from deoxyribonucleic acid (DNA), ribonucleic acid (RNA), or protein from cells, tissues, etc., gene expression data, genetic mutation with standard genomic data, DNA methylation ), etc. In general, genomic data includes sequence information obtained by analyzing a specific sample. Genomic data can be obtained in a variety of ways. For example, genome data can be generated through NGS analysis. Genomic data can be expressed as digital data understood by computers.

Machine learning, or learning, is a field of artificial intelligence, which refers to the field in which algorithms are developed so that computers can learn. A machine learning model or learning model means a model developed so that a computer can learn. Learning models include various models such as artificial neural networks and decision trees, depending on the approach.

The analysis device refers to a device that predicts the degree of MHC-peptide binding. The analysis device processes and analyzes the data using the installed program or code.

1 is an example of a system 100 for predicting the degree of MHC-peptide binding. In FIG. 1, the analysis devices 130, 140, 150 predict the degree of MHC-peptide binding. In FIG. 1, the analysis device is shown in the form of a server 130 and computer terminals 140 and 150. The server 130 may provide a service for predicting the degree of MHC-peptide binding on a network. The computer terminal 140 is connected to a network to receive genome data, and predicts MHC-peptide binding degree using an installed application. The computer terminal 150 receives input data from a medium (eg, USB, SD card, etc.) storing genome data, and predicts MHC-peptide binding degree using an installed application. The analysis devices 130, 140, and 150 may be implemented in various forms.

Analysis devices (130, 140, 150) analyze the MHC-peptide binding degree using the genome data. Here, the genome data includes information on the genome sequence. The genome analysis device 110 analyzes a sample and generates genome data. For example, the genome analysis device 110 may be an NGS analysis device. The genome analysis device 110 may store the generated genome data in a separate DB 120.

The genome analysis device 110 generates genome data using a genome library. The genome analysis device 110 may perform whole exome sequencing on the genome library. The genome library can be prepared using a commercial kit. For example, using the AllPrep DNA / RNA Mini Kit (Qiagen, 80204), AllPrep DNA / RNA Micro Kit (Qiagen, 80284), or QIAamp DNA FFPE Tissue Kit (Qiagen, 56404) to generate a genomic library for sequence analysis. I can.

Users (10, 20, 30) can check the degree of MHC-peptide binding to a specific sample. The user 10 may access the server 130 through a user terminal (PC, smartphone, etc.) and check the analysis result performed by the server 130. The user 20 can check the MHC-peptide binding degree through the computer terminal 140 used by the user. The user 30 can check the MHC-peptide binding degree through the computer terminal 150 used by the user 30.

Users (10, 20, 30) may be researchers who study the degree of MHC-peptide binding. Alternatively, the users 10, 20, and 30 may be medical staff who consider prescribing an anticancer drug for a specific patient.

2 is an example of a process 200 predicting the degree of MHC-peptide binding. The analysis apparatus 130, 140, or 150 may predict the degree of MHC-peptide binding through the prediction process 200 described in FIG. 2.

The analysis device receives the genome data of the sample (210). The sample refers to an individual to determine the degree of MHC-peptide binding. The sample genome data refers to the genome data of an individual to be analyzed.

The analysis device must detect the MHC amino acid sequence and the peptide amino acid sequence in the genomic data. The analysis device may use a program or model for predicting the MHC structure. For example, the analysis device may predict the HLA (Human Leukocyte Antigen) structure using HLAminer. In addition, the analysis device can identify the peptide amino acid sequence from the genomic data using a certain tool. For example, the analysis device may detect a peptide sequence by searching for an amino acid sequence flanking nonsynonymous mutations in genome data using an idfetch program.

The analysis device generates an interaction map using the genome data (220). The interaction map contains information on the binding affinity between the amino acids constituting the MHC and the amino acids of the peptide. The interaction map contains binding affinity information between amino acid pairs. The binding affinity information may be information that directly or indirectly represents the binding affinity between two amino acids. For example, the binding affinity information may be an interaction energy value of an amino acid pair in a certain protein structure.

The interaction map may be a matrix having a value indicating binding affinity. In addition, the interaction map may be a two-dimensional image in which the binding affinity between amino acid pairs is expressed as a color value. The interaction map will be described later.

The analysis device performs analysis by inputting the interaction map to the neural network model (230). FIG. 2 shows an example of a neural network model that extracts a combination feature from an interaction map and calculates information about the combination or non-combination. The analysis device provides information on the degree of MHC-peptide binding based on the information output from the neural network model.

The analysis device uses the genomic data to create an interaction map. Alternatively, a computer device other than the analysis device may generate an interaction map and transmit it to the analysis device. In this case, the analysis device can predict the degree of MHC-peptide binding by directly inputting the interaction map into the neural network model. Hereinafter, it is assumed that the analysis device generates an interaction map.

The analysis device may generate an interaction map based on information on the protein structure disclosed in the prior art. A database (database) published for academic or commercial use can hold information on protein structure. Based on native protein structures, the frequency of adjacent amino acids in the protein structure can be calculated. If the frequency of adjacent amino acid pairs in a protein is high, it can be said that the interaction energy of the amino acid pairs is high. If the interaction energy is calculated according to the amino acid pair constituting the protein structure, the analysis device may determine the interaction preference between the amino acid pair belonging to the MHC and the amino acid pair belonging to the peptide. In addition, the analysis device can generate an interaction map that numerically represents the interaction preference between amino acid pairs. Furthermore, the analysis device may generate an interaction map representing the interaction preference between amino acid pairs as color values.

The interaction preference between amino acid pairs can be determined in units of regions that uniformly divide the protein structure. For example, interaction preferences between amino acids can be determined based on contact between Cα atoms or some other atoms present in the protein structure. The contact of certain atoms in the protein structure means that the interactions or affinity between the atoms are high.

The researcher verified the effect on the neural network model to predict the degree of MHC-peptide binding. Some of the verification results will be described later. The experimental results showed the highest reliability when using the contact between Cα-Cα among the atoms of the protein structure. Thus, it may be desirable to create an interaction map based on the contact between Cα atoms. The Cα atom corresponds to an atom that forms the skeleton of a three-dimensional structure among atoms constituting a molecule.

3 is an example of an interaction map. The interaction map is a two-dimensional matrix with horizontal and vertical axes. 3 shows an image having color values as an example. The horizontal axis corresponds to the amino acid sequence labeled 1 to n, and the vertical axis corresponds to the amino acid sequence labeled a to z. One axis is the amino acid sequence of the MHC, and the other axis is the amino acid sequence of the antigenic peptide. The correlation preference (proximity) of amino acid pairs was expressed in color. Black means that the relationship preference is high. White means that the preference for interaction is low. Referring to FIG. 3, amino acid 7 on the horizontal axis generally has a high correlation preference. In addition, amino acids f and p on the vertical axis generally have high correlation preference. The correlation preference of amino acid pair 7-f is higher than that of amino acid pair 7-p.

The analysis device predicts the degree of MHC-peptide binding using a neural network model. As a neural network model, various models such as recurrent neural networks (RNN), feedforward neural networks (FFNN), and convolutional neural networks (CNN) may be used. Hereinafter, for convenience of explanation, the CNN model will be described.

4 is an example of a conventional CNN model. 4 is for explaining the general structure and operation of the CNN model.

The CNN includes a convolution layer (Conv), a pooling layer (Pool), and a fully connected layer. A number of convolutional layers and pooling layers may be repeatedly disposed. The CNN of FIG. 4 may have a structure of 5 convolutional layers, 2 pooling layers, and 2 fully connected layers.

The convolution layer outputs a feature map through a convolution operation on an input image. At this time, a filter that performs a convolution operation is also called a kernel. The size of the filter is called the filter size or kernel size. An operation parameter constituting the kernel is called a kernel parameter, a filter parameter, or a weight.

The convolutional layer performs convolution and nonlinear operations.

The convolution operation is performed on a window of a certain size. The window can be moved one by one from the upper left to the lower right of the image, and the size of the movement can be adjusted at a time. The size of the movement is called a stride. The convolutional layer performs a convolution operation on all areas of the input image while moving the window in the input image. The convolution layer can maintain the dimension of the input image after the convolution operation by padding the edge of the image.

The nonlinear operation layer is a layer that determines output values from neurons (nodes). The nonlinear operation layer uses a transfer function. Transfer functions include Relu and sigmoid functions.

The pooling layer subsamples the feature map obtained as a result of the operation in the convolutional layer. Pooling operations include max pooling and average pooling. Maximum pooling selects the largest sample value within the window. Average pooling is sampled as the average value of the values included in the window.

The all-connected layer finally classifies the input image. The all-connected layer receives all values output from the previous convolutional layer and performs a final classification. In FIG. 4, the all-connected layer outputs a classification result using a softmax function.

5 is an example of a CNN model 300 for predicting the degree of MHC-peptide binding. The CNN model 300 predicts the degree of MHC-peptide binding based on the input data (interaction map). The CNN model 300 includes a plurality of convolutional layers 310 and 320, a full connection layer 330 and an output layer 340. As shown in FIG. 5, the convolutional layer may be composed of two layers.

The convolutional layers 310 and 320 perform a convolution operation on input data, and output a value obtained by applying a ReLU (rectified linear unit) function to the convolutional value. The convolution operation is an operation that multiplies an input value by a weight matrix. Weights are provided through the learning process. The convolutional layers 310 and 320 extract interaction features from the input data. Kernels use interaction maps to find specific motifs important for peptide-MHC binding.

The input data may use the above-described interaction map. The input data includes parameters indicating the degree of interaction of the amino acid pair.

The all-connection layer 330 integrates input information. The all-connection layer 330 receives a value output from the convolutional layer 320 as an input. The all-connection layer 330 may perform a ReLU operation.

The output layer 340 outputs information on the binding possibility of a given protein and an MHC protein using a sigmoid function.

The convolutional layers 310 and 320 perform convolution using a specific number of kernels or weight matrices. The convolution may be a one-dimensional or two-dimensional operation. All convolution results are converted by ReLU. ReLU converts negative values to zero.

The first convolutional layer 310 (hereinafter referred to as a first convolutional layer) detects a combination pattern in input data. The first convolutional layer may use a window having a moving distance of 1. The high-level convolutional layer corresponds to a convolutional kernel that detects a combination pattern from an output value of a previous layer. The operation of the convolutional layer is shown in Equation 1 below. The second convolutional layer 320 (hereinafter, the second convolutional layer) may have the same structure as the convolutional layer 310. Alternatively, the second convolutional layer 320 may have a window size or a stride width different from that of the convolutional layer 310.

X is the input data, i is the index indicating the location of the output, and k is the index of the kernel. Each convolution kernel W _k corresponds to a weight matrix of size M×N. M is the window size, and N is the number of input channels.

The pooling layer is not used. That is, all output values of the convolutional layer are used for prediction. This is because amino acids that are relatively distant from each other can also affect the interaction between the MHC-peptide complex and the T-sero receptor.

The full connection layer 330 takes all outputs output from the second convolutional layer 320 as inputs. The all-connection layer 330 integrates input values output from the previous layer. The all-connected layer performs the ReLU(WX) function. X is an input value, and W is a weight matrix for the all-connected layer.

The output layer 340 may output a value between 0 and 1 according to the sigmoid function. A value output by the output layer 340 is a value for classifying a binding or non-binding state. The output layer 340 performs a sigmoid function Sigmoid(WX). X is the input value and W is the weight matrix for the sigmoid output layer.

The output layer 340 may use an activation function such as Softmax or ReLU, not sigmoid.

Hereinafter, the CNN model refers to a model for predicting MHC-peptide binding degree or binding possibility as shown in FIG. 5.

The CNN model is trained in the direction of minimizing the objective function. The learning process corresponds to the process of optimizing the weights used in the CNN model. For example, the weight optimization may use a gradient descent method.

The objective function is defined by the sum and regularization term of NLL (negative log likelihood). The objective function for the CNN model can be expressed as Equation 2 below.

이다.

to be.

s is the index of the training sample (data). t is the index of the interactive feature. Y _t ^s is the label value (0 or 1) for sample s and the interaction feature t. f _t (X ^s ) is the predicted probability for the interactive feature t of the input X ^s .

The normalization technique can use various techniques used in deep learning network training. Multiple normalization techniques combined in the CNN model can be applied. L2 regularization term ||W|| ₂ ² is the sum of the squares of all weight matrices. L1 regularization term ||H ^-1 || ₁ is the L1 norm for all outputs of all connected layers immediately before the output layer.

Optimization is dependent on the normalization condition. The normalization condition is ||W _m ⁿ || for any layer m and neuron n ₂ ≤λ3, or the L2 standard of weights of all neurons is greater than a certain value.

Various values can be used for hyperparameters for the CNN model. The learning rate can be [0.001, 0.01, 0.1]. The number of kernels for the first layer and the second layer (convolutional layer) may be [10, 30, 50]. The L1 and L2 normalization parameters may be [0.001, 0.01, 0.1]. The momentum may be [0,1, 0.5, 0.9].

In particular, the sizes of filters used to extract interaction features from the convolutional layer may vary. For example, a 1-5bp peptide, 1/2 length HLA, 2/3 length HLA, and full length HLA may each have different size filters.

6 is an example of an analysis device 400 for predicting the degree of MHC-peptide binding. The analysis device 400 is a device corresponding to the analysis device 130, 140, or 150 of FIG. 1.

The analysis device 400 may predict the degree of MHC-peptide binding using the neural network model described above. The analysis device 400 may be physically implemented in various forms. For example, the analysis device 400 may have a form such as a computer device such as a PC, a server of a network, or a chipset for image processing. The computer device may include a mobile device such as a smart device.

The analysis device 400 includes a storage device 410, a memory 420, an operation device 430, an interface device 440, a communication device 450, and an output device 460.

The storage device 410 stores a neural network model that predicts the degree of MHC-peptide binding. The neural network model must be trained in advance. Furthermore, the storage device 410 may store programs or source codes required for data processing. The storage device 410 may store input genome data and a predicted MHC-peptide binding degree.

The memory 420 may store data and information generated in the process of analyzing the data received by the analysis device 400.

The interface device 440 is a device that receives certain commands and data from the outside. The interface device 440 may receive dielectric data from an input device physically connected or an external storage device. The interface device 440 may receive a learning model for data analysis. The interface device 440 may receive training data, information, and parameter values for training a learning model.

The communication device 450 refers to a component that receives and transmits certain information through a wired or wireless network. The communication device 450 may receive genome data from an external object. The communication device 450 may also receive data for model training. The communication device 450 may transmit information on the MHC-peptide binding degree determined for the input sample to an external object.

The communication device 450 to the interface device 440 are devices that receive certain data or commands from the outside. The communication device 450 to the interface device 440 may be referred to as an input device.

The output device 460 is a device that outputs certain information. The output device 460 may output an interface required for a data processing process and an analysis result.

The computing device 430 may predict the degree of MHC-peptide binding to the input sample genome data using the neural network model stored in the storage device 410. The computing device 430 may predict the MHC-peptide binding degree by directly or uniformly processing a result output from the neural network model. The computing device 430 may train a learning model used for predicting MHC-peptide binding degree by using the given training data. The computing device 430 may be a device such as a processor, an AP, or a chip in which a program is embedded that processes data and processes certain operations.

Hereinafter, an experiment to verify the performance of the CNN model described above will be described. Training data for constructing a CNN model was used as an immune epitope database (eg, IEDB). An interaction map was created by estimating the binding preference of amino acid pairs using the DB for the modified protein structure. The CNN model was trained based on the two-dimensional interaction pattern between peptide-MHC.

For training data, data obtained from IEDB 3.0 were used. For prediction of binding between peptide-MHC class I. The database provides 57,173 data (training data) including binding affinity defined by IC ₅₀ /EC ₅₀ standards. The threshold for determining the affinity was an IC ₅₀ /EC ₅₀ value (500 nM). The binding state was set as the case of IC ₅₀ /EC ₅₀ <500nM, and the non-binding state was set as the case of IC ₅₀ /EC ₅₀ ≥ 500nM. That is, an interaction map was created from the training data, and the CNN model was trained based on the interaction map and the information on the association state for the corresponding training data. The training data were mainly data for 9mer and 10mer peptides, and data for HLA-A and HLA-B types of MHC class I.

7 is a result of evaluating a CNN model for predicting MHC-peptide binding. 7(A) is a result of predicting the degree of binding to 9-mer peptide and HLA-A. 7(B) is a result of predicting the degree of binding to the 10-mer peptide and HLA-A. 7(C) is a result of predicting the degree of binding to 9-mer peptide and HLA-B.

The area under the curve (AUC) for HLA-A and HLA-B was 0.93 and 0.94, respectively. Using IEDB's test data set, NetMHCpan, a conventionally known prediction tool, and CNN model were compared. For information on NetMHCpan, refer to "Automated benchmarking of peptide-MHC class i binding predictions, Bioinformatics 31, 2015". Compared with the conventional prediction tool, the CNN model showed higher performance in 70.5% of the total test data regardless of the HLA type.

In addition, the method for constructing a neural network model and the method for predicting MHC-peptide binding degree as described above may be implemented as a program (or application) including an executable algorithm that can be executed in a computer. The program may be provided by being stored in a non-transitory computer readable medium.

The non-transitory readable medium refers to a medium that stores data semi-permanently and can be read by a device, not a medium that stores data for a short moment, such as a register, cache, or memory. Specifically, the above-described various applications or programs may be provided by being stored in a non-transitory readable medium such as a CD, DVD, hard disk, Blu-ray disk, USB, memory card, and ROM.

The present embodiment and the accompanying drawings are merely illustrative of some of the technical ideas included in the above-described technology, and those skilled in the art will be able to easily within the scope of the technical ideas included in the specification and drawings It will be apparent that all of the modified examples and specific embodiments that can be inferred are included in the scope of the rights of the above-described technology.

Claims (20)

Receiving, by the analysis device, genome data of the sample;
Obtaining, by the analysis device, a first amino acid sequence of a major histocompatibility complex (MHC) and a second amino acid sequence of an antigen generated by cancer cells by using the genome data;
Generating, by the analysis device, a matrix representing information on an amino acid pair using the first amino acid sequence and the second amino acid sequence; And
The analysis device comprises the step of predicting the degree of binding of the MHC and the antigen by inputting the matrix into a neural network model learned in advance,
In the matrix, a first axis is the first amino acid sequence, a second axis is the second amino acid sequence, and for each amino acid pair formed by the first amino acid sequence and the second amino acid sequence, constituting the amino acid pair It contains information of the proximity of the distance between two amino acids,
The proximity is a method for predicting MHC-peptide binding degree on the surface of cancer cells, which is determined based on the frequency of proximity of the amino acid pair in the published protein structure database. The method of claim 1,
The analysis device is a method for predicting MHC-peptide binding degree on the surface of cancer cells for detecting the first amino acid sequence by using a program for detecting a human leukocyte antigen (HLA) allele in the genome data. The method of claim 1,
The second amino acid sequence is a new antigen generated by cancer cells, and a method for predicting MHC-peptide binding degree on the surface of cancer cells comprising a mutant sequence. The method of claim 1,
The matrix is a method for predicting the degree of MHC-peptide binding on the surface of cancer cells showing the binding preference of the amino acid pair based on Cα of the MHC and Cα of the antigen. delete The method of claim 1,
The neural network model is a CNN (Convolutional Neural Network), and a method for predicting MHC-peptide binding degree on the surface of cancer cells in which the kernel of the CNN detects a characteristic for MHC-peptide binding. The method of claim 1,
The neural network model is a CNN (Convolutional Neural Network) model consisting of a plurality of convolutional layers, a pre-connection layer, and a sigmoid output layer. The method of claim 7,
The convolutional layer performs a conversion by applying a convolution operation and a ReLU (Rectified Linear Unit) function,
The all-connected layer integrates the output values of the previous convolutional layer using the ReLU function,
The sigmoid output layer converts the output value of the pre-linked layer into a value representing a binding state between 0 and 1, the method for predicting MHC-peptide binding degree on the surface of cancer cells. The method of claim 7,
The convolutional layer is a method for predicting MHC-peptide binding degree on the surface of cancer cells that performs a convolution operation (convolution(X)) according to the following formula.

(X is the input data, i is the index indicating the location of the output, k is the index of the kernel, each convolution kernel W _k is the M×N weight matrix, M is the window size, and N is the number of input channels) The method of claim 1,
The neural network model is a method for predicting MHC-peptide binding degree on the surface of cancer cells in which weights are learned with the following objective function.

(NLL is the loss function that is the sum of the negative log likelihood, ||W|| ₂ ² is the L2 normalization function, which is the sum of squares of all weight matrices, and ||H ^-1 || ₁ is the fully connected layer. All output values of L1 are standard) The method of claim 1,
The neural network model is
A method for predicting MHC-peptide binding degree on the surface of cancer cells comprising a convolutional layer using filters of different sizes according to the length of the MHC and the amino acid length of the antigen. An input device for receiving dielectric data of a sample;
A storage device for storing a neural network model for predicting the degree of binding of the MHC and the antigen based on information indicating the proximity of each amino acid pair formed by the amino acid sequence constituting the major histocompatibility complex (MHC) and the amino acid sequence constituting the antigen; And
Detecting the first amino acid sequence of the MHC and the second amino acid sequence of the antigenic peptide from the genomic data, and generating a matrix representing the proximity of each amino acid pair formed by the amino acid of the first amino acid sequence and the second amino acid sequence, Comprising a computing device for predicting the degree of binding of the MHC and antigen to the sample by inputting the generated matrix into the neural network model,
In the matrix, a first axis is the first amino acid sequence, a second axis is the second amino acid sequence, and for each amino acid pair formed by the first amino acid sequence and the second amino acid sequence, constituting the amino acid pair It contains information of the proximity of the distance between two amino acids,
The proximity degree is an analysis device for predicting the MHC-peptide binding degree determined based on the frequency of the proximity of the amino acid pair in the published protein structure database. The method of claim 12,
The calculation device is an analysis device for predicting MHC-peptide binding degree for detecting the first amino acid sequence for the sample by using a program for detecting a human leukocyte antigen (HLA) allele in the genome data. The method of claim 12,
The second amino acid sequence is a new antigen generated by cancer cells, and an analysis device for predicting MHC-peptide binding degree including a mutant sequence. The method of claim 12,
The calculation device predicts the degree of MHC-peptide binding that generates the matrix representing the binding preference of the amino acid pair based on Cα of the MHC and Cα of the antigen. The method of claim 12,
The neural network model is a convolutional neural network (CNN) model consisting of a plurality of convolutional layers, a full-connection layer, and a sigmoid output layer. The method of claim 16,
The convolutional layer is an analysis device for predicting MHC-peptide binding degree that performs a convolution operation (convolution(X)) according to the following equation.

(X is the input data, i is the index indicating the location of the output, k is the index of the kernel, each convolution kernel W _k is the M×N weight matrix, M is the window size, and N is the number of input channels) The method of claim 12,
The neural network model is an analysis device that predicts the MHC-peptide binding degree of which weights are learned using the following objective function.

(NLL is the loss function that is the sum of the negative log likelihood, ||W|| ₂ ² is the L2 normalization function, which is the sum of squares of all weight matrices, and ||H ^-1 || ₁ is the fully connected layer. All output values of L1 are standard) The method of claim 12,
The neural network model is
An analysis device for predicting MHC-peptide binding degree including a convolutional layer using filters of different sizes according to the length of the MHC and the amino acid length of the antigen. A computer-readable recording medium in which a program for executing the method for predicting the degree of MHC-peptide binding on the surface of cancer cells according to any one of claims 1 to 4 and 6 to 11 is recorded on a computer.

Images