CN115497564A - A method for establishing an identification antigen model and an identification method for an antigen - Google Patents

Abstract

The invention provides an antigen identification model establishing method and an antigen identification method, and relates to the technical field of neural networks and immune response prediction. The invention relates to a method for establishing an antigen identification model, which comprises the following steps: constructing a neural network, wherein the neural network comprises a TCR feature extraction neural network, a pMHC feature extraction neural network and an antigen identification neural network; constructing a data set, wherein the data set comprises TCRCDR3, HLAI and antigen sequences; and inputting the data set into a neural network, and training the neural network to establish an antigen identification model. The technical scheme of the invention can realize the characteristic extraction of the sequence information of the TCRCDR3, the HLAI and the antigen, thereby identifying the related antigen of the activated T cell.

Description

A method for establishing an identification antigen model and an identification method for an antigen

technical field

The invention relates to the technical field of neural network and immune response prediction, in particular to a method for establishing an antigen identification model and an antigen identification method.

Background technique

Immunotherapy is a new technology that can be used in cancer treatment. The most important step in this method is the activation of the human adaptive immune response, namely, the recognition and destruction of diseased cells by mediating cytotoxic T lymphocytes.

In the prior art, direct detection methods, modeling and prediction methods, and machine learning methods are generally used to identify T cell antigens. However, these methods use limited data in the implementation process, so the peptides that can be identified Due to the complexity of this response, our understanding of which antigens can activate T lymphocyte immune responses is still very limited.

Contents of the invention

The technical problem to be solved by the present invention is how to realize the identification of relevant antigens of activated T cells.

In order to solve the above problems, the present invention provides a method for establishing an identification antigen model, comprising: constructing a neural network, wherein the neural network includes a TCR feature extraction neural network, a pMHC feature extraction neural network, and an antigen identification neural network; constructing a data set, Wherein, the data set includes TCR CDR3, HLA I and antigen sequences; the data set is input into the neural network, and the neural network is trained to establish an antigen identification model.

Preferably, the Z descriptor is used to describe the TCR CDR3, HLA I and antigen sequences in the data set, and the described matrix is normalized to determine the TCR CDR3 sequence matrix, HLA I sequence matrix and antigen sequence matrix respectively.

Preferably, the data set is input into the neural network, and the neural network is trained, including: using the TCR CDR3 sequence matrix as a training set, and training the TCR feature extraction neural network; The CDR3 sequence matrix is input into the trained TCR feature extraction neural network to obtain the TCR feature vector;

Using the HLA I sequence matrix and the antigen sequence matrix as a training set, the pMHC feature extraction neural network is trained; the HLA I sequence matrix and the antigen sequence matrix are input into the trained pMHC feature extraction In the neural network, the pMHC feature vector is obtained;

The TCR feature vector and the pMHC feature vector are used as a training set to train the antigen identification neural network.

Preferably, before training the antigen identification neural network, the method for establishing an antigen identification model further includes: performing balancing processing on the training set composed of the TCR feature vector and the pMHC feature vector using the SMOTE algorithm.

Preferably, training the neural network for antigen identification includes: taking the state of whether the TCR feature vector and the pMHC feature vector can be combined as a classification label.

Preferably, the training set composed of the TCR feature vector and the pMHC feature vector includes artificial negative binding data.

Preferably, the TCR feature extraction neural network includes an encoder module, a feature extraction module and a decoder module;

The convolution layer module adopted by the encoder module includes four layers of Cov1D layers;

The feature extraction module includes a fully connected layer for outputting TCR sequence features;

The decoder module includes four Conv1D layers.

Preferably, the pMHC feature extraction neural network includes an HLA feature extraction module, an antigen feature extraction module, a feature extraction module and a label training module;

The HLA feature extraction module includes four layers of Cov1D layers, one layer of Reshape layer, and one layer of fully connected layer;

The antigen feature extraction module includes four layers of Cov1D layers, one layer of Reshape layer, and one layer of fully connected layer;

The feature extraction module includes a fully connected layer for outputting pMHC sequence features;

The label training module includes two fully connected layers.

Preferably, the antigen identification neural network includes a TCR feature learning module, a pMHC feature learning module and an output module;

The TCR feature learning module includes two fully connected layers;

The pMHC feature learning module includes two fully connected layers;

The output module employs three fully connected layers for outputting antigen identification results.

The identification antigen model establishment method described in the present invention, by describing the sequence features, using different neural networks to reduce the dimensionality of the sequence features, and using the supervised deep learning neural network, by learning the sequence features after dimensionality reduction, the activation of identification is realized. The target of T cell-associated antigen; meanwhile, in order to ensure the training effect of the model, the present invention uses the SMOTE method to reconstruct the training data set, so that the data in the training set is more balanced, avoiding over-fitting of a small amount of data, and improving the model effect; in addition , in the case of a fixed model structure, try a variety of training set and test set ratios, each ratio for multiple training, the general high verification rate proves that the model has high stability and prediction accuracy; finally, this A series of neural networks designed by the invention can extract the features of the sequence information of TCR CDR3, HLA I and antigens, so as to identify the relevant antigens of activated T cells.

The present invention also provides a method for identifying antigens, comprising: obtaining the TCR CDR3, HLA I and antigen sequences that need to be detected, and inputting the TCR CDR3, HLA I and antigen sequences that need to be detected into any of the identification antigens described above In the identification antigen model established by the model building method, the identification result of the antigen is obtained. The method for identifying antigens has the same advantages as the above method for identifying antigen models over the prior art, and will not be repeated here.

Description of drawings

Fig. 1 is the flow chart of the establishment method of identification antigen model described in the embodiment of the present invention;

Fig. 2 is the structural representation of the TCR feature extraction neural network described in the embodiment of the present invention;

FIG. 3 is a schematic structural diagram of the pMHC feature extraction neural network described in the embodiment of the present invention;

Fig. 4 is a schematic structural diagram of the antigen identification neural network described in the embodiment of the present invention;

Figure 5 is a schematic flow chart of the identification of activated T cell-associated antigens described in the embodiment of the present invention;

Fig. 6 is one of the structural schematic diagrams of the establishment device for identifying antigen models described in the embodiment of the present invention;

Fig. 7 is the second schematic diagram of the structure of the device for identifying antigen models described in the embodiment of the present invention.

detailed description

In order to make the above objects, features and advantages of the present invention more comprehensible, specific embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings.

As shown in Figure 1, an embodiment of the present invention provides a method for establishing an antigen identification model, including the following steps: constructing a neural network, wherein the neural network includes a TCR feature extraction neural network, a pMHC feature extraction neural network, and an antigen identification neural network Constructing a data set, wherein the data set includes TCR CDR3, HLA I and antigen sequences; inputting the data set into the neural network, and training the neural network to establish a model for identifying antigens.

Specifically, the method for establishing an antigen identification model includes: constructing a TCR feature extraction neural network, and using the TCR CDR3 sequence matrix as the input of the TCR feature extraction module for unsupervised training. Train an autoencoder neural network that can extract TCR sequence features; pMHC feature extraction neural network, HLA I sequence matrix and antigen sequence sequence matrix are used as the input of pMHC feature extraction module, and train a convolution-based algorithm that can extract pMHC sequence features Layer neural network; antigen identification neural network, TCR feature vector and pMHC feature vector as a training set, known whether the two can be combined with the state as a classification label, neural network training for antigen identification.

The data set was constructed, and the TCR CDR3, HLA I and antigen sequences in the data set were described by Z descriptor, and the described matrix was normalized to determine the TCR CDR3 sequence matrix, HLA I sequence matrix and antigen sequence matrix respectively.

Input the data set into the neural network, train the neural network to establish an identification antigen model, use the TCR CDR3 sequence matrix as the training set, and train the TCR feature extraction neural network; input the TCR CDR3 sequence matrix into the trained TCR feature extraction In the neural network, the TCR feature vector is obtained;

The HLA I sequence matrix and the antigen sequence matrix are used as the training set to train the pMHC feature extraction neural network; the HLA I sequence matrix and the antigen sequence matrix are input into the trained pMHC feature extraction neural network to obtain the pMHC feature vector;

The TCR feature vector and pMHC feature vector were used as the training set to train the antigen identification neural network.

In this embodiment, by constructing a neural network, input the data set into the constructed neural network, train the neural network to obtain the feature vector, use the feature vector as the training set, train the antigen identification neural network, and establish an identification antigen model , to achieve the goal of identifying antigens associated with activated T cells.

Preferably, the method for establishing an antigen identification model further includes: using the Z descriptor to describe the TCR CDR3, HLA I and antigen sequences in the data set, and normalizing the described matrix to respectively determine the TCR CDR3 sequence matrix, HLA I Sequence Matrix and Antigen Sequence Matrix.

Specifically, the Z-descriptor is a highly condensed descriptor derived from principal component analysis (PCA) of 29 experimental or calculated physicochemical properties of all naturally occurring amino acids, with each amino acid in a protein sequence represented by The three Z descriptors represent the hydrophobicity, spatial characteristics and polarity of amino acids, respectively; each protein sequence is described as a 3*N matrix, where N represents the length of the protein sequence. In the normalized matrix, each value is between -1 and 1.

In this embodiment, by normalizing the described matrix, the subsequent data processing is more convenient and the learning speed is accelerated.

Preferably, said inputting said data set into said neural network, and training said neural network includes:

The TCR CDR3 sequence matrix is used as a training set to train the TCR feature extraction neural network; the TCR CDR3 sequence matrix is input into the trained TCR feature extraction neural network to obtain a TCR feature vector;

Using the HLA I sequence matrix and the antigen sequence matrix as a training set, the pMHC feature extraction neural network is trained; the HLA I sequence matrix and the antigen sequence matrix are input into the trained pMHC feature extraction In the neural network, the pMHC feature vector is obtained;

The TCR feature vector and the pMHC feature vector are used as a training set to train the antigen identification neural network.

Specifically, the binding affinity between HLA and the antigen is used as the training label of the module to perform neural network training, and the binding affinity is a value between 0 and 1. The closer the value is to 1, the higher the binding affinity between the two, on the contrary, the lower the binding affinity between the two. Through this step, the neural network can combine a two-dimensional matrix of HLA features and a two-dimensional matrix of antigen features through feature Dimensionality reduction was extracted, and binding features were extracted based on binding affinity. Extracted pMHC sequence features The features are represented by a series of vectors with a length of 50.

Preferably, before the training of the neural network for antigen identification, the method for establishing the antigen identification model further includes: using the SMOTE algorithm to balance the training set composed of the TCR feature vector and the pMHC feature vector.

Specifically, the SMOTE algorithm is used to balance the training set, which also includes: constructing a new minority class sample, and combining the new sample with the original data to form a new training set. In the published data sets, there are few reports on the combination of negative data, and the imbalance problem will lead to overfitting problems in the process of training the deep learning network. STOME randomly selects its adjacent samples according to the existing minority class samples. Synthesize a new minority sample as a training set to train the neural network.

In this embodiment, the SMOTE algorithm is used to balance the training set, analyze and simulate the samples of a few categories, and add artificially simulated new samples to the data set, so that the categories in the original data set are no longer seriously unbalanced, in order to solve This imbalance problem prevents overfitting problems.

Preferably, the training of the antigen identification neural network includes: using the state of whether the TCR feature vector and the pMHC feature vector can be combined as a classification label.

Specifically, the obtained TCR eigenvector and pMHC eigenvector are used as the training set to train the antigen identification neural network. It is known whether the two can be combined with the state as the classification label, and the neural network training of the antigen identification module is performed. The classification label uses 0 and 1 represent the negative and positive combinations of the two, respectively.

In this embodiment, classification labels are used to represent the combined states of the two, so that the neural network can better identify antigens in subsequent training and learning.

Preferably, training the neural network for antigen identification also includes: testing the trained neural network on a test set.

Specifically, keep the neural network structure unchanged, change the data ratio of the verification set and the test set, and perform 50 trainings for each ratio, test the trained neural network on the test set, and select the network with the best effect. As the final antigen identification module.

In this embodiment, multiple trainings are adopted, and the verification rate of the model is maintained at a high level continuously under the condition of model parameter changes, which ensures the accuracy and stability of the model.

Preferably, the training set composed of the TCR feature vector and the pMHC feature vector includes artificial negative binding data.

Specifically, randomly replace any one or more amino acids in the TCR sequence in the research database. If the sequence does not exist in the real TCR sequence database, it will be used as an artificial negative TCR sequence and put into the negative TCR sequence library. By matching any TCR sequence in the negative TCR sequence library with any known pMHC sequence, a binding negative TCR-pMHC pair can be obtained.

In this embodiment, by adding a small amount of artificial negative binding data in the training set, the antigen identification neural network can learn the characteristics of the artificial negative binding data, so that the network can learn more error modes and make its training more accurate.

Preferably, as shown in FIG. 2, the TCR feature extraction neural network includes an encoder module, a feature extraction module and a decoder module;

The convolution layer module adopted by the encoder module includes four layers of Cov1D layers;

The feature extraction module includes a fully connected layer for outputting TCR sequence features;

The decoder module includes four Conv1D layers.

Specifically, the neural network is an autoencoder neural network. The network is unsupervised training, and the parameters are updated by comparing the similarity between the input and the output, so that the similarity between the input and the output reaches the specified threshold; the structure of the encoder module and the decoder module is mirror-symmetrical; the first two layers of Conv1D in the encoder module The convolution kernel size of the first layer is 3, and the convolution kernel size of the last two layers is 2.

In this embodiment, the TCR sequence feature extraction module and submodules can effectively extract the TCR CDR3 sequence matrix to improve the training effect of the neural network.

Preferably, as shown in Figure 3, the pMHC feature extraction neural network includes an HLA feature extraction module, an antigen feature extraction module, a feature extraction module and a label training module;

The HLA feature extraction module includes four layers of Cov1D layers, one layer of Reshape layer, and one layer of fully connected layer;

The antigen feature extraction module includes four layers of Cov1D layers, one layer of Reshape layer, and one layer of fully connected layer;

The feature extraction module includes a fully connected layer for outputting pMHC sequence features;

The label training module includes two fully connected layers.

Specifically, the convolution kernel size of the Conv1D layer in the HLA feature extraction module is 2, the convolution kernel size of the Conv1D layer in the antigen feature extraction module is 2, and the Reshape layer reduces the two-dimensional structure of the output result of the convolution layer to one-dimensional structure.

In this embodiment, according to the pMHC sequence feature extraction module and submodules, the TCRCDR3 sequence matrix can be effectively extracted to improve the training effect of the neural network.

Preferably, as shown in Figure 4, the antigen identification neural network includes a TCR feature learning module, a pMHC feature learning module and an output module;

The TCR feature learning module includes two fully connected layers;

The pMHC feature learning module includes two fully connected layers;

The output module employs three fully connected layers for outputting antigen identification results.

Specifically, the input of the network is the output result of the TCR sequence feature extraction network and the pMHC sequence feature extraction network.

In this embodiment, the antigen sequence matrix can be effectively extracted according to the antigen identification module, so as to improve the training effect of the neural network.

As shown in Figure 5, another embodiment of the present invention also provides a method for identifying antigens, including: obtaining the TCR CDR3, HLA I and antigen sequences that need to be detected, and inputting the TCR CDR3, HLA I and antigen sequences that need to be detected for identification In the identified antigen model established by the antigen model establishment method, the identification result of the antigen is obtained.

As shown in Figure 6, another embodiment of the present invention also provides a device for establishing an identification antigen model, including: a building block, the building block is used to build a neural network, wherein the neural network includes a TCR feature extraction neural network, a pMHC feature extraction neural network Network and antigen identification neural network; also be used for constructing and constructing data set, wherein, data set comprises TCR CDR3, HLA I and antigen sequence; Training module, training module is used for inputting data set into neural network, neural network is trained, with Establish identification antigen model.

As shown in Figure 7, another embodiment of the present invention provides a device for identifying antigen models, including a memory and a processor: the memory is used to store a computer program; the processor is used to execute the computer program , implement the antigen identification method as above.

Another embodiment of the present invention also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program. When the computer program is read and run by a processor, the above method for establishing an antigen identification model or the method for identifying an antigen is realized.

Although the disclosure of the present invention is as above, the protection scope of the disclosure of the present invention is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will all fall within the protection scope of the present invention.

Claims (10)

1. An identification antigen model building method, comprising:

constructing a neural network, wherein the neural network comprises a TCR feature extraction neural network, a pMHC feature extraction neural network and an antigen identification neural network;

constructing a data set, wherein the data set comprises TCR CDR3, HLA I, and an antigen sequence;

and inputting the data set into the neural network, and training the neural network to establish an antigen identification model.

2. The method of claim 1, further comprising:

and describing TCR CDR3, HLA I and antigen sequences in the data set by adopting a Z descriptor, and normalizing the described matrixes to respectively determine a TCR CDR3 sequence matrix, an HLA I sequence matrix and an antigen sequence matrix.

3. The method of claim 2, wherein inputting the data set into the neural network, training the neural network comprises:

taking the TCR CDR3 sequence matrix as a training set, and training the TCR feature extraction neural network; inputting the TCR CDR3 sequence matrix into a trained TCR feature extraction neural network to obtain a TCR feature vector;

training the pMHC characteristic extraction neural network by taking the HLA I sequence matrix and the antigen sequence matrix as training sets; inputting the HLA I sequence matrix and the antigen sequence matrix into a trained pMHC feature extraction neural network to obtain a pMHC feature vector;

and taking the TCR characteristic vector and the pMHC characteristic vector as a training set, and training the antigen identification neural network.

4. The method of claim 3, wherein before the training of the antigen-identifying neural network, the method further comprises:

and balancing a training set consisting of the TCR characteristic vector and the pMHC characteristic vector by adopting a SMOTE algorithm.

5. The method of claim 3, wherein training the antigen-identifying neural network comprises:

and taking the binding state of the TCR characteristic vector and the pMHC characteristic vector as a classification label.

6. The method of claim 3, wherein the training set of TCR and pMHC signature vectors comprises artificial negative binding data.

7. The method of claim 1, wherein the TCR feature extraction neural network comprises an encoder module, a feature extraction module, and a decoder module;

the convolutional layer module adopted by the encoder module comprises four Cov1D layers;

the characteristic extraction module comprises a full connection layer for outputting TCR sequence characteristics;

the decoder module includes four layers of Conv1D layers.

8. The method for building an identification antigen model according to claim 1, wherein the pMHC feature extraction neural network comprises an HLA feature extraction module, an antigen feature extraction module, a feature extraction module, and a tag training module;

the HLA feature extraction module comprises four Cov1D layers, a Reshape layer and a full connection layer;

the antigen feature extraction module comprises four Cov1D layers, a Reshape layer and a full connection layer;

the characteristic extraction module comprises a full connection layer for outputting pMHC sequence characteristics;

the label training module comprises two full-connection layers.

9. The method for building an antigen-identifying model according to claim 1, wherein the antigen-identifying neural network comprises a TCR feature learning module, a pMHC feature learning module, and an output module;

the TCR feature learning module comprises two full connection layers;

the pMHC feature learning module comprises two full connection layers;

the output module adopts three full-connection layers for outputting antigen identification results.

10. A method for identifying an antigen, comprising:

obtaining TCR CDR3, HLA I and an antigen sequence to be detected, inputting the TCR CDR3, HLA I and the antigen sequence to be detected into an identification antigen model established by the identification antigen model establishing method of any one of claims 1 to 9, and obtaining an identification result of the antigen.

Images