A method, system, device and storage medium for predicting immunogenic peptide presentation
By training the model to extract the characteristics of immunogenic peptides and perform fusion prediction, the problem of predicting immunogenic peptide presentation in the prior art is solved, and efficient and low-cost peptide presentation prediction is achieved.
Patent Information
- Application Number
- CN202310872953.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-07-14
- Publication Date
- 2025-08-15
- Estimated Expiration
- 2043-07-14
AI Technical Summary
The prior art is difficult to predict the presentation of immunogenic peptides quickly and accurately, especially the process of presentation of peptides to T cells, and the experimental methods are costly and time-consuming, and there are limitations in deep learning and machine learning models.
Through the trained shear prediction model, pre-trained protein language model and binding prediction model, the shear characteristics, amino acid sequence characteristics and binding characteristics of the immunogenic peptide are extracted, and then fused and entered into the presentation prediction model to perform presentation prediction of the immunogenic peptide.
It achieves more accurate and faster prediction of immunogenic peptide presentation, improves prediction accuracy and generalization capabilities, and reduces cost and time requirements.
Smart Images
Figure CN116994643B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of biological information processing technology, and in particular to a prediction method, system, device and storage medium for immunogenic peptide presentation. Background Art
[0002] In vivo, polypeptide chains synthesized from proteins must be cleaved into smaller peptide fragments to exert their biological functions, which involves complex enzymatic hydrolysis and processing. Immunogenic peptides are peptides that can elicit an immune response. These peptides can bind to major histocompatibility complex (MHC) molecules and be presented on the cell surface, thereby activating T cells and triggering immune responses such as antibody production and cytotoxic T lymphocyte-mediated killing. MHC molecules are receptors located on the cell membrane that can bind and present endogenous or exogenous antigenic peptides. Only peptides that can be presented on the cell surface by MHC have the potential to be recognized by T cells. Therefore, predicting the presentation of MHC class II immunogenic peptides is a key issue in immunology research and vaccine design.
[0003] Experimental methods commonly use techniques such as mass spectrometry, ELISPOT, and lymphocyte proliferation assays to predict immunogenic peptides. These methods require expensive experimental equipment and consumables, are costly, and typically take weeks to months to analyze samples. Sample purity and quantity requirements are also high. Currently, deep learning and machine learning models have limitations. They can only predict peptide binding to MHC but cannot effectively predict peptide presentation. Immunogenic peptides also need to meet other specific requirements before they can be presented to T cells. For example, they must have a specific length and amino acid sequence, as well as an appropriate spatial conformation, in order to be recognized by T cell receptors. In addition, many cells regulate the presentation of immunogenic peptides through different mechanisms to avoid inadvertently triggering autoimmune reactions or excessive immune responses. Summary of the Invention
[0004] In view of this, an object of the embodiments of the present invention is to provide a method, system, device and storage medium for predicting immunogenic peptide presentation, which can more accurately and quickly predict the presentation of immunogenic peptides.
[0005] In one aspect, an embodiment of the present invention provides a method for predicting immunogenic peptide presentation, comprising the following steps:
[0006] The immunogenic peptide, upstream amino acid sequence, and downstream amino acid sequence are input into the trained splicing prediction model to extract the peptide splicing features;
[0007] Inputting the immunogenic peptide into a pre-trained protein language model to extract the peptide amino acid sequence features;
[0008] Inputting the immunogenic peptide and major histocompatibility complex allele sequence into a trained binding prediction model to extract the binding characteristics of the immunogenic peptide and major histocompatibility complex;
[0009] The peptide segment cleavage feature, the peptide segment sequence feature and the binding feature are fused to obtain a fusion feature, and the fusion feature is input into a presentation prediction model to obtain a presentation prediction result.
[0010] Optionally, sample data for model training is obtained by the following steps:
[0011] Obtaining immunogenic peptide data samples from a first public database and cleaning the immunogenic peptide data, wherein the immunogenic peptide data includes several immunogenic peptide amino acid sequence samples, corresponding major histocompatibility complex allele names, and corresponding protein names;
[0012] Obtaining amino acid full-length sequence samples of several major histocompatibility complex alleles through a second public database;
[0013] The full-length amino acid sequence samples of several proteins were obtained through a third public database.
[0014] Optionally, the cleaning process of the immunogenic peptide data is as follows:
[0015] Immunogenic peptides whose lengths were not within the range of human peptide lengths and repeated immunogenic peptides were removed.
[0016] Optionally, the training process of the shear prediction model is as follows:
[0017] The peptide segments that do not correspond to the full-length protein in the cleaned immunogenic peptide data are removed to obtain the first positive sample peptide segments;
[0018] Randomly intercepting an amino acid sequence of the length of a human peptide in the protein series, and if the intercepted amino acid sequence does not belong to the first positive sample peptide, the intercepted amino acid sequence is used as the first negative sample peptide;
[0019] intercepting a first type of amino acid sequence upstream and downstream of the first positive sample peptide segment, and intercepting a second type of amino acid sequence upstream and downstream of the first negative sample peptide segment;
[0020] The first positive sample peptide segment, the first negative sample peptide segment, the first type of amino acid sequence and the second type of amino acid sequence are divided into a first training set and a first test set, the first training set is used to train the shear prediction model, and the first test set is used to test the shear prediction model until the shear prediction model meets the first preset requirement.
[0021] Optionally, the training process of the combined prediction model is as follows:
[0022] Peptides lacking corresponding major histocompatibility complex allele information are removed from the cleaned immunogenic peptide data to obtain sample peptides; among them, the sample peptides that are determined to be able to bind to the major histocompatibility complex according to the experimental results are second positive sample peptides, and the remaining sample peptides are second negative sample peptides;
[0023] The second positive sample peptide segment and its major histocompatibility complex allele sample sequence and the second negative sample peptide segment and its major histocompatibility complex allele sample sequence are divided into a second training set and a second test set, the second training set is used to train the binding prediction model, and the second test set is used to test the binding prediction model until the binding prediction model meets the second preset requirement.
[0024] Optionally, the training process of the presentation prediction model is as follows:
[0025] The peptide segment that meets the shearing screening conditions, the binding screening conditions and can be presented is used as the third positive sample peptide segment; the third positive sample peptide segment also includes the name of the major histocompatibility complex allele and the corresponding protein name;
[0026] Randomly intercepting an amino acid sequence of the length of a human peptide from the protein sequence as a third negative sample; the third negative sample also includes a randomly matched major histocompatibility complex allele name, allele sequence, the full-length protein sequence of all peptides matched according to the protein name, and the upstream and downstream amino acids of the peptides;
[0027] The third positive sample peptide segment and the third negative sample are divided into a third training set and a third test set, the third training set is used to train the presentation prediction model, and the third test set is used to test the presentation prediction model until the presentation prediction model meets the third preset requirement.
[0028] In another aspect, an embodiment of the present invention provides a system for predicting immunogenic peptide presentation, comprising:
[0029] The first module is used to input the immunogenic peptide, upstream amino acid sequence and downstream amino acid sequence into the trained splicing prediction model to extract the peptide splicing features;
[0030] The second module is used to input the immunogenic peptide segment into the trained pre-trained protein language model to extract the peptide segment amino acid sequence features;
[0031] The third module is used to input the immunogenic peptide segment and the major histocompatibility complex allele sequence into the trained binding prediction model to extract the binding characteristics of the immunogenic peptide segment and the major histocompatibility complex;
[0032] The fourth module is used to fuse the peptide segment cleavage feature, the peptide segment sequence feature and the binding feature to obtain a fusion feature, and input the fusion feature into the presentation prediction model to obtain a presentation prediction result.
[0033] In another aspect, an embodiment of the present invention provides a device for predicting immunogenic peptide presentation, comprising:
[0034] at least one processor;
[0035] at least one memory for storing at least one program;
[0036] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.
[0037] On the other hand, an embodiment of the present invention provides a storage medium storing a program executable by a processor. When the program is executed by the processor, it is used to perform the above method.
[0038] On the other hand, an embodiment of the present invention provides a prediction system for immunogenic peptide presentation, comprising a computer device and a plurality of storage devices connected to the computer device; wherein,
[0039] Several storage devices are used to store sample data for model training;
[0040] The computer device comprises:
[0041] at least one processor;
[0042] at least one memory for storing at least one program;
[0043] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.
[0044] The implementation of the embodiments of the present invention includes the following beneficial effects: in this embodiment, the upstream amino acid sequence and the downstream amino acid sequence of the immunogenic peptide are input into a trained splicing prediction model to extract the peptide splicing features, the immunogenic peptide is input into a trained pre-trained protein language model to extract the peptide amino acid sequence features, the immunogenic peptide and the major histocompatibility complex allele sequence are input into a trained binding prediction model to extract the binding features of the immunogenic peptide and the major histocompatibility complex, and the peptide splicing features, the peptide amino acid sequence features and the binding features are fused to obtain a fusion feature, and the fusion feature is input into a presentation prediction model to obtain a presentation prediction result, thereby mining the intrinsic relationship information of the peptide from splicing to binding to presentation, so as to achieve more accurate and faster prediction of the presentation of the immunogenic peptide. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] Figure 1 1 is a schematic flow chart of the steps of a method for predicting immunogenic peptide presentation provided by an embodiment of the present invention;
[0046] Figure 2 This is another schematic flow chart of the steps of a method for predicting immunogenic peptide presentation provided by an embodiment of the present invention;
[0047] Figure 3 This is a schematic diagram of the structure of a Transformer pre-trained protein language model provided by an embodiment of the present invention;
[0048] Figure 4 1 is a schematic diagram of a training process of a shear prediction model provided by an embodiment of the present invention;
[0049] Figure 5 Schematic diagram of a training process for a combined prediction model provided by an embodiment of the present invention;
[0050] Figure 6 This is a prediction result diagram of a Class II amino acid provided in an embodiment of the present invention;
[0051] Figure 7 This is a structural block diagram of a prediction system for immunogenic peptide presentation provided by an embodiment of the present invention;
[0052] Figure 8 This is a structural block diagram of a device for predicting immunogenic peptide presentation provided by an embodiment of the present invention;
[0053] Figure 9 This is another structural block diagram of a prediction system for immunogenic peptide presentation provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0054] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The step numbers in the following embodiments are provided for ease of description only and do not limit the order of the steps. The order of execution of the steps in the embodiments can be adaptively adjusted based on the understanding of those skilled in the art.
[0055] The embodiment of the present invention involves multiple prediction models, each of which is used to predict different performance characteristics. Each prediction model needs to be trained first, and the training data uses sample data. When the trained prediction model meets the preset requirements, the trained prediction model is used to predict the performance characteristics based on the prediction data.
[0056] like Figure 1 As shown, the present invention provides a method for predicting immunogenic peptide presentation, comprising the following steps:
[0057] S100: Input the immunogenic peptide, upstream amino acid sequence and downstream amino acid sequence into the trained splicing prediction model to extract the peptide splicing features.
[0058] S200: Input the immunogenic peptide segment into a pre-trained protein language model to extract peptide segment amino acid sequence features.
[0059] S300: Input the immunogenic peptide segment and major histocompatibility complex allele sequence into a trained binding prediction model to extract the binding characteristics of the immunogenic peptide segment and major histocompatibility complex.
[0060] S400: Fusing the peptide segment cleavage feature, the peptide segment sequence feature, and the binding feature to obtain a fusion feature, and inputting the fusion feature into a presentation prediction model to obtain a presentation prediction result.
[0061] See Figure 2 First, the upstream amino acid sequence and the immunogenic peptide are jointly input into a trained splicing prediction model to extract the peptide splicing features, the immunogenic peptide is input into a trained pre-trained protein language model to extract the peptide amino acid sequence features, and the immunogenic peptide and MHC (major histocompatibility complex) allele sequence are input into a trained binding prediction model to extract the binding features of the immunogenic peptide and MHC; then, the extracted peptide splicing features, peptide amino acid sequence features, and binding features are fused to obtain a fusion feature; finally, the fusion feature is input into a presentation prediction model to obtain a presentation prediction result. It can be understood by those skilled in the art that the presentation prediction result includes whether the immunogenic peptide can be presented to T cells.
[0062] Specifically, the splicing prediction model, pre-trained protein language model, combined pre-trained model, and presentation prediction model are determined according to actual applications, and this embodiment does not impose specific restrictions. For example, in this embodiment, the splicing prediction model, pre-trained protein language model, combined prediction model, and presentation prediction model can all be determined based on the Transformer pre-trained protein language model, but the structural parameters of each prediction model may be different. Figure 3 , Figure 3 This is a structural diagram of the Transformer pre-trained protein language model. The Transformer pre-trained protein language model includes an encoder and a decoder. The encoder consists of several layers of multi-head attention mechanism, residual connection and normalization, feedforward neural network, etc. The decoder consists of several layers of multi-head attention mechanism, residual connection and normalization, feedforward neural network, linear layer, normalization layer, etc.
[0063] In a specific embodiment, the shear prediction model is determined based on the Transformer pre-trained protein language model, the hidden layer size is set to 512, there are 6 hidden layers, 8 attention heads, the middle layer size of the encoder is set to 512, and the maximum position embedding is 100; the linear layer input feature size is 512, and the output feature size is 2; the optimizer selects AdamW, the learning decay selects cosine annealing to adjust the learning rate CosineAnnealingLR, the initial learning rate is 0.0001, and the batch size is 128.
[0064] In a specific embodiment, the pre-trained model is determined based on the Transformer pre-trained protein language model, the hidden layer size is set to 512, there are 6 hidden layers in total, the attention heads are 8 in total, the middle layer size of the encoding layer is set to 512, and the maximum position embedding is 400; the linear layer input feature size is 512, and the output feature size is 2; the optimizer selects AdamW, the learning decay selects cosine annealing to adjust the learning rate CosineAnnealingLR, the initial learning rate is 0.0001, and the batch size is 256.
[0065] In a specific embodiment, the presentation model is constructed based on the Transformer pre-trained protein language model, the hidden layer size is set to 512, there are 6 hidden layers in total, the attention heads are 8 in total, the middle layer size of the Transformer encoding layer is set to 512, and the maximum position embedding is 50; the linear module includes three linear layers, the input feature size of the first linear layer is 1536, the input size of the second linear layer is 512, and the output size of the third linear layer is 2; the optimizer selects AdamW, the learning decay selects cosine annealing to adjust the learning rate CosineAnnealingLR, the initial learning rate is 0.0001, and the batch size is 256.
[0066] Optionally, sample data for model training is obtained by the following steps:
[0067] S001. Obtain immunogenic peptide data samples from a first public database and clean the immunogenic peptide data. The immunogenic peptide data includes several immunogenic peptide amino acid sequence samples, corresponding major histocompatibility complex allele names, and corresponding protein names.
[0068] The first public database is determined based on actual application and is not specifically limited in this embodiment. In this embodiment, the immunogenic peptide data sample is from the public database IEDB, which contains 1.45 million immunogenic peptide amino acid sequences identified through actual laboratory testing and comprehensive literature analysis, as well as the corresponding MHC allele names, corresponding protein names, etc.
[0069] Optionally, the cleaning process of the immunogenic peptide data is as follows:
[0070] S0011. Eliminate immunogenic peptides whose lengths are not within the range of human peptide lengths and repeated immunogenic peptides.
[0071] Specifically, data with immunogenic peptide lengths outside the range of human peptide lengths (8 to 26 amino acids for class II and 8 to 14 amino acids for class I) and duplicates were removed (only one duplicate was retained).
[0072] S002. Obtain amino acid full-length sequence samples of several major histocompatibility complex alleles through a second public database.
[0073] The second public database is determined according to actual application and is not specifically limited in this embodiment. In this embodiment, the MHC allele sequences are from the public database EMBL-EBI, which has collected more than 20,000 full-length amino acid sequences of MHC alleles.
[0074] S003. Obtain full-length amino acid sequence samples of several proteins through a third public database.
[0075] The third public database is determined based on actual application and is not specifically limited in this embodiment. It should be noted that proteins include human proteins and non-human proteins. In this embodiment, for example, protein data comes from the public database NCBI, which has collected the full-length amino acid sequences of approximately 28,000 proteins.
[0076] Optionally, the training process of the shear prediction model is as follows:
[0077] S011, removing peptides that do not correspond to the full-length protein from the cleaned immunogenic peptide data to obtain the first positive sample peptide;
[0078] S012. Randomly intercepting an amino acid sequence of the length of a human peptide in the protein series, and if the intercepted amino acid sequence does not belong to the first positive sample peptide, using the intercepted amino acid sequence as the first negative sample peptide;
[0079] S013, intercepting the first type of amino acid sequence upstream and downstream of the first positive sample peptide segment, and intercepting the second type of amino acid sequence upstream and downstream of the first negative sample peptide segment;
[0080] S014. Divide the first positive sample peptide segment, the first negative sample peptide segment, the first type of amino acid sequence, and the second type of amino acid sequence into a first training set and a first test set, use the first training set to train the shear prediction model, and use the first test set to test the shear prediction model until the shear prediction model meets the first preset requirement.
[0081] Peptides that could not be found in the cleaned immunogenic peptide data were removed as positive sample peptides. Amino acid sequences of the length of human peptides were randomly intercepted from the protein sequence. If the intercepted sequence was not in the positive sample set, it was included as a negative sample. It should be noted that because the task of this model is to predict whether a peptide fragment can be enzymatically hydrolyzed from the protein, the model input also needs to include the upstream and downstream sequences of the peptide fragment in the original protein. The upstream and downstream amino acid sequences are obtained by intercepting the 15 amino acid sequences upstream and downstream of the peptide from the protein in which the peptide is located.
[0082] See Figure 4 The input training set data of the shear prediction model includes: immunogenic peptides and upstream and downstream sequences. The protein language model is pre-trained through Transformer to extract sequence features; then the linear layer is used to output the binary prediction score, which is compared with the actual result, and the loss is calculated using the cross-entropy function. During training, the loss function gradient backpropagation updates all model parameters; the parameters of the large protein language model can also be frozen, and the loss function gradient backpropagation only updates the remaining parameters. Repeat the training until the AUC of the model on the validation set no longer continues to improve. Among them, AUC (Area under the Curve of ROC) is the area under the ROC curve, which is the criterion for judging the quality of the binary prediction model. The closer the value is to 1, the higher the model performance.
[0083] It should be noted that the specific ratio of the training set and the test set is determined according to actual application and is not specifically limited in this embodiment. For example, 80% of the sample data is used as the training set and 20% of the sample data is used as the test set.
[0084] Optionally, the training process of the combined prediction model is as follows:
[0085] S021. Eliminate peptide segments that lack corresponding major histocompatibility complex allele information from the cleaned immunogenic peptide data to obtain sample peptide segments; wherein the sample peptide segments that are determined to be able to bind to the major histocompatibility complex according to the experimental measurement results are second positive sample peptide segments, and the remaining sample peptide segments are second negative sample peptide segments;
[0086] S022. Dividing the second positive sample peptide segment and its major histocompatibility complex allele sample sequence and the second negative sample peptide segment and its major histocompatibility complex allele sample sequence into a second training set and a second test set, using the second training set to train the binding prediction model, and using the second test set to test the binding prediction model until the binding prediction model meets the second preset requirement.
[0087] The peptides without alleles in the cleaned immunogenic peptide data were removed; each peptide had a corresponding allele name, and then the corresponding allele sequence was matched according to the allele name.
[0088] See Figure 5 The input data for predictive model training includes immunogenic peptides and MHC molecule amino acid sequences. A Transformer-based pretrained protein language model extracts amino acid sequence features. A linear layer then outputs a binary prediction score, which is compared with the true result, and the loss is calculated using the cross-entropy function. During training, the gradient of the loss function is backpropagated to update all model parameters. Alternatively, the parameters of the large protein language model can be frozen, and the gradient of the loss function is backpropagated to update only the remaining parameters. Training is repeated until the model's AUC on the validation set no longer improves.
[0089] Optionally, the training process of the presentation prediction model is as follows:
[0090] S031. A peptide segment that meets the shearing screening conditions, the binding screening conditions, and can be presented is used as a third positive sample peptide segment; the third positive sample peptide segment also includes a major histocompatibility complex allele name and a corresponding protein name;
[0091] S032. Randomly intercepting an amino acid sequence of the length of a human peptide from the protein sequence as a third negative sample; the third negative sample also includes a randomly matched major histocompatibility complex allele name, allele sequence, the full-length protein sequence of all peptides matched according to the protein name, and the upstream and downstream amino acids of the peptides;
[0092] S033. Divide the third positive sample peptide segment and the third negative sample into a third training set and a third test set, use the third training set to train the presentation prediction model, and use the third test set to test the presentation prediction model until the presentation prediction model meets the third preset requirement.
[0093] After cleaning, the immunogenic peptide data must meet both the shearing and binding screening criteria, retaining only peptides that can be presented. These peptides serve as positive samples, each containing the corresponding MHC allele name and protein name. Amino acid sequences of the length of human peptides are randomly truncated from the protein sequence as negative samples, and the MHC allele names are randomly matched. All peptides are matched to the corresponding allele sequences based on the MHC allele name; all peptides are matched to the full-length protein sequence based on the corresponding protein name, and the upstream and downstream amino acids are truncated at the peptide location.
[0094] The input data for the presentation prediction model includes peptide splicing features extracted by the splicing prediction model, binding features extracted by the binding prediction model, and peptide sequence features extracted by a pre-trained protein language model. The peptide sequence features, peptide splicing features, and binding features are concatenated and input to the feature fusion module to generate fused features. A linear module then outputs a binary classification prediction score, which is compared with the actual result and the loss is calculated using the cross-entropy function. During training, the parameters of the pre-trained splicing model, the pre-trained binding model, and the pre-trained protein language model are frozen, and the loss function gradient is back-propagated to update only the remaining parameters. Training is repeated until the model's AUC on the validation set no longer improves.
[0095] In a specific embodiment, see Figure 6 , the AUC of class II was 0.977. Figure 6 The accuracy of each allele is also very high. Ten allele subtypes of Class II were randomly selected and shown in Table 1.
[0096] Table 1
[0097]
[0098]
[0099] Input the immunogenic peptide, allele subtype, upstream amino acid, and downstream amino acid, and each model will automatically match the allele sequence according to the allele subtype name. The label is the result identified by actual laboratory testing and comprehensive analysis of literature: P (Positive) indicates that the peptide is presented; N (Negative) means it will not be presented. The prediction score is the prediction result of the presentation model of the present invention, and the score (0-1) is: the closer to 1, the more likely the peptide is to be presented; the closer to 0, the less likely it is to be presented.
[0100] Table 2 below shows the peptides randomly sampled and tested, and the model prediction results are very consistent with the actual results.
[0101] Table 2
[0102]
[0103]
[0104] The implementation of the embodiment of the present invention includes the following beneficial effects: in this embodiment, by inputting the upstream amino acid sequence and the downstream amino acid sequence of the immunogenic peptide into a trained cleavage prediction model to extract the peptide cleavage characteristics, the stability of the peptide is better linked to the instability of the cleavage position, so that the naturally cleaved immunogenic peptide can be predicted with high accuracy; by inputting the immunogenic peptide into a trained pre-trained protein language model to extract the peptide amino acid sequence characteristics, by inputting the immunogenic peptide and the major histocompatibility complex allele sequence into a trained binding prediction model to extract the immunogenic peptide and major histocompatibility complex allele sequence. The binding characteristics of the complex improve the accuracy of binding of immunogenic peptides to MHC molecules, are compatible with various MHC allele subtypes, have good generalization ability, and fuse the peptide cleavage characteristics, peptide amino acid sequence characteristics and binding characteristics to obtain fusion characteristics. By inputting the fusion characteristics into the presentation prediction model to obtain the presentation prediction results, the model predicts whether the immunogenic peptide can be presented. It is not limited to the number of sample data, which improves the accuracy and generalization of the model, thereby mining the intrinsic relationship information of the peptide from cleavage to binding to presentation, so as to achieve more accurate and faster prediction of the presentation of immunogenic peptides.
[0105] See Figure 7 , an embodiment of the present invention provides a prediction system for immunogenic peptide presentation, comprising:
[0106] The first module is used to input the immunogenic peptide, upstream amino acid sequence and downstream amino acid sequence into the trained splicing prediction model to extract the peptide splicing features;
[0107] The second module is used to input the immunogenic peptide segment into the trained pre-trained protein language model to extract the peptide segment amino acid sequence features;
[0108] The third module is used to input the immunogenic peptide segment and the major histocompatibility complex allele sequence into the trained binding prediction model to extract the binding characteristics of the immunogenic peptide segment and the major histocompatibility complex;
[0109] The fourth module is used to fuse the peptide segment cleavage feature, the peptide segment sequence feature and the binding feature to obtain a fusion feature, and input the fusion feature into the presentation prediction model to obtain a presentation prediction result.
[0110] It can be seen that the contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.
[0111] See Figure 8 , an embodiment of the present invention provides a device for predicting immunogenic peptide presentation, comprising:
[0112] at least one processor;
[0113] at least one memory for storing at least one program;
[0114] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.
[0115] It can be seen that the contents of the above method embodiments are all applicable to the present device embodiments. The functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.
[0116] In addition, the embodiments of the present application further disclose a computer program product or computer program, which is stored in a computer-readable storage medium. The processor of a computer device can read the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device performs the above-mentioned method. Similarly, the contents of the above-mentioned method embodiment are all applicable to the present storage medium embodiment, and the functions specifically implemented by the present storage medium embodiment are the same as those of the above-mentioned method embodiment, and the beneficial effects achieved are also the same as those achieved by the above-mentioned method embodiment.
[0117] See Figure 9 , an embodiment of the present invention provides a prediction system for immunogenic peptide presentation, comprising a computer device and a plurality of storage devices connected to the computer device; wherein,
[0118] Several storage devices are used to store sample data for model training;
[0119] The computer device comprises:
[0120] at least one processor;
[0121] at least one memory for storing at least one program;
[0122] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.
[0123] Specifically, the public database is mainly implemented through a storage device, and it may specifically include at least one memory. The number of public databases is determined according to actual applications and is not specifically limited in this embodiment. As for the computer device, it can be different types of electronic devices, including but not limited to desktop computers, laptops and other terminals.
[0124] It can be seen that the contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.
[0125] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.
Claims
1. A method for predicting immunogenic peptide presentation, characterized in that: include: The immunogenic peptide, upstream amino acid sequence, and downstream amino acid sequence are input into the trained splicing prediction model to extract the peptide splicing features; Inputting the immunogenic peptide into a pre-trained protein language model to extract the peptide amino acid sequence features; Inputting the immunogenic peptide and major histocompatibility complex allele sequence into a trained binding prediction model to extract the binding characteristics of the immunogenic peptide and major histocompatibility complex; fusing the peptide segment cleavage feature, the peptide segment amino acid sequence feature, and the binding feature to obtain a fusion feature, and inputting the fusion feature into a presentation prediction model to obtain a presentation prediction result; The training process of the shear prediction model is as follows: The peptide segments that do not correspond to the full-length protein in the cleaned immunogenic peptide data are removed to obtain the first positive sample peptide segments; Randomly intercepting an amino acid sequence of the length of a human peptide in the protein series, and if the intercepted amino acid sequence does not belong to the first positive sample peptide, the intercepted amino acid sequence is used as the first negative sample peptide; intercepting a first type of amino acid sequence upstream and downstream of the first positive sample peptide segment, and intercepting a second type of amino acid sequence upstream and downstream of the first negative sample peptide segment; Dividing the first positive sample peptide segment, the first negative sample peptide segment, the first type of amino acid sequence, and the second type of amino acid sequence into a first training set and a first test set, training the splicing prediction model using the first training set, and testing the splicing prediction model using the first test set until the splicing prediction model meets a first preset requirement; The training process of the combined prediction model is as follows: Peptides lacking corresponding major histocompatibility complex allele information are removed from the cleaned immunogenic peptide data to obtain sample peptides; among them, the sample peptides that are determined to be able to bind to the major histocompatibility complex according to the experimental results are second positive sample peptides, and the remaining sample peptides are second negative sample peptides; dividing the second positive sample peptide segment and its major histocompatibility complex allele sample sequence and the second negative sample peptide segment and its major histocompatibility complex allele sample sequence into a second training set and a second test set, using the second training set to train the binding prediction model, and using the second test set to test the binding prediction model until the binding prediction model meets a second preset requirement; The training process of the presentation prediction model is as follows: The peptide segment that meets the shearing screening conditions, the binding screening conditions and can be presented is used as the third positive sample peptide segment; the third positive sample peptide segment also includes the name of the major histocompatibility complex allele and the corresponding protein name; Randomly intercepting an amino acid sequence of the length of a human peptide from the protein sequence as a third negative sample; the third negative sample also includes a randomly matched major histocompatibility complex allele name, allele sequence, the full-length protein sequence of all peptides matched according to the protein name, and the upstream and downstream amino acids of the peptides; The third positive sample peptide segment and the third negative sample are divided into a third training set and a third test set, the third training set is used to train the presentation prediction model, and the third test set is used to test the presentation prediction model until the presentation prediction model meets the third preset requirement.
2. The method according to claim 1, characterized in that The sample data for model training is obtained through the following steps: Obtaining immunogenic peptide data samples from a first public database and cleaning the immunogenic peptide data, wherein the immunogenic peptide data includes several immunogenic peptide amino acid sequence samples, corresponding major histocompatibility complex allele names, and corresponding protein names; Obtaining amino acid full-length sequence samples of several major histocompatibility complex alleles through a second public database; The full-length amino acid sequence samples of several proteins were obtained through a third public database.
3. The method according to claim 2, characterized in that The cleaning process of the immunogenic peptide data is as follows: Immunogenic peptides whose lengths were not within the range of human peptide lengths and repeated immunogenic peptides were removed.
4. A prediction system for immunogenic peptide presentation, characterized in that include: The first module is used to input the immunogenic peptide, upstream amino acid sequence and downstream amino acid sequence into the trained splicing prediction model to extract the peptide splicing features; The second module is used to input the immunogenic peptide segment into the trained pre-trained protein language model to extract the peptide segment amino acid sequence features; The third module is used to input the immunogenic peptide segment and the major histocompatibility complex allele sequence into the trained binding prediction model to extract the binding characteristics of the immunogenic peptide segment and the major histocompatibility complex; The fourth module is used to fuse the peptide segment cleavage feature, the peptide segment amino acid sequence feature and the binding feature to obtain a fusion feature, and input the fusion feature into a presentation prediction model to obtain a presentation prediction result; The training process of the shear prediction model is as follows: The peptide segments that do not correspond to the full-length protein in the cleaned immunogenic peptide data are removed to obtain the first positive sample peptide segments; Randomly intercepting an amino acid sequence of the length of a human peptide in the protein series, and if the intercepted amino acid sequence does not belong to the first positive sample peptide, the intercepted amino acid sequence is used as the first negative sample peptide; intercepting a first type of amino acid sequence upstream and downstream of the first positive sample peptide segment, and intercepting a second type of amino acid sequence upstream and downstream of the first negative sample peptide segment; Dividing the first positive sample peptide segment, the first negative sample peptide segment, the first type of amino acid sequence, and the second type of amino acid sequence into a first training set and a first test set, training the splicing prediction model using the first training set, and testing the splicing prediction model using the first test set until the splicing prediction model meets a first preset requirement; The training process of the combined prediction model is as follows: Peptides lacking corresponding major histocompatibility complex allele information are removed from the cleaned immunogenic peptide data to obtain sample peptides; among them, the sample peptides that are determined to be able to bind to the major histocompatibility complex according to the experimental results are second positive sample peptides, and the remaining sample peptides are second negative sample peptides; dividing the second positive sample peptide segment and its major histocompatibility complex allele sample sequence and the second negative sample peptide segment and its major histocompatibility complex allele sample sequence into a second training set and a second test set, using the second training set to train the binding prediction model, and using the second test set to test the binding prediction model until the binding prediction model meets a second preset requirement; The training process of the presentation prediction model is as follows: The peptide segment that meets the shearing screening conditions, the binding screening conditions and can be presented is used as the third positive sample peptide segment; the third positive sample peptide segment also includes the name of the major histocompatibility complex allele and the corresponding protein name; Randomly intercepting an amino acid sequence of the length of a human peptide from the protein sequence as a third negative sample; the third negative sample also includes a randomly matched major histocompatibility complex allele name, allele sequence, the full-length protein sequence of all peptides matched according to the protein name, and the upstream and downstream amino acids of the peptides; The third positive sample peptide segment and the third negative sample are divided into a third training set and a third test set, the third training set is used to train the presentation prediction model, and the third test set is used to test the presentation prediction model until the presentation prediction model meets the third preset requirement.
5. A device for predicting immunogenic peptide presentation, characterized in that include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 3.
6. A storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to perform the method according to any one of claims 1 to 3 when executed by the processor.
7. A prediction system for immunogenic peptide presentation, characterized in that It includes a computer device and several storage devices connected to the computer device; wherein, Several storage devices are used to store sample data for model training; The computer device comprises: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 3.