Immune epitope prediction method based on antigen presentation and immunogenic feature fusion

By integrating antigen presentation and immunogenicity prediction methods, the characteristics of peptides and HLA-I proteins are extracted, and the antigen presentation and immunogenicity model are combined to generate immunogenicity scores, solving the problem of low prediction accuracy in the prior art , achieving higher immunoepitope prediction performance and model interpretability.

CN119028435AActive Publication Date: 2024-11-26TSINGHUA UNIVERSITY

Patent Information

Application Number
CN202310593175.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-24
Publication Date
2024-11-26
Estimated Expiration
2043-05-24

AI Technical Summary

Technical Problem

Existing immunoepitope prediction methods have low prediction accuracy when screening tumor neogenic antigens or designing cancer vaccines, mainly because they only focus on the binding process of peptides to HLA-I proteins, and ignore other key knowledge of antigen presentation, or The lack of immunogenicity or TCR binding information leads to insufficient accuracy and recall.

Method used

Using an immunoepitope prediction method based on the fusion of antigen presentation and immunogenic characteristics, the peptide and HLA-I protein characteristics are extracted from the immunoepitope database by constructing a feature extraction module, combining antigen presentation and immunogenic prediction models. Generate immunogenic scores of peptide-activated T cells to achieve prediction of immunoepitopes. This method uses a transformer encoder to extract sequence information, calculate attention scores, and predict through a three-layer fully connected neural network.

Benefits of technology

It improves the prediction accuracy of immunoepitopes, especially in antigen presentation and immunogenicity prediction, outperforms existing methods, enhances the interpretability of the model, and provides guidance on screening of neogenic antigens in practical applications.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119028435A_ABST
    Figure CN119028435A_ABST
Patent Text Reader

Abstract

The invention provides an immune epitope prediction method based on antigen presentation and immunogenicity feature fusion. The method comprises the following steps: extracting peptide fragments and I-type human leukocyte antigen protein features from an immune epitope database by using a constructed feature extraction module; inputting the peptide fragment and the I-type human leukocyte antigen protein characteristics into a pre-trained antigen presentation prediction model, and obtaining the antigen presentation probability based on the antigen presentation prediction model; and inputting the peptide fragment, the I-type human leukocyte antigen protein characteristics and antigen presentation probability information into an immunogenicity prediction model to realize fusion of antigen presentation and immunogenicity characteristics, and obtaining an immunogenicity score representing the T cell activation probability of the peptide fragment based on the immunogenicity prediction model, thereby realizing prediction of the immunoepitope. According to the invention, the prediction precision of the immune epitope is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence, bioinformatics, biomedicine and tumor immunology technology, and in particular to an immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics. Background Art

[0002] Understanding the rules of T cell-mediated immune response is of great significance for the treatment of diseases including cancer. Antigen recognition is the first and most critical step in the entire immune response process. It mainly depends on the antigen presentation function of the cell. Antigen presentation can display various antigen information synthesized or ingested by cells on the cell surface for T cell recognition. Among them, the presentation pathway based on major histocompatibility complex (MHC) molecules is the most important. MHC proteins related to antigen presentation can be divided into MHC class I and MHC class II. MHC class I molecules are expressed in nucleated cells and mainly present endogenous antigen fragments, which can activate CD8+T cells; while MHC class II molecules are expressed in antigen-presenting cells (APC, such as dendritic cells, macrophages, B cells), mainly present exogenous antigen fragments, which can activate CD4+T cells. MHC I proteins in the human body are also called class I human leukocyte antigen (HLA-I) proteins.

[0003] MHC proteins can bind to peptides produced by antigen degradation to form peptide-MHC complexes, which are then transported to the cell surface for display. Due to T cell immune tolerance, endogenous protein information presented by normal cells will not activate T cells. However, if cells express non-endogenous proteins due to cancer, viral infection or allogeneic transplantation, it is very likely that T cells will be activated through the MHC protein-mediated presentation pathway, triggering subsequent immune responses. Finding T cell immune epitopes, that is, MHC ligands that can be presented and activate T cell immune responses, has a very important guiding significance for medical treatment. Especially in tumor immunotherapy, the tumor neoantigen prediction algorithm developed based on the above rules can filter and screen the candidate set of neoantigens, reduce the workload of experimental verification, and is regarded as a key tool for the development of tumor vaccines. In addition, the tumor neoantigens obtained by these prediction methods can also be used as targets for CAR-T therapy.

[0004] Antigen presentation includes multiple processes, such as protease degradation, TAP transport, MHC protein binding, etc. Among them, the binding affinity between peptides and MHC proteins is the most important indicator for screening CD8+T cell immune epitopes. Studies have found that in most CD8+T cell immune epitopes, the affinity IC50 between peptides and MHC proteins is <500nM, so 500nM has become the most commonly used affinity threshold for screening CD8+T cell immune epitopes in the past two decades. However, the binding of peptides to MHC proteins is only one link in antigen presentation, and affinity prediction cannot describe the entire antigen presentation process. This also leads to the fact that there will be a performance ceiling for screening immune epitopes using only affinity prediction models. However, with the advancement of experimental technology in recent years, a large amount of eluted ligand data has been accumulated in this field by eluting peptide-MHC complexes on the cell surface and determining sequence information using mass spectrometry experiments. This type of data describes all antigen presentation processes including protease degradation, TAP transport, and peptide binding to MHC proteins, and can achieve end-to-end antigen presentation modeling, thus greatly improving the performance of antigen presentation prediction. However, there is a lot of evidence that peptides that can be presented do not necessarily activate T cells. One of the key influencing factors is the binding of the T cell receptor (TCR) to the peptide-MHC complex. Only peptides that are presented to the cell surface by MHC proteins and can bind to TCR are immunogenic peptides. The Immune Epitope Database (IEDB) has accumulated a large amount of immunogenicity data obtained from T cell immune response measurement experiments. This type of data describes the immune response process including antigen presentation and TCR binding, and can be used to train immunogenicity prediction models.

[0005] Based on the natural properties of the training data and the meaning of the model output, the existing technology divides the immune epitope prediction methods into three categories. The first category of models is trained on affinity data, and the model output represents the binding affinity or probability of the peptide to the MHC protein. For example, the model ACME uses a convolutional neural network module to extract the amino acid sequence features of the peptide and MHC, and introduces an attention mechanism to explain the peptide-MHC binding mode. This type of model uses the affinity data open to the immune epitope database IEDB for training, and its output is an affinity value normalized to the range of 0 to 1. The current best affinity prediction model TransPHLA applies the multi-head self-attention mechanism that has been very popular in natural language processing in recent years to this field, and its performance exceeds 14 peptide-MHC protein binding prediction methods. When applied to the screening of neoantigens, the positive prediction rate of TransPHLA reached 96%. In addition, as pan-allele-specific models, ACME and TransPHLA can make accurate predictions for MHC typing that has not been seen in the training data samples. The second category of models is trained on eluted ligand data, or the final output of the model is an antigen presentation score. For example, NetMHCpan4.1, the most popular tool in the field, is a simple feedforward neural network with only one hidden layer. During the training process, the model used mono-allelic and multi-allelic eluted ligand data to train the antigen presentation prediction network, and also used affinity data to train the affinity prediction network. The two networks use hard parameter sharing to synergistically improve each other's performance. In addition, MHCflurry2.0 decomposes the antigen presentation problem into two modules: antigen processing prediction and affinity prediction. The model uses convolutional neural networks to build a prediction network for antigen processing, and uses fully connected layers to build a prediction network for affinity. MHCflurry2.0 uses logistic regression to fuse the outputs of the two networks to obtain the final antigen presentation score. The third category of models is trained on immunogenicity data, and its output represents the probability that the peptide is immunogenic. For example, DeepHLApan uses gated recurrent units (a variant of LSTM) to build the network and adds an attention layer at the end of the network. The output immunogenicity score is used as an additional filter to improve the precision of affinity models or antigen presentation models when screening immune epitopes. In addition, the latest immunogenicity prediction model DeepImmuno-CNN uses a convolutional neural network and uses a Beta distribution probability model to set different immunogenicity labels for each pair of peptide-MHC complexes during training. When applied to the test sets of dengue virus, new coronavirus, and cancer neoantigens, DeepImmuno-CNN has improved both recall and precision.

[0006] When designing personalized cancer vaccines or searching for CD8+T cell immune epitopes of infectious viruses, the importance of immune epitope prediction algorithms is self-evident. However, the three types of methods currently proposed all have limitations. The affinity model only considers the binding process between peptides and MHC proteins, which means that the remaining candidate peptides after screening can only guarantee that they may bind to MHC proteins, but they are not necessarily products of protease degradation or targets of TAP transport. This leads to the fact that the use of affinity prediction models to screen immune epitopes will naturally have a higher recall rate and a lower precision rate. Although the biological process reflected by the antigen presentation prediction model is longer than that of the affinity prediction model, it is still insufficient in precision because the training data still lacks information related to TCR binding. In addition, the immunogenicity prediction model that seems to describe the entire process also has limitations. This is because whether a specific peptide-MHC complex can activate T cells is closely related to the TCR library of the test individual. The negative sample measured in the experiment only means that it cannot activate T cells in the current experiment, and it does not mean that the peptide is non-immunogenic. Therefore, the label noise that may exist at the data level greatly limits the performance of current immunogenicity prediction models.

[0007] Existing immune epitope prediction methods either only focus on the binding process between peptides and HLA-I proteins, ignoring other key knowledge of antigen presentation, or lack immunogenicity or TCR binding information. This leads to the inevitable problem of low prediction accuracy when using existing immune epitope prediction methods to screen tumor neoantigens or design cancer vaccines and antiviral vaccines. Summary of the invention

[0008] The present invention provides an immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics, which is used to improve the prediction accuracy of immune epitopes.

[0009] The present invention provides an immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics, comprising:

[0010] The constructed feature extraction module was used to extract peptide segments and human leukocyte antigen type I protein features from the immune epitope database;

[0011] Inputting the peptide segment and the type I human leukocyte antigen protein features into a pre-trained antigen presentation prediction model, and obtaining the probability of antigen presentation based on the antigen presentation prediction model;

[0012] The peptide segment and type I human leukocyte antigen protein characteristics and antigen presentation probability information are input into the immunogenicity prediction model, and the antigen presentation and immunogenicity characteristics are realized based on the immunogenicity prediction model to obtain an immunogenicity score representing the probability of the peptide segment activating T cells, so as to realize the prediction of the immune epitope.

[0013] According to an immune epitope prediction method based on the fusion of antigen presentation and immunogenicity features provided by the present invention, the constructed feature extraction module is used to extract peptide segments and type I human leukocyte antigen protein features from the immune epitope protein database, specifically comprising:

[0014] The feature extraction module extracts sequence information based on the transformer encoder and calculates the attention score to mine peptide segments and type I human leukocyte antigen protein sequence sites that have a key impact on the final output of the respective models for the antigen presentation prediction model and the immunogenicity prediction model;

[0015] Based on the sequence embedding and position encoding method, the amino acid word vector of each site in the sequence is constructed, and three weight matrices are multiplied by the word vector to obtain three different vectors: Query, Key, and Value;

[0016] The attention weight of each word vector is calculated by Queries and Keys to generate an attention weight matrix;

[0017] The elements related to each sequence position in the attention weight matrix are extracted, and their average value is calculated to obtain the attention value of each position, which reflects the influence of the position on the final output of the antigen presentation prediction model and the immunogenicity prediction model.

[0018] According to an immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics provided by the present invention, the training of the antigen presentation prediction model and the immunogenicity prediction model requires the use of positive sample and negative sample data of eluted ligand data and immunogenicity data;

[0019] Collect positive sample data of elution culture data and immunogenicity data from the existing database, delete sample data whose peptide length is not within the set range, and collect the remaining sample data to construct a positive training data set;

[0020] The eluted ligand data and immunogenicity data are peptide sample data randomly extracted from the protein database PDB, and negative samples ten times the number of positive samples are generated to augment the training data. The randomly sampled peptide data and experimentally verified non-immunogenic peptide data together constitute a negative training data set.

[0021] According to an immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics provided by the present invention, the immunogenicity prediction model is divided into an antigen presentation branch and an immunogenicity branch;

[0022] The antigen presentation branch and the immunogenicity branch are both composed of a three-layer fully connected neural network, and the features extracted by the feature extraction module are passed through two fully connected layers with 1024 and 256 neurons respectively and then batch normalized;

[0023] Both fully connected layers use rectified linear units as activation functions, and the immunogenicity score representing the probability of peptide activating T cells is generated by a fully connected layer with only one neuron.

[0024] According to an immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics provided by the present invention, the antigen presentation branch of the immunogenicity prediction model fully utilizes the network weights of the antigen presentation prediction model, and its parameters remain unchanged during the training of the immunogenicity prediction model involving immunogenicity data;

[0025] The loss gradient resulting from the immunogenicity score is propagated back to the immunogenicity branch only.

[0026] According to an immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics provided by the present invention, the antigen presentation prediction model and the immunogenicity prediction model use a five-fold cross-validation method to find better parameters and a prediction model with the best performance;

[0027] In the training of the prediction model, five neural networks with better performance were obtained based on five-fold cross validation. The final prediction model was a collection of five neural networks. The average of the output values ​​of the five neural networks was calculated as the final output value of the prediction model, and the performance of the prediction model was evaluated using a variety of evaluation indicators.

[0028] The present invention also provides an immune epitope prediction system based on the fusion of antigen presentation and immunogenicity characteristics, the system comprising:

[0029] A feature extraction module is used to extract peptide segments and type I human leukocyte antigen protein features from an immune epitope database through a constructed feature extraction module;

[0030] An antigen presentation module, used for inputting the peptide segment and the type I human leukocyte antigen protein features into a pre-trained antigen presentation prediction model, and obtaining the probability of antigen presentation based on the antigen presentation prediction model;

[0031] The immunogenicity module is used to input the characteristics of peptide segments and type I human leukocyte antigen protein and antigen presentation probability information into the immunogenicity prediction model, and obtain the immunogenicity score representing the probability of peptide segments activating T cells based on the immunogenicity prediction model, thereby realizing the prediction of immune epitopes.

[0032] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for predicting immune epitopes based on the fusion of antigen presentation and immunogenicity characteristics as described above is implemented.

[0033] The present invention provides an immune epitope prediction method based on the fusion of antigen presentation and immunogenicity features. The amino acid sequences of polypeptides and type I human leukocyte antigen proteins are effectively encoded through a feature extraction module, achieving advanced performance in antigen presentation prediction. The knowledge of antigen presentation is integrated into the network structure of immunogenicity prediction, thereby greatly improving the prediction performance of immunogenicity. The analysis of the output distribution enhances the interpretability of the model and provides guidance in searching for immune epitopes. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 This is one of the flow diagrams of the immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics provided by the present invention;

[0036] Figure 2 It is a schematic diagram of module connection of the immune epitope prediction system based on the fusion of antigen presentation and immunogenicity characteristics provided by the present invention;

[0037] Figure 3 It is a schematic diagram of the network structure of the antigen presentation prediction model IEPAPI-EL provided by the present invention;

[0038] Figure 4 It is a schematic diagram of the network structure of the immunogenicity prediction model IEPAPI-IM provided by the present invention;

[0039] Figure 5 It is a schematic diagram of the feature extraction module in the models IEPAPI-EL and IEPAPI-IM provided by the present invention;

[0040] Figure 6 It is a schematic diagram of the network structure of the immunogenicity prediction model baseline-IM provided by the present invention;

[0041] Figure 7 It is a detailed schematic diagram of network parameters of the fully connected layer of each network model provided by the present invention;

[0042] Figure 8 It is a schematic diagram of calculating the attention score of the input peptide segment and HLA protein provided by the present invention;

[0043] Fig. 9 It is a schematic diagram of performance comparison between the baseline-IM model and the IEPAPI-IM model on the validation set 1 provided by the present invention;

[0044] Fig.10 It is a schematic diagram of the peptide patterns displayed by the five HLA-I subtypes provided by the present invention in antigen presentation and immunogenicity;

[0045] Fig.11 is a schematic diagram of attention scores of four HLA-B27 subtypes associated with ankylosing spondylitis provided by the present invention;

[0046] Fig.12 It is a schematic diagram of the output distribution of the model IEPAPI on three independent test sets provided by the present invention;

[0047] Fig.13 It is a structural schematic diagram of the electronic device provided by the present invention.

[0048] Reference numerals:

[0049] 110: feature extraction module; 120: antigen presentation module; 130: immunogenicity module;

[0050] 1310: processor; 1320: communication interface; 1330: memory; 1340: communication bus. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] Combine the following Figure 1 The present invention describes an immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics, comprising:

[0053] S100, extracting peptide segments and human leukocyte antigen type I protein features from the immune epitope database using the constructed feature extraction module;

[0054] S200, inputting the peptide segment and the type I human leukocyte antigen protein features into a pre-trained antigen presentation prediction model, and obtaining the probability of antigen presentation based on the antigen presentation prediction model;

[0055] S300, inputting the peptide segment and type I human leukocyte antigen protein characteristics and antigen presentation probability information into an immunogenicity prediction model, realizing antigen presentation and immunogenicity characteristics based on the immunogenicity prediction model to obtain an immunogenicity score representing the probability of the peptide segment activating T cells, so as to realize the prediction of immune epitopes.

[0056] In the present invention, a new immune epitope prediction method, called IEPAPI, is proposed. First, IEPAPI trains a transformer encoder-based feature extraction module in the antigen presentation prediction task to obtain the features of peptides and HLA-I proteins. Then IEPAPI inputs the peptide and type I human leukocyte antigen protein features and antigen presentation probability information obtained in the antigen presentation prediction task into the network for immunogenicity prediction. Quantitative comparison results on an independent antigen presentation test dataset show that IEPAPI outperforms the most advanced methods NetMHCpan4.1 and mhcflurry2.0 on 100% (25 / 25) and 76% (19 / 25) of HLA-I subtypes, respectively. In addition, IEPAPI also shows good prediction accuracy on two independent neoantigen datasets, which shows that IEPAPI can become an important tool for finding neoantigens.

[0057] The constructed feature extraction module is used to extract peptide segments and type I human leukocyte antigen protein features from the immune epitope database, specifically including:

[0058] The feature extraction module extracts sequence information based on the transformer encoder and calculates the attention score to mine peptide segments and type I human leukocyte antigen protein sequence sites that have a key impact on the final output of the respective models for the antigen presentation prediction model and the immunogenicity prediction model;

[0059] Based on the sequence embedding and position encoding method, the amino acid word vector of each site in the sequence is constructed, and the three weight matrices are multiplied by the word vector to obtain three different vectors: Query, Key, and Value;

[0060] The attention weight of each word vector is calculated by Queries and Keys to generate an attention weight matrix;

[0061] The elements related to each sequence position in the attention weight matrix are extracted, and their average value is calculated to obtain the attention value of each position, which reflects the influence of the position on the final output of the antigen presentation prediction model and the immunogenicity prediction model.

[0062] like Figure 5As shown, the feature extraction module first fills the character "*" to the right end of the peptide sequence to form a maximum length of 11, and then connects the peptide sequence with the pseudo sequence of the HLA-I protein. Since the HLA protein is represented by a pseudo sequence consisting of residues at 34 sites on the binding pocket, the final spliced ​​sequence length is 45. The present invention designs tokens for 20 common amino acids, filler characters "*" and unknown amino acid symbols "X". The connected sequence is tokenized to form the input of the model: (c1, c2, ..., c 22 )→(t1,t2,…,t 22 ). Where c i is a character, t i is its corresponding token, which is also an integer. Unlike the previous algorithm that uses the BLOSUM matrix to encode amino acids, the present invention uses a trainable embedding layer to encode each token into a 128-dimensional word vector. In addition, the present invention also uses a trainable parameter matrix to encode the position information of the input sequence. The embedding layer and position encoding are the first two layers of the end-to-end model, and their parameters are learned during the training process. The encoded sequence is processed by a self-attention layer with 8 heads. Finally, the present invention averages all word vectors except the padding symbol '*' to obtain a 128-dimensional feature vector to characterize the peptide and HLA information.

[0063] like Figure 8 As shown, because the present invention uses a transformer encoder to extract sequence information, an attention score can be calculated to show which sequence parts are more critical to the final output. After sequence embedding and position encoding, the word vector is multiplied by three weight matrices to obtain three different vectors: Query, Key, and Value. The attention weight of each word vector is calculated by Queries and Keys, and then an attention weight matrix is ​​obtained. The present invention extracts the elements related to each sequence position in the attention weight matrix, and calculates its average value to obtain the attention value of each position, which reflects the degree of influence of the position on the final output of the antigen presentation prediction model and the immunogenicity prediction model.

[0064] The training of the antigen presentation prediction model and the immunogenicity prediction model requires the use of positive sample data and negative sample data of the eluted ligand data and the immunogenicity data;

[0065] Collect positive sample data of elution culture data and immunogenicity data from the existing database, delete sample data whose peptide length is not within the set range, and collect the remaining sample data to construct a positive training data set;

[0066] The eluted ligand data and immunogenicity data are peptide sample data randomly extracted from the protein database PDB, and negative samples ten times the number of positive samples are generated to augment the training data. The randomly sampled peptide data and experimentally verified non-immunogenic peptide data together constitute a negative training data set.

[0067] In the present invention, there are two types of data used: 1) eluted ligands (EL) data: peptides eluted from the cell surface that can bind to HLA proteins and contain antigen presentation information; 2) immunogenicity (IM) data: peptides that can activate T cell immune responses and contain immune information. The EL training data of the present invention comes from the training data of NetMHCpan4.1 and MHCflurry2.0. The present invention only uses the positive EL samples provided in the two studies, including 124,579 non-redundant samples of MHCflurry2.0 and 197,523 non-redundant samples of NetMHCpan4.1. The present invention deletes samples whose peptide lengths are not in the range of 8-11, and collects the remaining samples to construct a training data set. Subsequently, the present invention randomly samples peptides from the Protein Data Bank (PDB) to produce ten times more negative samples than positive samples. This strategy of randomly splitting peptides from the protein database as negative samples is consistent with NetMHCpan4.1 and mhcflurry2.0. In addition, the present invention obtained IM data for training from three different studies. Before training, the present invention used random peptides obtained from PDB to augment IM data. The number of random peptides is 10 times that of positive samples. These random peptides and experimentally verified non-immunogenic peptides together constitute the negative training data of IM.

[0068] In order to demonstrate the advantages of IEPAPI in predicting antigen presentation, the present invention first compared IEPAPI with two state-of-the-art methods, NetMHCpan4.1 and mhcflurry2.0, on a recently published independent EL test dataset. Secondly, for the prediction of immunogenicity, the present invention compared IEPAPI with existing methods on two independent cancer neoantigen datasets, including a glioblastoma (GBM) dataset and a neoantigen dataset collected by the Tumor Neoantigen Selection Alliance (TESLA).

[0069] The immunogenicity prediction model is divided into an antigen presentation branch and an immunogenicity branch;

[0070] The antigen presentation branch and the immunogenicity branch are both composed of a three-layer fully connected neural network, and the feature vectors extracted by the feature extraction module are passed through two fully connected layers with 1024 and 256 neurons respectively and then batch normalized;

[0071] Both fully connected layers use rectified linear units as activation functions, and the immunogenicity score representing the probability of peptide activating T cells is generated by a fully connected layer with only one neuron.

[0072] In the present invention, the antigen presentation prediction model is called IEPAPI-EL, and the immunogenicity prediction model is called IEPAPI-IM. Figure 3 As shown in Figure 1, IEPAPI-EL aims to predict the probability of antigen presentation from peptide fragments and HLA-I information. Figure 4 As shown, IEPAPI-IM predicts the probability of a peptide activating T cells. The present invention assumes that the input variables X1 and X2 represent the pseudo sequence of the HLA-I protein and the amino acid sequence of the peptide, respectively. The output Y represents the probability of the peptide presented by the HLA-I protein, and the output Z is the immunogenicity probability of the peptide. Both Y and Z are restricted to the range of 0 to 1 by the sigmoid function. The mapping function between input and output can be expressed as Y=f1(X1,X2), Z=g1(X1,X2,Y). Figure 5 As shown in FIG. 1 , the present invention uses a feature extraction module to obtain feature vectors of peptides and HLA-I proteins. IEPAPI-IM and IEPAPI-EL share the same feature extraction module, and the module is trained on a large-scale EL dataset.

[0073] like Figure 6 As shown in Figure 1, in order to verify the effectiveness of the IEPAPI-IM network structure, the present invention constructs a control model named baseline-IM, whose corresponding function mapping is represented by Z = g2(X1, X2). The model baseline-IM abandons the antigen presentation branch, and its feature extraction module is trained on IM data. The detailed parameters of the three models proposed in this study are shown in Figure 1. Figure 7 .

[0074] like Figure 4 As shown in Figure 2, the prediction model IEPAPI-IM can be divided into the antigen presentation branch and the immunogenicity branch. Figure 7As shown, both the antigen presentation branch and the immunogenicity branch are composed of three layers of fully connected neural networks. The 128-dimensional feature vector extracted by the feature extraction module first passes through two fully connected layers with 1024 and 256 neurons respectively. Then, the two fully connected layers are batch normalized to speed up the convergence of the neural network. Both fully connected layers use rectified linear units (ReLU) as activation functions. Finally, the prediction scores of the antigen presentation branch and the immunogenicity branch are generated by a fully connected layer with only one neuron. The present invention uses the sigmoid function as the activation function to limit the output to the range of [0,1]. The antigen presentation branch of IEPAPI-IM fully utilizes the network weights of IEPAPI-EL and remains unchanged in IM data-driven training. Therefore, the loss gradient generated by the IM score is only passed back to the immunogenicity branch. In addition, the network structure difference between the baseline-IM and immunogenicity branches of IEPAPI-IM is that the third fully connected layer of the immunogenicity branch of IEPAPI-IM also takes the EL score of the antigen presentation branch as input. This design is because the network structure of IEPAPI-IM can reflect the connection between antigen presentation and immunogenicity to a certain extent, that is, antigen presentation is an important upstream process of T cell immune response.

[0075] Considering that antigen presentation and immunogenicity prediction are essentially classification problems, the present invention adopts cross entropy (CE) as the loss function, that is,

[0076]

[0077] Where P = [p1, p2, ..., p N ] is the predicted value, G=[g1,g2,…,g N ] is the label and N is the batch size.

[0078] In model training, the present invention first uses a large amount of EL data to train the model IEPAPI-EL. Secondly, the present invention uses the network weights of IEPAPI-EL to initialize the feature extraction block and antigen presentation branch of IEPAPI-IM and freeze them. The training of the immunogenicity branch in IEPAPI-IM is completed on IM data. The present invention adopts a five-fold cross-validation method to find the optimal parameters and the model with the best generalization performance. For EL data or IM data, the present invention randomly divides it into five non-overlapping, equal-sized parts, each of which is used as a validation set in turn, and the remaining four parts are combined into a training set. During the training process, the maximum number of iterations is set to 200 times and the learning rate is set to 0.00005. When the performance on the validation set does not improve for 50 consecutive times, the present invention believes that the network has converged and the training process is terminated early. Finally, the present invention saves the neural network that achieves the best performance on the validation set. In the five-fold cross-validation, the present invention obtains an optimal neural network for each training. The final model is a collection of these five neural networks, and their predicted values ​​are averaged as the final output. The AdamW optimizer was used during training with a weight decay parameter of 0.01 to control the complexity of the network.

[0079] Model training was implemented on a Centos Linux 7.9.2009 (core) system. The CPU is an Intel(R) Xeon(R) CPU E5-2678 v3@2.50GHz with 14 logical processors. The GPU is 8 Tesla K80s, and the available memory for each GPU is 11G. The computer program is based on Python 3.7.9 and the deep learning framework Pytorch1.8.0.

[0080] The present invention uses a variety of evaluation indicators to evaluate the performance of the model, including AUC, AUPRC, accuracy, precision and F1 score:

[0081]

[0082]

[0083]

[0084] Where TP is the number of true positive samples, FP is the number of false positive samples, TN is the number of true negative samples, and FN is the number of false negative samples.

[0085] The present invention downloaded 207,304 human protein sequences from the UniProt database and randomly cut them to generate 100,000 peptides with 9 amino acids. Given a specific HLA-I protein typing, the present invention selected the peptides with the top 1% EL score or IM score. The attention scores corresponding to these 1000 peptides were summed to form the total attention value of the 20 common amino acids at the 9 sites of the peptide. Finally, the cumulative attention values ​​of the 20 amino acids corresponding to the 9 sites were presented in the form of a heat map.

[0086] The following is verified by a specific numerical calculation process. In order to prove the effectiveness of IEPAPI-EL in predicting antigen presentation, the present invention compares it with two long-established methods in this field: NetMHCpan4.1 and MHCflurry2.0. The monoallelic EL training data used in the present invention comes from the training data of NetMHCpan4.1 and MHCflurry2.0. Since the peptide lengths of most of the collected EL data are between 8 and 11, the present invention only discusses peptides with lengths within this range. The output of IEPAPI-EL is a probability score ranging from 0 to 1. For fairness, the present invention uses the EL score of NetMHCpan 4.1 and the presentation score of mhcflurry2.0 for comparison instead of using their %rank values. The positive samples in the independent antigen presentation test data set come from recent mass spectrometry experiments. In order to prevent data leakage, all positive samples in the test data set that appear in the training data are deleted. Negative samples include natural peptides randomly extracted from the protein database PDB. The final antigen presentation test dataset has 289,723 entries, of which 9.09% are positive samples. In addition, 25 common HLA subtypes are covered in the test data, which makes the comparison of models more sufficient.

[0087] The present invention uses AUC and AUPRC to evaluate the presentation prediction performance of each method. The present invention compares the performance of the three models on the entire independent test data set. IEPAPI-EL (AUC = 0.9556, AUPRC = 0.8364) exceeds mhcflurry2.0 (AUC = 0.9449, AUPRC = 0.8248) and NetMHCpan4.1-EL (AUC = 0.9328, AUPRC = 0.8101) in both evaluation indicators. This is due to the efficient encoding of peptides and HLA by the feature extraction module based on the transformer encoder. Secondly, the present invention shows the performance improvement and decline of IEPAPI-EL over existing antigen presentation prediction methods on each HLA-I subtype in the test set. Compared with NetMHCpan4.1-EL, IEPAPI-EL achieved better AUC on all 25 HLA-I subtypes in the test set. The most obvious improvement was in HLA-A*02:01, which is a common HLA-I subtype in the population. Compared with mhcflurry2.0, IEPAPI-EL achieved an improvement in AUC on 76% (19 / 25) of HLA-I subtypes. The subtype with the most obvious AUC improvement was HLA*B15:13.

[0088] When searching for tumor neoantigens, it is crucial to accurately predict the immune peptides that can activate CD8+T cells. In order to demonstrate the advantages of IEPAPI-IM in predicting immunogenicity, the present invention compares it with three existing immunogenicity prediction algorithms, including DeepImmuno, IEDB method, and DeepHLApan. In addition, the present invention also designs a control model, baseline-IM, to show the improvement of IEPAPI-IM in model structure. Figure 4 and Figure 6As shown, the difference between baseline-IM and IEPAPI-IM is that baseline-IM is trained entirely on IM data. In contrast, the feature extraction block and antigen presentation branch of IEPAPI-IM are trained on EL data, but its immunogenicity branch is trained on IM data and incorporates the prediction of antigen presentation into its input. The present invention uses AUC, AUPRC, accuracy, precision and F1 score as evaluation indicators. For baseline-IM, IEPAPI-IM, DeepHLApan and DeepImmuno, whose outputs are distributed between 0 and 1, the present invention uses 0.5 as the classification threshold. For the IEDB method whose output is distributed from -1 to 1, the present invention uses 0 as the classification threshold. It is worth noting that the IEDB method only uses peptide information to predict immunogenicity, while other methods use information of HLA-peptide complexes as input. All immunogenicity prediction algorithms were evaluated on two independent cancer neoantigen test datasets.

[0089] like Fig. 9 , the present invention first evaluated the performance of baseline-IM and IEPAPI-IM on the validation set during the training process. In the initial stage of the training process, the AUC of IEPAPI-IM on the validation set was already much higher than that of baseline-IM. In addition, IEPAPI-IM also reached the highest performance point faster than baseline-IM, which means the effectiveness of the feature extraction module trained with large-scale EL data. As the iterative training progressed, both baseline-IM and IEPAPI-IM overfitted due to insufficient IM training data. However, because the present invention selected the model that performed best on the validation set, the retained IEPAPI-IM model (AUC=0.9289) was still better than the retained baseline-IM model (AUC=0.9190). The above results show that after incorporating the feature extraction block and prediction score of antigen presentation, the performance of IEPAPI-IM is significantly improved.

[0090] Secondly, the present invention compared the models on two independent cancer neoantigen test datasets. One of them is the GBM test dataset, which contains 123 HLA-A*02:01 samples. The other is the TESLA test dataset, which contains 399 samples of 7 HLA-I subtypes. These two test datasets do not overlap with the IM training data. The test results show that it is obvious that baseline-IM does not perform better than DeepImmuno, IEDB method and DeepHLApan on the two neoantigen test datasets. In the comparison of existing methods on the GBM test dataset, the IEDB method achieved a higher AUC than DeepHLApan and DeepImmuno. However, compared with the IEDB method, IEPAPI-IM further improved the AUC. The performance of IEPAPI-IM on the GBM test dataset far exceeds that of the latest method DeepImmuno (AUC increased by 40.0%). Similarly, in the TESLA test dataset, the AUC indicators of IEPAPI-IM are also better than all other methods.

[0091] The present invention sets a classification threshold to evaluate the performance of immunogenicity prediction. IEPAPI-IM achieved the best performance in almost all indicators, including the highest accuracy (GBM: 0.8130, TESLA: 0.7644), precision (GBM: 0.6000, TESLA: 0.1319) and F1 score (GBM: 0.3429, TESLA: 0.2034) on two independent test data sets. In particular, on the TESLA test data set, the prediction accuracy of IEPAPI-IM is almost twice that of DeepImmuno, which indicates that accurate antigen presentation prediction can improve the performance of IEPAPI in immunogenicity prediction.

[0092] Since the present invention uses a feature extraction module based on a transformer encoder to obtain the features of peptides and HLA-I, the self-attention mechanism can help explain the contribution of each residue in the peptide to the model output. After specifying the HLA-I subtype, the present invention uses IEPAPI-EL or IEPAPI-IM to predict 100,000 random 9-mer peptides extracted from the UniProt dataset. The attention scores are calculated for the top 1% of the peptides in the prediction results. Fig.10 As shown, the present invention accumulates their attention scores to obtain an attention heat map. The brighter the color, the higher the cumulative score, and the greater the impact on the output. Fig.10Some of the patterns shown in have certain biological basis. For example, for HLA-B*40:01, it can be seen that GLU (E) at position 2 (P2) has a great influence on both antigen presentation and immunogenicity. Studies have shown that this site belongs to the anchor amino acid and is important for the binding of the peptide to the HLA-B*40:01 protein. Therefore, it will naturally have an important impact on downstream biological processes. In addition, in previous studies, results have shown that patterns such as Lys (K) at P9 of HLA-A*11:01, Trp (W) at P9 of HLA-B57:03, Arg (R) at P9 of HLA-A68:01, and Glu (E) at P2 of HLA-B44:02 are all related to binding. Encouragingly, their effects on antigen presentation and immunogenicity were successfully captured by the IEPAPI method.

[0093] In addition to displaying the patterns of peptides, the calculated attention weights can also be used to reveal the impact of HLA-I subtypes on the final immunogenicity score. For example, the risk of ankylosing spondylitis (AS) was found to be closely associated with HLA-B*27. Current studies have shown that HLA-B*27:04 and HLA-B*27:05 are closely associated with AS, while HLA-B*27:06 and HLA-B*27:09 are not associated or weakly associated with AS. The only difference between HLA-B*27:05 and HLA-B*27:09 is a single amino acid at P116 (Asp and His). Similarly, the difference between HLA-B*27:04 and HLA-B*27:06 is only two amino acids, P114 (His and Asp) and P116 (Asp and Tyr).

[0094] The present invention calculates the immunogenicity scores of 100,000 random 9-mer peptides with the above four HLA-B*27 subtypes. Similar to the attention weight calculation of peptides, the present invention calculates the cumulative attention weight of HLA-B*27 subtypes in the first 1000 score predictions. Fig.11 The normalized attention weights of the four HLA-B*27 pseudo sequences are shown. It can be observed that Asp at P116 of HLA-B*27:04 and HLA-B*27:05 contribute greatly to the output of the model, while Tyr at P116 of HLA-B*27:06 and His at P116 of HLA-B*27:09 have much less effect on the output of the model. This phenomenon suggests that Asp at P116 of HLA-B*27 can strongly affect the risk of developing the autoimmune disease AS through antigen presentation.

[0095] The method of the present invention can simultaneously provide probability scores for antigen presentation and T cell immune response. It is generally believed that only peptides presented on the cell surface can further bind to TCR and generate signals to activate T cells. Therefore, it is very meaningful to explore whether the two outputs of IEPAPI are consistent with the above relationship. Fig.12 As shown, the present invention uses the tool seaborn to calculate and visualize the kernel density estimation of the distribution of prediction scores for positive and negative samples. It can be seen that the recall rate of the independent antigen presentation test data set is 70.4%, and only 17.7% of the presented peptides are predicted to be immunogenic. In addition, 98.8% of the negative samples generated by random sampling were simultaneously determined by the method of the present invention to be neither presentable nor immunogenic, indicating that both EL and IM scores can well identify random peptides. Less than 1% of positive and negative samples were predicted to be non-presentable but immunogenic, indicating that the predictions of antigen presentation and immunogenicity provided by IEPAPI are consistent with the relationship that only presented peptides can further activate T cells.

[0096] like Fig.12 As shown, in terms of judging positive immunogenic peptides, in the GBM test dataset, the EL and IM scores identified 28% and 24% of the real immunogenic peptides, respectively. For the TESLA test dataset, the EL score achieved a high recall rate of 96.3%, while the IM score only identified 44.4% of the positive peptides. In terms of judging non-immunogenic peptides, the ability of the EL score is far inferior to the IM score. On the GBM test dataset, the IM score accurately predicted 95.9% of the experimentally verified non-immunogenic peptides, while the EL score only found 69.4% of them. For the TESLA test dataset, the EL and IM scores found 40.1% and 78.8% of the non-immunogenic peptides, respectively. The above results show that the EL score is suitable for scenarios requiring high recall, while the IM score is suitable for scenarios requiring high precision. Similar to antigen presentation prediction, in the GBM and TESLA datasets, few positive or negative examples are predicted as low EL scores and high IM scores at the same time, which reveals that the method of the present invention can reflect the relationship between antigen presentation and immunogenicity.

[0097] refer to Figure 2 The present invention also discloses an immune epitope prediction system based on the fusion of antigen presentation and immunogenicity characteristics, the system comprising:

[0098] A feature extraction module 110 is used to extract peptide segments and type I human leukocyte antigen protein features from an immune epitope database through a constructed feature extraction module;

[0099] An antigen presentation module 120, for inputting the peptide segment and the type I human leukocyte antigen protein features into a pre-trained antigen presentation prediction model, and obtaining the probability of antigen presentation based on the antigen presentation prediction model;

[0100] The immunogenicity module 130 is used to input the peptide segment and type I human leukocyte antigen protein characteristics and antigen presentation probability information into the immunogenicity prediction model, and obtain the immunogenicity score representing the probability of the peptide segment activating T cells based on the immunogenicity prediction model, thereby realizing the prediction of immune epitopes.

[0101] The immune epitope prediction system based on the fusion of antigen presentation and immunogenicity features provided by the present invention effectively encodes the amino acid sequences of polypeptides and type I human leukocyte antigen proteins through a feature extraction module, achieving advanced performance in antigen presentation prediction; the knowledge of antigen presentation is integrated into the network structure of immunogenicity prediction, reflecting the relationship between the two processes, thereby greatly improving the prediction performance of immunogenicity; the analysis of the output distribution enhances the interpretability of the model and provides guidance for the practical application of searching for cancer neoantigens.

[0102] Fig.13 An example of a physical structure diagram of an electronic device is shown in FIG. Fig.13 As shown, the electronic device may include: a processor 1310, a communication interface 1320, a memory 1330 and a communication bus 1340, wherein the processor 1310, the communication interface 1320, and the memory 1330 communicate with each other through the communication bus 1340. The processor 1310 may call the logic instructions in the memory 1330 to execute the immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics, and the method includes:

[0103] The constructed feature extraction module was used to extract peptide segments and human leukocyte antigen type I protein features from the immune epitope database;

[0104] Inputting the peptide segment and the type I human leukocyte antigen protein features into a pre-trained antigen presentation prediction model, and obtaining the probability of antigen presentation based on the antigen presentation prediction model;

[0105] The peptide segment and type I human leukocyte antigen protein characteristics and antigen presentation probability information are input into the immunogenicity prediction model to realize antigen presentation and immunogenicity characteristics based on the immunogenicity prediction model to obtain an immunogenicity score representing the probability of the peptide segment activating T cells, so as to realize the prediction of immune epitopes.

[0106] In addition, the logic instructions in the above-mentioned memory 1330 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0107] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics provided by the above methods, and the method includes:

[0108] The constructed feature extraction module was used to extract peptide segments and human leukocyte antigen type I protein features from the immune epitope database;

[0109] Inputting the peptide segment and the type I human leukocyte antigen protein features into a pre-trained antigen presentation prediction model, and obtaining the probability of antigen presentation based on the antigen presentation prediction model;

[0110] The peptide segment and type I human leukocyte antigen protein characteristics and antigen presentation probability information are input into the immunogenicity prediction model to realize antigen presentation and immunogenicity characteristics based on the immunogenicity prediction model to obtain an immunogenicity score representing the probability of the peptide segment activating T cells, so as to realize the prediction of immune epitopes.

[0111] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to perform the immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics provided by the above methods, the method comprising:

[0112] The constructed feature extraction module was used to extract peptide segments and human leukocyte antigen type I protein features from the immune epitope database;

[0113] Inputting the peptide segment and the type I human leukocyte antigen protein features into a pre-trained antigen presentation prediction model, and obtaining the probability of antigen presentation based on the antigen presentation prediction model;

[0114] The peptide segment and type I human leukocyte antigen protein characteristics and antigen presentation probability information are input into the immunogenicity prediction model to realize antigen presentation and immunogenicity characteristics based on the immunogenicity prediction model to obtain an immunogenicity score representing the probability of the peptide segment activating T cells, so as to realize the prediction of immune epitopes.

[0115] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0116] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics, characterized in that: include: The constructed feature extraction module was used to extract peptide segments and human leukocyte antigen type I protein features from the immune epitope database; Inputting the peptide segment and the type I human leukocyte antigen protein features into a pre-trained antigen presentation prediction model, and obtaining the probability of antigen presentation based on the antigen presentation prediction model; The peptide segment and type I human leukocyte antigen protein characteristics and antigen presentation probability information are input into the immunogenicity prediction model to realize antigen presentation and immunogenicity characteristics based on the immunogenicity prediction model to obtain an immunogenicity score representing the probability of the peptide segment activating T cells, so as to realize the prediction of immune epitopes.

2. The immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics according to claim 1, characterized in that: The constructed feature extraction module is used to extract peptide segments and type I human leukocyte antigen protein features from the immune epitope database, specifically including: The feature extraction module extracts sequence information based on the transformer encoder and calculates the attention score to mine peptide segments and type I human leukocyte antigen protein sequence sites that have a key impact on the final output of the respective models for the antigen presentation prediction model and the immunogenicity prediction model; Based on the sequence embedding and position encoding method, the amino acid word vector of each site in the sequence is constructed, and the three weight matrices are multiplied by the word vector to obtain three different vectors: Query, Key, and Value; The attention weight of each word vector is calculated by Queries and Keys to generate an attention weight matrix; The elements related to each sequence position in the attention weight matrix are extracted, and their average value is calculated to obtain the attention value of each position, which reflects the influence of the position on the final output of the antigen presentation prediction model and the immunogenicity prediction model.

3. The immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics according to claim 1, characterized in that: The training of the antigen presentation prediction model and the immunogenicity prediction model requires the use of positive sample data and negative sample data of the eluted ligand data and the immunogenicity data; Collect positive sample data of elution culture data and immunogenicity data from the existing database, delete sample data whose peptide length is not within the set range, and collect the remaining sample data to construct a positive training data set; The eluted ligand data and immunogenicity data are peptide sample data randomly extracted from the protein database PDB, and negative samples ten times the number of positive samples are generated to augment the training data. The randomly sampled peptide data and experimentally verified non-immunogenic peptide data together constitute a negative training data set.

4. The immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics according to claim 1, characterized in that: The immunogenicity prediction model is divided into an antigen presentation branch and an immunogenicity branch; The antigen presentation branch and the immunogenicity branch are both composed of a three-layer fully connected neural network, and the features extracted by the feature extraction module are passed through two fully connected layers with 1024 and 256 neurons respectively and then batch normalized; Both fully connected layers use rectified linear units as activation functions, and the immunogenicity score representing the probability of a peptide activating T cells is generated by a fully connected layer with only one neuron.

5. The immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics according to claim 4, characterized in that: The antigen presentation branch of the immunogenicity prediction model fully utilizes the network weights of the antigen presentation prediction model, and its parameters remain unchanged during the training of the immunogenicity prediction model with the participation of immunogenicity data; The loss gradient resulting from the immunogenicity score is only propagated back to the immunogenicity branch.

6. The immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics according to claim 1, characterized in that: The antigen presentation prediction model and the immunogenicity prediction model use a five-fold cross validation method to find better parameters and a prediction model with the best performance; In the training of the prediction model, five neural networks with better performance were obtained based on five-fold cross validation. The final prediction model was a collection of five neural networks. The average of the output values ​​of the five neural networks was calculated as the final output value of the prediction model, and the performance of the prediction model was evaluated using a variety of evaluation indicators.

7. An immune epitope prediction system based on the fusion of antigen presentation and immunogenicity characteristics, characterized in that: The system comprises: A feature extraction module is used to extract peptide segments and type I human leukocyte antigen protein features from an immune epitope database through a constructed feature extraction module; An antigen presentation module, used for inputting the peptide segment and the type I human leukocyte antigen protein features into a pre-trained antigen presentation prediction model, and obtaining the probability of antigen presentation based on the antigen presentation prediction model; The immunogenicity module is used to input the characteristics of peptide segments and type I human leukocyte antigen protein and antigen presentation probability information into the immunogenicity prediction model, and obtain the immunogenicity score representing the probability of peptide segments activating T cells based on the immunogenicity prediction model, thereby realizing the prediction of immune epitopes.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the immune epitope prediction method based on the fusion of antigen presentation and immunogenicity characteristics as described in any one of claims 1 to 6 is implemented.

Citation Information

Patent Citations

  • Method and device for model training, antibody modification and binding site prediction

    CN113764037A

  • Glycopeptide epitope prediction method, storage medium and terminal equipment

    CN113851187A

  • Integrated biomarkers for cancer immunotherapy

    CN115715322A

Cited By

  • Prediction method for binding affinity of human leukocyte antigen class I molecules and peptide fragments

    CN119339780A

  • TCR-pMHC combined prediction method, model and system based on deep learning

    CN121215045A