Method for intelligently recognizing T cell antigen at atomic level

Abstract

The invention discloses a method for intelligently recognizing T cell antigens at an atomic level. The method comprises the following steps: expressing the antigens, HLA (human leukocyte antigen) and TCR (T cell receptor) as a graph taking atoms as nodes and chemical bonds as edges; constructing an intelligent prediction model for combining the antigen at the atomic level with the HLA, and screening the antigen which can be presented by the HLA; constructing an intelligent prediction model for the interaction between the antigen and the TCR at the atomic level, and screening the antigen which can be recognized by the TCR; and synthesizing the T cell cloning frequency to calculate the immunogenicity of the antigen and recognize the high-immunogenicity antigen, thereby forming the intelligent recognition method for the T cell antigen. According to the invention, the problem that most candidate antigens cannot cause immune response due to the fact that T cell antigens are only recognized at a sequence or residue level in the prior art is solved, and tumor neoantigens with immunogenicity can be accurately screened at an atom level.

Description

Technical Field

[0001] The present invention relates to the field of bioinformation technology, in particular to a method for intelligently identifying T cell antigens at the atomic level. Background Art

[0002] T cell antigens are short peptide epitopes that can be presented by major histocompatibility complex (MHC) molecules (also known as human leukocyte antigens, HLA in humans) and recognized by T cell receptors (TCR) on the surface of T cells, which can activate immune responses. T cell antigen recognition is of great significance for infection, autoimmune diseases and tumor immunotherapy, especially in the development of tumor mRNA vaccines. There are a large number of antigens in nature, and the diversity of TCRs generated by VDJ recombination is as high as 10 16 -10 18 , and the binding or interaction between antigens and HLA or TCR is multi-specific. Therefore, biological experimental methods such as high-throughput sequencing, mass spectrometry flow cytometry, and microfluidics cannot verify the immunogenicity of a large number of potential antigens one by one, limiting the recognition of T cell antigens. Therefore, using artificial intelligence to identify T cell antigens is crucial to accelerate the development of mRNA vaccines.

[0003] The two key elements of using artificial intelligence technology to identify T cell antigens are antigen-HLA binding prediction and antigen-TCR interaction prediction. Existing methods only use sequence or residue-level binding information to predict whether an antigen can be presented by HLA or recognized by TCR. Most of the identified T cell antigens fail to trigger T cell immune responses. Recent studies have shown that atomic-level contacts between antigens and TCRs, especially non-covalent interactions, are crucial in determining the activation of T cell immune responses. During the dissociation of antigens and TCRs, newly formed hydrogen bonds or salt bridges can prolong the contact time between immunogenic antigens and TCRs (reverse lock bonds), while non-immunogenic antigens do not form new atomic-level non-covalent interactions with TCRs. Reverse lock bonds have been used to precisely regulate TCR activity and have shown potential in cancer immunotherapy. However, the development of computational methods that can predict antigen-HLA binding and antigen-TCR interactions at the atomic level and further infer antigen immunogenicity has not been explored.

[0004] Therefore, making full use of the information at the atomic level of antigens, HLA and TCR and the crystal structure of the complex to develop computational methods to reveal the binding or interaction mechanism between antigens and HLA or TCR at the atomic level, and comprehensively considering the two processes of antigen presentation and T cell recognition to infer the immunogenicity of the antigen, will greatly improve the screening efficiency and accuracy of T cell antigens and provide a new breakthrough for the development of tumor mRNA vaccines. Summary of the invention

[0005] The purpose of the present invention is to address the defects of the above-mentioned existing T cell antigen recognition technology, and aims to utilize the structural information of the antigen and HLA or TCR complex, combined with artificial intelligence algorithms, to provide a method for efficiently and accurately predicting the binding or interaction between antigen and HLA or TCR at the atomic level, integrating biotechnology and information technology to achieve efficient and accurate recognition of massive T cell antigens.

[0006] In order to achieve the above objectives, the technical solution adopted by the present invention is: a method for intelligently identifying T cell antigens at the atomic level, comprising the following steps: 1) Antigen, HLA, and TCR data representation and feature encoding: The amino acid sequences of antigens, HLA, and TCR are represented as graphs with atoms as nodes and chemical bonds as edges using the amino acid structural formula, and the graphs are digitally encoded in combination with the physical and chemical properties of atoms and chemical bonds; 2) Construction of intelligent prediction model for antigen-HLA binding: Based on the graphical representation of antigen and HLA, an artificial intelligence model is established to predict the binding probability and atomic-level contact sites between antigen and HLA class I or HLA II molecules, and to screen out antigens that can be presented by HLA; 3) Construction of intelligent prediction model for antigen-TCR interaction: Based on the graph representation of antigen and TCR, an artificial intelligence model is established to predict the binding probability and atomic-level contact sites between antigen and TCR, and to screen out antigens that can be recognized by TCR; 4) Construction of intelligent T cell antigen recognition method: Based on the probability of interaction between the antigen and each TCR and the T cell clone frequency, the comprehensive immunogenicity score of the antigen that can be presented by HLA is calculated, and antigens with high immunogenicity are screened to obtain the method for intelligently identifying T cell antigens at the atomic level.

[0007] The above antigen, HLA, and TCR data representation and feature encoding include the following steps: 1.1) According to the HLA polymorphic sites, the amino acids at the corresponding positions were extracted from the full-length sequences of the HLA α and β chains and the HLA pseudo-sequences were formed in order. The CDR3 sequence was extracted from the full-length sequence of the TCR β chain according to the comparison results of the IMGT database; 1.2) Parse each amino acid in the input antigen sequence, HLA pseudo sequence and CDR3 sequence, and convert each amino acid into a molecular structure according to the structural formula of 20 common amino acids; 1.3) Connect the amino acids in sequence, the carboxyl group of the previous amino acid and the amino group of the next amino acid form a peptide bond through dehydration condensation, and hydrogen atoms are added to the structure to meet the valence requirements; 1.4) Convert the molecular structure into a graph with atoms as nodes and chemical bonds as edges. The nodes are digitally encoded by atomic properties (atom type, degree, hybridization state, formal charge, aromaticity, chirality), and the edges are digitally encoded by chemical bond properties (bond type, conjugation, ring, stereo conformation) to obtain a graphical representation of antigens, HLA, and TCR.

[0008] The construction of the above antigen-HLA combination intelligent prediction model includes the following steps: 2.1) Use graph convolutional networks to extract atomic-level features of antigens and HLA, identify atoms that play a key role in the interaction from antigens and HLA respectively, use a multi-head cross-attention mechanism to predict the binding reactivity between antigens and HLA, and adjust the parameters of the prediction model through training and validation set data; 2.2) The Euclidean distance between paired atoms between antigen and HLA is calculated using the resolved 3D structure of the antigen-HLA complex, which is used as prior information to fine-tune the model parameters, guide the graph convolutional network to identify atoms that play a key role in the binding process between antigen and HLA, and predict the contact probability between key atoms.

[0009] The construction of the above-mentioned intelligent prediction model for antigen-TCR interaction includes the following steps: 3.1) Use graph convolutional networks to extract atomic-level features of antigens and TCRs, select atoms that play a key role in the interaction from antigens and TCRs respectively, use a multi-head cross-attention mechanism to predict the binding reactivity between antigens and HLA, and adjust the parameters of the prediction model through training and validation set data; 3.2) The 3D structure of the antigen-TCR complex is used to calculate the Euclidean distance of paired atoms between the antigen and TCR, which is used to fine-tune the model parameters using the resolved prior information, guide the graph convolutional network to identify atoms that play a key role in the interaction between the antigen and TCR, and predict the contact probability between key atoms.

[0010] In the above steps 2.1) and 3.1), the graph convolutional network is used to extract the atomic-level features of antigens, HLA and TCR, including the following steps: S1) The graphical representation of each antigen, HLA or TCR sequence is , represents the initial features of the i-th node, Represents the features of the edge between the i-th node and the j-th node. The initial features of the atom are converted to Space, as shown in formula (1): (1) is a learnable weight matrix, is a nonlinear activation function; S2) The atomic features are processed by an L-layer graph convolutional network. In each iteration, message aggregation, information update, and gated recurrent unit (GRU) processing are used in turn, as shown in formulas (2), (3), and (4): (2) (3) (4) in Indicates that the i-th node is in l Aggregate features of the layers, Indicates that the i-th node is in l The updated features of the layer, Indicates that the i-th node after GRU reset and update is in the l The characteristics of the layer, represents all neighbor nodes of the i-th node, Represents a serial operation, , ; S3) Score each atom through a neural network, as shown in formula (5): (5) , represents the hyperbolic tangent activation function, and the Top-k pooling method is used to select the k atoms with the highest scores to represent the atomic-level features of antigens, HLA, and TCR.

[0011] S4) Calculate the pairwise interaction characteristics F between the Top-k atoms of the antigen and the Top-k atoms of HLA or TCR, as shown in formulas (6) and (7): (6) (7) represents the atomic features of the antigen after being updated by L layers of graph convolution, represents the atomic features of HLA or TCR after being updated by L layers of graph convolution, M represents the number of attention heads, Represents the average value of multi-head attention weights; S5) Based on the interaction feature F, the probability P (value range 0-1) of antigen binding to HLA or interacting with TCR is calculated using a multi-layer perceptron, as shown in formula (8): (8) , a binary discrimination can be made based on the interaction probability, i.e., whether the antigen binds to HLA or interacts with TCR, and the model parameters are optimized using the cross entropy loss function.

[0012] In the above steps 2.2) and 3.2), the model is fine-tuned using the 3D structure information of the complex, including the following steps: S6) Add position information to atomic features through position encoding, as shown in formulas (9), (10), and (11): (9) (10) (11) S7) Calculate the joint score of the antigen and the paired atoms in HLA or TCR, and use the loss function based on the Pearson correlation coefficient to make the joint score negatively correlated with the actual distance, as shown in formulas (12), (13), and (14): (12) (13) (14) is the covariance, is the standard deviation, S and D represent the joint score and actual distance between the antigen and the paired atoms in HLA or TCR, respectively. are the coordinates of the atom in space; S8) While ensuring that the model can identify key atoms, the cross-entropy loss function is used to further fine-tune the model to predict the contact probability of pairs of atoms.

[0013] The above-mentioned T cell antigen intelligent recognition method is constructed, comprising the following steps: 4.1) Using the intelligent prediction model for antigen-HLA binding, the binding strength between antigen and HLA allele is predicted, and antigens with binding strength greater than 0.5 are selected as candidates; 4.2) The antigen-HLA combination intelligent prediction model is used to predict the probability of interaction between the candidate antigen and each TCR clone in the TCR library, and the interaction probability is weighted according to the TCR clone frequency to calculate the immunogenicity score of the i-th candidate antigen, as shown in formula (15): (15) where p ij is the interaction strength between the ith candidate antigen and the jth TCR in the TCR library, freq j is the clonal frequency of the jth TCR; 4.3) Sort candidate antigens according to their immunogenicity scores, select antigens with high immunogenicity scores as T cell antigens, and output the contact sites between the antigens and the atoms of HLA or TCR.

[0014] Beneficial effects of the present invention: The present invention utilizes information at the atomic level of antigens, HLA and TCR, and uses graph neural networks and multi-head attention mechanisms to accurately capture the differences between T cell antigens and non-T cell antigens at the atomic level. By combining rare atomic-level complex crystal structure data with rich sequence-level interaction data, potential atomic-level contact information is effectively integrated from sequence-level data, thereby accurately revealing the contact between antigens and HLA or TCR at the atomic level, solving the problem of limited crystal structure data and insufficient prediction accuracy of antigen binding to HLA or interaction with TCR. The beneficial effects of the present invention specifically include the following: 1. Based on the general structural formula of amino acids and the dehydration condensation between amino acids, the present invention constructs a graph representation with atoms as nodes and chemical bonds as edges, and refines the representation of antigens, HLA and TCR from the sequence level or residue level to the atomic level, laying a foundation for distinguishing the differences between T cell antigens and non-T cell antigens at the atomic level and the contacts between atoms.

[0015] 2. The present invention encodes the graph through the biochemical properties of atoms and chemical bonds, extracts high-dimensional potential features through convolutional neural networks, and uses neural networks to score each atom. It can identify a few atoms that play a key role in the interaction process, while reducing the amount of calculation, realizing atomic-level antigen-HLA binding prediction models and antigen-TCR interaction prediction models, and improving prediction accuracy.

[0016] 3. The present invention first pre-trains the model on rich sequence-level binding data to capture potential atomic-level contact information, and then fine-tunes the model using a small amount of atomic-level complex structure data, thereby solving the problem of limited atomic-level data and achieving accurate prediction of the contact between antigens and HLA or TCR atoms through small sample learning, providing a more comprehensive and clear perspective for meaningful biological discoveries such as key contact site detection and mutation effect prediction.

[0017] 4. The present invention constructs an intelligent T cell antigen recognition method, fully utilizes artificial intelligence methods and omics data analysis technology, comprehensively considers the two processes of antigen presentation by HLA and recognition by TCR, improves the accuracy of T cell antigen recognition, and will greatly accelerate the screening of immunogenic tumor neoantigens and promote the development of clinical tumor mRNA vaccines. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1It is a flow chart of a method for intelligently identifying T cell antigens at the atomic level of the present invention; Figure 2 This is a flowchart of deepAntigen, a T cell antigen recognition framework based on graph convolutional networks; Figure 3 This is the result diagram of deepAntigen predicting the binding performance between antigen and HLA molecules; Figure 4 This is the result diagram of deepAntigen predicting the performance of antigen-TCR interaction; Figure 5 This is the result map of deepAntigen identifying key contact sites at the atomic level; Figure 6 This is the result of an ELISA test to verify the immunogenicity of new tumor antigens identified by deepAntigen in clinical cancer patients. DETAILED DESCRIPTION

[0019] A specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings, but it should be understood that the protection scope of the present invention is not limited by the specific implementation.

[0020] See also Figure 1-Figure 6 The present invention discloses a method for intelligently identifying T cell antigens at the atomic level. Figure 1 As shown, the specific implementation steps are as follows: 1) Data preprocessing: Sequence-level data of antigen-HLA binding determined by experimental methods such as binding affinity measurement and natural elution peptide mass spectrometry identification were obtained from public resource libraries such as IEDB as positive samples, including antigen-HLA class I pairing data and antigen-HLA class II pairing data. Each entry of the data includes two features: antigen sequence and HLA allele type. The HLA allele searches for the corresponding protein sequence according to the IMGT database. For each antigen, an antigen sequence of equal length was randomly cut on the gene where it was located as a negative sample. In addition, atomic-level antigen-HLA complex crystal structure data were obtained from the protein data bank and the atomic pair distance matrix between antigen and HLA was calculated. Atom pairs with a distance less than 5Å were defined as contact atom pairs, and the remaining atom pairs were non-contact atom pairs. Sequence-level data of antigen-TCR interaction verified by biological experiments were collected from databases such as VDJdb, PIRD, McPas-TCR and ImmuneAccess as positive samples. Each entry of the data includes antigen sequence and TCR sequence. For each antigen, non-interacting TCRs were randomly sampled from about 600 million inactive TCRs from healthy people as negative samples to balance positive samples. In addition, atomic-level antigen-TCR complex crystal structure data were obtained from the protein data bank and the atomic pair distance matrix between antigen and TCR was calculated. Atom pairs with a distance less than 5Å were defined as contact atom pairs, and the rest of the atom pairs were defined as non-contact atom pairs.

[0021] 2) Antigen, HLA, and TCR data representation and feature encoding: The amino acid sequences of antigens, HLA, and TCR are represented as graphs with atoms as nodes and chemical bonds as edges using the amino acid structural formula, and the graphs are digitally encoded in combination with the physical and chemical properties of atoms and chemical bonds, laying the foundation for identifying T cell antigens using atomic-level features, such as Figure 2 As shown in a and b.

[0022] 2.1) According to the HLA polymorphic sites, the amino acids at the corresponding positions were extracted from the full-length sequences of the HLA α and β chains and the HLA pseudo-sequences were formed in order. The complementary determining region 3 (CDR3) sequence was extracted from the full-length sequence of the TCR β chain according to the comparison results of the IMGT database; 2.2) Parse each amino acid in the input antigen sequence, HLA pseudo sequence and CDR3 sequence, and convert each amino acid into a molecular structure according to the structural formula of 20 common amino acids; 2.3) Connect the amino acids in sequence, the carboxyl group of the previous amino acid and the amino group of the next amino acid form a peptide bond through dehydration condensation, and hydrogen atoms are added to the structure to meet the valence requirements; 2.4) Convert the molecular structure into a graph with atoms as nodes and chemical bonds as edges. The nodes are represented as attribute vectors of length 25 by one-hot encoding of atomic attributes (atom type, degree, hybridization state, formal charge, aromaticity, chirality), and the edges are represented as attribute vectors of length 11 by one-hot encoding of chemical bond attributes (bond type, conjugation, ring, stereo conformation), and obtain the graph representation of antigens, HLA, and TCR; 2.5) During the model training phase, some nodes and their associated edges in the graph are randomly deleted with a probability of 5% to perform data enhancement on the graph, expand the data set, and avoid model overfitting. During the model testing phase, this process is ignored.

[0023] 3) Construction of an intelligent prediction model for antigen-HLA binding: Based on the graphical representation of antigens and HLA, an artificial intelligence model is established to predict the binding probability and atomic-level contact sites between antigens and HLA class I or HLA II molecules, and to screen out antigens that can be presented by HLA.

[0024] 3.1) Use graph convolutional networks to extract atomic-level features of antigens and HLAs, identify atoms that play a key role in the interaction from antigens and HLAs respectively, use a multi-head cross-attention mechanism to predict the binding reactivity between antigens and HLAs, and adjust the parameters of the prediction model through training and validation set data, such as Figure 2 As shown in a.

[0025] 3.1.1) The graphical representation of each antigen or HLA sequence is , represents the 25-dimensional initial features of the i-th node, Represents the 11-dimensional feature of the edge between the i-th node and the j-th node. The initial atomic features are converted to Space, h=128, as shown in formula (1): (1) is a learnable weight matrix, is a nonlinear activation function; 3.1.2) The atomic features are processed by a graph convolutional network with L = 5 layers. In each iteration, message aggregation, information update and gated recurrent unit GRU are used in turn, as shown in formulas (2), (3) and (4): (2) (3) (4) in Indicates that the i-th node is in l Aggregate features of the layers, Indicates that the i-th node is in l The updated features of the layer, Indicates that the i-th node after GRU reset and update is in the l The characteristics of the layer, represents all neighbor nodes of the i-th node, Represents a serial operation, , ; 3.1.3) Score each atom through a neural network, as shown in formula (5): (5) , represents the hyperbolic tangent activation function, and the Top-k pooling method is used to select the k atoms with the highest scores to represent the antigen and HLA atomic-level features. For antigen-HLA class I molecule binding prediction, the k value is set to 10; for antigen-HLA class II molecule binding prediction, the k value is set to 15.

[0026] 3.1.4) Calculate the pairwise interaction characteristics F between the Top-k atoms of the antigen and the Top-k atoms of the HLA, as shown in formulas (6) and (7): (6) (7) represents the atomic features of the antigen after being updated by L=5 layers of graph convolution. represents the HLA or TCR atomic features after being updated by L layers of graph convolution, M represents the number of attention heads is 4, Represents the average value of multi-head attention weights; 3.1.5) Based on the interaction feature F, the probability P (value range 0-1) of antigen binding to HLA is calculated using a multi-layer perceptron, as shown in formula (8): (8) , a binary judgment can be made based on the interaction probability, i.e., whether the antigen binds to HLA, and the model parameters are optimized using the cross entropy loss function; 3.1.6) The sequence-level antigen-HLA binding data were split into training and test sets in a ratio of 8:2, and the model parameters were optimized on the training set through ten-fold cross validation.

[0027] 3.2) The Euclidean distances of paired atoms between antigen and HLA are calculated using the resolved 3D structure of the antigen-HLA complex, which is used as prior information to fine-tune the model parameters, guide the graph convolutional network to identify atoms that play a key role in the binding process between antigen and HLA, and predict the contact probability between key atoms, such as Figure 2 As shown in c.

[0028] 3.2.1) Add position information to atomic features through position encoding, as shown in formulas (9), (10), and (11): (9) (10) (11) 3.2.2) Calculate the joint score of the antigen and the paired atoms in HLA or TCR. The loss function based on the Pearson correlation coefficient makes the joint score negatively correlated with the actual distance, as shown in formulas (12), (13), and (14): (12) (13) (14) is the covariance, is the standard deviation, S and D represent the joint score and actual distance between the antigen and the paired atoms in HLA or TCR, respectively. are the coordinates of the atom in space; 3.2.3) While ensuring that the model can identify key atoms, the cross-entropy loss function is used to further fine-tune the model to predict the contact probability of pairs of atoms; 3.2.4) The atomic-level antigen-HLA crystal structure data were divided into training set and validation set according to the leave-one-out method, and the model parameters were optimized through cross-validation.

[0029] 4) Construction of intelligent prediction model for antigen-TCR interaction: Based on the graph representation of antigen and TCR, an artificial intelligence model is established to predict the binding probability and atomic-level contact sites between antigen and TCR, and to screen out antigens that can be recognized by TCR, such as Figure 2 As shown in a, c. Using the antigen-TCR sequence-level binding data, the antigen-TCR interaction prediction model was constructed according to the specific implementation steps 3.1.1)-3.1.6); using the antigen-TCR atomic-level complex crystal structure data, the antigen-TCR atomic-level contact site prediction model was constructed according to the specific implementation steps 3.2.1)-3.2.4). The difference is that the antigen and TCR screen k=20 key atoms.

[0030] 5) Construction of intelligent T cell antigen recognition method: Based on the probability of interaction between the antigen and each TCR and the T cell clone frequency, the comprehensive immunogenicity score of the antigen that can be presented by HLA is calculated, and antigens with high immunogenicity are screened to obtain the method for intelligently identifying T cell antigens at the atomic level.

[0031] 5.1) Use the intelligent prediction model for antigen-HLA binding to predict the binding strength between antigen and HLA allele, and select antigens with binding strength greater than 0.5 as candidates; 5.2) The antigen-HLA combination intelligent prediction model is used to predict the probability of interaction between the candidate antigen and each TCR clone in the TCR library, and the interaction probability is weighted according to the TCR clone frequency to calculate the immunogenicity score of the i-th candidate antigen, as shown in formula (15): (15) where p ij is the interaction strength between the ith candidate antigen and the jth TCR in the TCR library, freq j is the clonal frequency of the jth TCR; 5.3) Sort candidate antigens according to their immunogenicity scores, select antigens with high immunogenicity scores as T cell antigens, and output the contact sites between the antigens and the atoms of HLA or TCR.

[0032] 6) Model performance evaluation: The performance of the model is evaluated using the test set. In the present invention, both sequence-level binding prediction and atomic-level contact prediction are binary classification tasks. A confusion matrix is ​​generated using binding probability or contact probability, including true positives (TP), true negatives (TN), false positives (FP) and false negatives (FN). Then, the sensitivity and specificity can be calculated as shown in formulas (16) and (17).

[0033] (16) (17) The receiver operating characteristic (ROC) curve can be drawn based on the specificity and sensitivity, and the area under the ROC curve (AUROC) can be calculated to quantitatively evaluate the overall performance of the model.

[0034] In addition, the precision can be calculated according to formula (18).

[0035] (18) The recall rate (Recall) can be calculated according to formula (16). Based on the recall rate and precision, a precision-recall rate (PR) curve can be drawn, and the area under the PR curve (AUPR) can be calculated to quantitatively evaluate the performance in identifying positive samples.

[0036] Figure 3 a, b show the performance of antigen-HLA class I molecule and antigen-HLA class II molecule binding reactivity prediction, respectively. Figure 4 The performance of antigen-TCR interaction prediction is demonstrated. Figure 5 It shows that the atomic-level contact site prediction reflects the actual binding conformation. Figure 6 The new tumor antigens identified by deepAntigen in breast cancer and lung cancer patients were demonstrated to be immunogenic by ELISPOT assay.

[0037] In summary, the present invention expands to the atomic level on the basis of sequence or residues, designs an artificial intelligence-based computational method to identify T cell antigens from the biological mechanism, comprehensively considers antigen presentation and T cell recognition processes, and identifies whether the antigen can activate an immune response in one stop, and provides high-precision atomic-level contact sites between antigens and HLA or TCR, overcoming the coarse-grained and low-precision problems of the prior art in identifying T cell antigens, and can efficiently and accurately screen T cell antigens and expand to the prediction of interactions between small molecule drugs, nucleic acids, etc. and proteins with atomic precision.

[0038] The above disclosures are only several specific embodiments of the present invention. However, the embodiments of the present invention are not limited thereto. Any changes that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A method for intelligently identifying T cell antigens at the atomic level, characterized in that: The steps include: 1) Antigen, HLA, and TCR data representation and feature encoding: The amino acid sequences of antigens, HLA, and TCR are represented as graphs with atoms as nodes and chemical bonds as edges using the amino acid structural formula, and the graphs are digitally encoded in combination with the physical and chemical properties of atoms and chemical bonds; 2) Construction of intelligent prediction model for antigen-HLA binding: Based on the graphical representation of antigen and HLA, an artificial intelligence model is established to predict the binding probability and atomic-level contact sites between antigen and HLA class I or HLA II molecules, and to screen out antigens that can be presented by HLA; 3) Construction of intelligent prediction model for antigen-TCR interaction: Based on the graph representation of antigen and TCR, an artificial intelligence model is established to predict the binding probability and atomic-level contact sites between antigen and TCR, and to screen out antigens that can be recognized by TCR; 4) Construction of intelligent T cell antigen recognition method: Based on the probability of interaction between the antigen and each TCR and the T cell clone frequency, the comprehensive immunogenicity score of the antigen that can be presented by HLA is calculated, and antigens with high immunogenicity are screened to obtain the method for intelligently identifying T cell antigens at the atomic level.

2. A method for intelligently identifying T cell antigens at the atomic level according to claim 1, characterized in that: The antigen, HLA, TCR data representation and feature encoding comprises the following steps: 1) According to the HLA polymorphic sites, the amino acids at the corresponding positions were extracted from the full-length sequences of the HLA α and β chains and the HLA pseudo-sequences were formed in order. The CDR3 sequence was extracted from the full-length sequence of the TCR β chain according to the comparison results of the IMGT database; 2) Analyze each amino acid in the input antigen sequence, HLA pseudo sequence and CDR3 sequence, and convert each amino acid into a molecular structure according to the structural formula of common amino acids; 3) Connect the amino acids in sequence, the carboxyl group of the previous amino acid and the amino group of the next amino acid form a peptide bond through dehydration condensation, and hydrogen atoms are added to the structure to meet the valence requirements; 4) Convert the molecular structure into a graph with atoms as nodes and chemical bonds as edges. The nodes are digitally encoded by atomic properties, and the edges are digitally encoded by chemical bond properties to obtain a graph representation of antigens, HLA, and TCR.

3. The method for intelligently identifying T cell antigens at the atomic level according to claim 1, characterized in that: The construction of the antigen-HLA combination intelligent prediction model includes the following steps: 1) Use graph convolutional networks to extract atomic-level features of antigens and HLA, identify atoms that play a key role in the interaction from antigens and HLA respectively, use a multi-head cross-attention mechanism to predict the binding reactivity between antigens and HLA, and adjust the parameters of the prediction model through training and validation set data; 2) The Euclidean distance between paired atoms between antigen and HLA is calculated using the resolved 3D structure of the antigen-HLA complex, which is used as prior information to fine-tune the model parameters, guide the graph convolutional network to identify atoms that play a key role in the binding process between antigen and HLA, and predict the contact probability between key atoms.

4. The method for intelligently identifying T cell antigens at the atomic level according to claim 1, characterized in that: The construction of the intelligent prediction model of antigen-TCR interaction includes the following steps: 1) Use graph convolutional networks to extract atomic-level features of antigens and TCRs, select atoms that play a key role in the interaction from antigens and TCRs respectively, use a multi-head cross-attention mechanism to predict the binding reactivity between antigens and HLA, and adjust the parameters of the prediction model through training and validation set data; 2) The 3D structure of the antigen-TCR complex is used to calculate the Euclidean distance of paired atoms between the antigen and TCR, which is used to fine-tune the model parameters using the resolved prior information, guide the graph convolutional network to identify atoms that play a key role in the interaction between the antigen and TCR, and predict the contact probability between key atoms.

5. A method for intelligently identifying T cell antigens at the atomic level according to claim 3 or 4, characterized in that: The method of extracting atomic-level features of antigens, HLA and TCR using a graph convolutional network includes the following steps: 1) The graphical representation of each antigen, HLA or TCR sequence is , represents the initial features of the i-th node, Represents the features of the edge between the i-th node and the j-th node. The initial features of the atom are converted to space, as shown in formula (1): （1） is a learnable weight matrix, is a nonlinear activation function; 2) The atomic features are processed by an L-layer graph convolutional network. In each iteration, message aggregation, information update, and gated recurrent unit (GRU) processing are used in turn, as shown in formulas (2), (3), and (4): （2） （3） （4） in Indicates that the i-th node is in l Aggregate features of the layers, Indicates that the i-th node is in l The updated features of the layer, It means that the i-th node is in the first l The characteristics of the layer, represents all neighbor nodes of the i-th node, Represents a serial operation, , ; 3) Score each atom through a neural network, as shown in formula (5): （5） , represents the hyperbolic tangent activation function, and the Top-k pooling method is used to select the k atoms with the highest scores to represent the atomic-level features of antigens, HLA, and TCR.

6. A method for intelligently identifying T cell antigens at the atomic level according to claim 3 or 4, characterized in that: The method of using a multi-head cross attention mechanism to predict the interaction between antigen and HLA or TCR comprises the following steps: 1) Calculate the pairwise interaction characteristics F between the Top-k atoms of the antigen and the Top-k atoms of HLA or TCR, as shown in formulas (6) and (7): （6） （7） represents the atomic features of the antigen after being updated by L layers of graph convolution, represents the atomic features of HLA or TCR after being updated by L layers of graph convolution, M represents the number of attention heads, Represents the average value of multi-head attention weights; 2) Based on the interaction feature F, the probability P (value range 0-1) of antigen binding to HLA or interacting with TCR is calculated using a multi-layer perceptron, as shown in formula (8): （8） , a binary discrimination can be made based on the interaction probability, i.e., whether the antigen binds to HLA or interacts with TCR, and the model parameters are optimized using the cross entropy loss function.

7. A method for intelligently identifying T cell antigens at the atomic level according to claim 3 or 4, characterized in that: The method of fine-tuning the model using the 3D structural information of the complex comprises the following steps: 1) Add position information to atomic features through position encoding, as shown in formulas (9), (10), and (11): （9） （10） （11） 2) Calculate the joint score of the antigen and the paired atoms in HLA or TCR. The loss function based on the Pearson correlation coefficient makes the joint score negatively correlated with the actual distance, as shown in formulas (12), (13), and (14): （12） （13） （14） is the covariance, is the standard deviation, S and D represent the joint score and actual distance between the antigen and the paired atoms in HLA or TCR, respectively. are the coordinates of the atom in space; 3) While ensuring that the model can identify key atoms, the cross-entropy loss function is used to further fine-tune the model to predict the contact probability of pairs of atoms.

8. The method for intelligently identifying T cell antigens at the atomic level according to claim 1, characterized in that: The construction of the T cell antigen intelligent recognition method comprises the following steps: 1) Use the intelligent prediction model for antigen-HLA binding to predict the binding strength between antigen and HLA allele, and select antigens with high binding strength as candidates; 2) The antigen-HLA combination intelligent prediction model is used to predict the probability of interaction between the candidate antigen and each TCR clone in the TCR library, and the interaction probability is weighted according to the TCR clone frequency to calculate the immunogenicity score of the i-th candidate antigen, as shown in formula (15): （15） where p ij is the interaction strength between the ith candidate antigen and the jth TCR in the TCR library, freq j is the clonal frequency of the jth TCR; 3) Sort candidate antigens according to their immunogenicity scores, select antigens with high immunogenicity scores as T cell antigens, and output the contact sites between the antigens and the atoms of HLA or TCR.

Images