Antibody-oriented double-branch structure feature prediction method

Abstract

The invention discloses an antibody-oriented double-branch structure feature prediction method, and relates to the technical field of protein expression learning and protein structure prediction, and the method comprises the following steps: 1, taking a given antibody amino acid sequence as an input, extracting three types of features including an MSA feature, a Pair feature and a template feature of an antibody, carrying out collaborative fusion on the three types of features; 2, constructing a double-branch structure feature prediction module, and optimizing an antibody structure; step 3, model training; and 4, testing the model to obtain an antibody structure. By adopting the antibody-oriented double-branch structure feature prediction method disclosed by the invention, efficient and accurate antibody structure feature prediction is realized, and powerful support is provided for related research and application of antibodies.

Description

Technical Field

[0001] The present invention relates to the technical field of protein representation learning and protein structure prediction, and in particular to a method for predicting double-branch structure features of antibodies. Background Art

[0002] Antibody structure prediction is to predict the three-dimensional structure of antibodies by computational methods when the amino acid sequence of antibodies is known. Antibodies are special proteins produced by the immune response of organisms to invading pathogens. Their main function is to recognize antigens and bind specifically to them, thereby effectively removing foreign substances such as bacteria and fungi that invade the human body, neutralizing toxins released by these pathogens, and maintaining the balance of the immune system, which plays a significant role in maintaining human health. As a key component of the immune system, the function of antibodies is closely related to their structure. Therefore, accurate modeling and prediction of antibody structure are of great significance for understanding the mechanism of antibody-antigen interaction, as well as drug design, vaccine development, and disease treatment.

[0003] The variable region of an antibody is the key structural basis for its ability to bind to various antigens. The variable region consists of a framework region (FR) and a complementary determining region (CDR). The framework region provides basic structural support, while the high variability of the complementary determining region, especially the heavy chain complementary determining region 3 (H3), enables antibodies to adapt to the structure of different antigens. The complementary determining region can form a variety of shapes of ring structures that can complement the epitope of the antigen in space. This extreme diversity makes it difficult for general methods to effectively predict its structure, and its importance requires that the predicted results need to be highly accurate.

[0004] There are many methods for studying antibody structure prediction tasks:

[0005] Traditional methods: Traditional methods can be basically divided into homology modeling and ab initio prediction, which mainly rely on experimental methods such as X-ray crystal diffraction and nuclear magnetic resonance imaging. These methods are not only time-consuming but also costly.

[0006] Artificial intelligence-based methods: Deep learning builds a neural network model and uses a large amount of known antibody structure and sequence data for training. It can learn complex patterns and features from a large amount of data, thereby achieving high-precision prediction of antibody structure. However, the prediction of antibody structure requires not only considering the information of the amino acid sequence, but also its spatial conformation, that is, the three-dimensional structure. Traditional sequence-based deep learning methods have difficulty directly processing such structured data, which limits their application in antibody structure prediction. Summary of the invention

[0007] The purpose of the present invention is to provide a method for predicting double-branch structural features for antibodies, by extracting multidimensional features from the antibodies and performing aggregation updates, using the paired features therein to capture the complex relationships within the antibodies, and predicting the structural features of the antibodies to assist the task, so as to achieve the purpose of more accurate prediction of the three-dimensional structure of the antibodies.

[0008] To achieve the above object, the present invention provides a method for predicting double-branched structural features of antibodies, comprising the following steps:

[0009] Step 1: Taking a given antibody amino acid sequence as input, extract the MSA feature, Pair feature and template feature of the antibody, and synergistically fuse the three features;

[0010] Step 2: Construct a double-branch structure feature prediction module to optimize the antibody structure;

[0011] Step 3: Model training;

[0012] Step 4: Model testing to obtain the antibody structure.

[0013] Preferably, in step 2, the target antibody structural features are represented as a set of six relative distance and direction matrices in residue pairs L×L, including three inter-residue distance matrices and three inter-residue direction matrices.

[0014] Preferably, in step 2, a structural modal parsing network based on residue cross attention is constructed for the three inter-residue distance matrices and the two inter-residue direction matrices θ and φ for prediction, comprising the following steps:

[0015] S11, add the Pai r feature and its transposed matrix to obtain the average, and obtain a symmetric matrix as input;

[0016] S12, through a module composed of three stacked residual blocks, the processed features are added to the original input using residual connections and the output is activated again to perform deep feature extraction and conversion;

[0017] S13, projecting the dimension to the dimension corresponding to the number of intervals in the set range through a convolution layer, and the output features successively pass through two layers of cross attention to mine the correlation information between the features of the residue pairs;

[0018] S14, the features are processed again through a sequence of operations consisting of a convolution layer with a convolution kernel of 3*3, batch normalization, and ReLU activation;

[0019] S15. Adjust the dimension of the feature through a linear transformation layer and a transpose operation, and output the distance and direction matrices respectively.

[0020] Preferably, in S13,CA , d CB , d NO , θ is projected to 37, and φ is projected to 18.

[0021] Preferably, for the inter-residue direction matrix, using Pai r features as input, a structural symmetry state deduction network guided by dilated convolution is constructed for prediction, comprising the following steps:

[0022] S21, Pai r features and their transposed matrices are added and averaged to obtain a symmetric matrix as input;

[0023] S22, through multiple convolution modules, including convolution layer, instance normalization layer, ELU activation function layer, the features of multi-layer convolution output are added to the original input and activated by ELU to realize residual connection;

[0024] S23. After a convolution layer and a linear transformation layer, the dimension is adjusted. After the transpose operation, the output is a dihedral angle ω direction matrix with a shape of 1,nres,nres,37, where nres is the number of amino acid residues in the antibody sequence.

[0025] Preferably, the convolution layers in S22 use convolution kernels with increasing sizes and dilation rates in sequence to perform dilation convolution to expand the receptive field.

[0026] Preferably, in step three, the RoseTTAFol d2 network is used to extract features of the antibody sequence in the model training.

[0027] Preferably, in step three, the activation function used in the model training is the exponential linear unit ELU and the rectified linear unit ReLU, the dropout is set to 0.1 to avoid overfitting, the optimizer is Adam, the training is 10 cycles, the learning rate is set to 0.001, and the learning rate scheduler MultiStepLR is used to reduce the learning rate by 0.5 times in the 2nd, 3rd, 4th, 6th, and 8th cycles.

[0028] Therefore, the present invention adopts the above-mentioned dual-branch structural feature prediction method for antibodies to achieve efficient and accurate antibody structural feature prediction, providing strong support for antibody-related research and applications. By constructing a dual-branch structural feature prediction module, a structural modal parsing network based on residue cross attention is used to predict and output d CA Distance matrix, d CB Distance matrix, d NOFive structural features, including distance matrix, dihedral angle θ direction matrix, torsion angle φ direction matrix, etc.; using the structural symmetry state inference network guided by void convolution, the dihedral angle ω direction matrix is ​​predicted and output separately. The six antibody structural features can provide the necessary constraints for antibody structure prediction training, making the predicted structure more in line with biological reality, thereby effectively improving the accuracy and reliability of the task, and enhancing the adaptability and effectiveness of the overall prediction effect in biological application scenarios.

[0029] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A schematic diagram of an embodiment of a method for predicting double-branched structural features of antibodies according to the present invention;

[0031] Figure 2 This is a front view of an embodiment of a method for predicting double-branched structural features of antibodies according to the present invention. DETAILED DESCRIPTION

[0032] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0033] Unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0034] Example

[0035] See also Figure 1-2The present invention provides a method for predicting double-branch structural features for antibodies. Based on the input antibody amino acid sequence, the homologous sequence is searched, and the feature extraction is performed to obtain the multiple sequence alignment (MSA) feature, the paired (Pai r) feature and the template feature of the sequence. The three types of features are interactively updated to promote the full fusion of information between different features, so that the features can more accurately reflect the structure and function relationship of the antibody. A double-branch structural feature prediction module is constructed, and the Pa ir feature is processed separately. Different prediction networks in two branches are used to predict a total of six relative distances and direction matrices between residue pairs in the sequence, which can provide assistance for the training process of antibody structure prediction to optimize the final predicted structure.

[0036] Step 1: For a given antibody amino acid sequence as input, use a specific tool to extract the msa file containing its multiple sequence alignment content, and combine the template to extract the antibody's MSA features, Pa ir features, and template features. The three types of features are synergistically integrated to learn feature information and potential associations from each other, and the antibody prediction structure is preliminarily obtained.

[0037] Step 2: A dual-branch structural feature prediction module, one branch is a structural modal analysis network based on residue cross attention, and the other branch is a structural symmetry state deduction network guided by dilated convolution. The three types of features extracted from antibodies are multiple sequence alignment MSA features, paired Pa ir features, and template features, which respectively carry key information of different dimensions of antibodies, such as one-dimensional gene sequence, two-dimensional folding rotation, and three-dimensional spatial conformation. The three are fused and updated to learn from each other's feature information and potential associations.

[0038] Considering the great differences and instability of variable region structures in different antibodies, the prediction of structural features can play a certain auxiliary role in antibody structure prediction. For a target antibody sequence with L residues, the target antibody structural features can be represented as a set of six relative distance and direction matrices in residue pairs (L×L), including three inter-residue distance matrices: C α Between atoms (d CA :C α -C α ), C β Between atoms (d CB :C β -C β ), between NO atoms (d NO : NO); and three inter-residue orientation matrices: two dihedral angles (ω: C α -C β -C β -C α ,θ:NC α -C β -Cβ ), a torsion angle (φ: C α -C β -C β ). Each specific value in the six matrices is classified and mapped to different intervals: For the three distance matrices, the distance between residues is divided into 37 intervals, and the width of the first 36 intervals is represent The length range of the 37th interval represents the distance between residues exceeding For two dihedral angle matrices, the inter-residue dihedral angles were divided into 36 intervals with a width of 10°, representing the angle range of 0-360°; for one torsion angle matrix, the inter-residue torsion angles were divided into 18 intervals with a width of 10°, representing the angle range of 0-180°.

[0039] A dual-branch structural feature prediction module is constructed to predict the relative distance and direction matrix between residues. Pair features are used as module input, and the relative distance and direction matrix between residues are used as module output. A structural modal analysis network based on residue cross attention and a structural symmetry state deduction network guided by dilated convolution are constructed to perform dual-branch prediction on the relative distance and direction matrix between residues. The following is the specific network architecture of the two branches:

[0040] For the three inter-residue distance matrices, dihedral angle θ and torsion angle φ direction matrices, a structural modal parsing network based on residue cross-attention is constructed with Pai r features as input for prediction. CA d CB The Pai r feature and its transposed matrix are added and averaged to obtain a symmetric matrix as input. The features are sequentially passed through a module consisting of three residual blocks. Each residual block is first filled with appropriate padding to ensure the consistency of the size before and after convolution. Then, it includes operations such as convolution, instance normalization, ELU activation, and Dropout. The residual connection is used to add the processed features to the original input and then activate the output, so as to perform deep feature extraction and conversion. Then, a convolution layer is used to project the dimension to the dimension corresponding to the number of intervals in the set range (where d CA , d CB , d NO, θ is projected to 37, φ is projected to 18). Then the output features are successively passed through two layers of Cross-Attention (Cross-Attention), and the association information between the features of residue pairs is further mined through its unique query, key, value convolution, dimension transformation, attention calculation and other operations, so that each output residue pair can fully aggregate the information of all other residue pairs. Then, the features are processed again through an operation sequence consisting of a convolution layer with a convolution kernel of 3*3, batch normalization and ReLU activation. Finally, the features are dimensionally adjusted through a linear transformation layer and a transpose operation, and the distance and direction matrices with shapes of [1,nres,nres,37] are output respectively (where φ is [1,nres,nres,18]), where nres is the number of amino acid residues in the antibody sequence.

[0041] For the more complex symmetric dihedral angle ω direction matrix, the Pai r feature is used as input to construct a structural symmetry state deduction network guided by dilated convolution for prediction. Due to the symmetric nature of ω, the Pai r feature and its transposed matrix are added and averaged to obtain a symmetric matrix as input. The feature passes through multiple convolution modules, each of which includes a convolution layer, an instance normalization layer, and an ELU activation function layer. The features output by multiple convolution layers are added to the original input and activated by ELU to achieve residual connection. The convolution layers use convolution kernels of increasing sizes (3*3, 5*5, etc.) and dilation rates (1, 2, 4, 8, etc.) in order to perform dilated convolution to expand the receptive field. In this way, the symmetrical Pai r features in the antibody structure can better capture the feature relationship at different distance scales, such as taking into account the symmetric relationship between close and long-distance amino acid residues at the same time, so as to extract symmetric features more comprehensively. Finally, the features are passed through a convolutional layer and a linear transformation layer to adjust the dimension. After the transpose operation, the output is a dihedral angle ω direction matrix with a shape of [1,nres,nres,37], where nres is the number of amino acid residues in the antibody sequence.

[0042] After predicting the six relative distance and direction matrices, the cross entropy loss is calculated with the true structure to optimize the network.

[0043] Step 3: Model training phase

[0044] 3.1 Training Dataset

[0045] To construct the dataset used in the training process, we selected all paired structural antibodies collected before July 2021 from the structural antibody dataset website SAbDab, excluding antibody data with missing FV, incomplete heavy and light chains in the FV region, and missing heavy and light chains in the FV region, and screened 2,640 antibody samples. The heavy and light chains of these antibodies were separated, and the Chothia sequences of these single chains were taken. The msa files were obtained through the HHb l its tool as the model training dataset.

[0046] 3.2 Training Process

[0047] Perform training on the training dataset according to the steps in Section 2. The specific details are as follows:

[0048] The RoseTTAFol d2 network was used to extract features from the antibody sequence, and the MSA features were 256-dimensional, the Pair features were 128-dimensional, and the template features were 64-dimensional. The number of cross-attention heads in the residue-based cross-attention structural modal parsing network was set to 8. The activation function used exponential linear unit (ELU) and rectified linear unit (ReLU), the dropout was set to 0.1 to avoid overfitting, the optimizer used Adam, and the training was 10 cycles. The learning rate was set to 0.001, and the learning rate scheduler Mult iStepLR was used to reduce the learning rate by 0.5 times in the 2nd, 3rd, 4th, 6th, and 8th cycles.

[0049] 3.3 Training Loss Function

[0050] We use a combination of multiple loss functions to construct the overall loss function of the training process. The specific contents are as follows:

[0051] L=λ1L CA +λ2L CB +λ3L NO +λ4L ω +λ5L θ +λ6L φ ;

[0052] Among them, L CA ,L CB ,L NO ,L ω ,L θ ,L φ They are d CA ,d CB ,d NO ,ω,θ,φ are the losses, λ1~λ6 are the weight parameters corresponding to each loss, so as to adjust each loss to a suitable order of magnitude. All losses are calculated in the form of cross entropy loss.

[0053] Finally, the training is completed and the model is obtained.

[0054] Step 4: Model testing phase

[0055] In the testing phase, it is only necessary to use the model obtained by the above training steps, input the antibody sequence (fasta file) of the test data set into the model, and obtain the antibody structural features and final antibody structure predicted by the model.

[0056] Therefore, the present invention adopts the above-mentioned dual-branch structural feature prediction method for antibodies to achieve efficient and accurate antibody structural feature prediction, providing strong support for antibody-related research and applications. By constructing a dual-branch structural feature prediction module, a structural modal parsing network based on residue cross attention is used to predict and output d CA Distance matrix, d CB Distance matrix, d NO Five structural features, including distance matrix, dihedral angle θ direction matrix, torsion angle φ direction matrix, etc.; using the structural symmetry state inference network guided by void convolution, the dihedral angle ω direction matrix is ​​predicted and output separately. The six antibody structural features can provide the necessary constraints for antibody structure prediction training, making the predicted structure more in line with biological reality, thereby effectively improving the accuracy and reliability of the task, and enhancing the adaptability and effectiveness of the overall prediction effect in biological application scenarios.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for predicting double-branched structural features of antibodies, characterized in that: The following steps are involved: Step 1: Taking a given antibody amino acid sequence as input, extract the MSA feature, Pair feature, and template feature of the antibody by combining the template, and then synergistically fuse the three features; Step 2: Construct a double-branch structure feature prediction module to optimize the antibody structure; Step 3: Model training; Step 4: Model testing to obtain the antibody structure.

2. A method for predicting double-branched structural features of antibodies according to claim 1, characterized in that: In step 2, the target antibody structural features are represented as a set of six relative distance and direction matrices in residue pairs L×L, including three inter-residue distance matrices and three inter-residue direction matrices.

3. A method for predicting double-branched structural features of antibodies according to claim 2, characterized in that: In step 2, a structural modal parsing network based on residue cross attention is constructed for the three inter-residue distance matrices and the two inter-residue direction matrices θ and φ for prediction, including the following steps: S11, add the Pair feature and its transposed matrix to get the average, and obtain a symmetric matrix as input; S12, through a module composed of three stacked residual blocks, the processed features are added to the original input using residual connections and the output is activated again to perform deep feature extraction and conversion; S13, projecting the dimension to the dimension corresponding to the number of intervals in the set range through a convolution layer, and the output features successively pass through two layers of cross attention to mine the correlation information between the features of the residue pairs; S14, the features are processed again through a sequence of operations consisting of a convolution layer with a convolution kernel of 3*3, batch normalization, and ReLU activation; S15. Adjust the dimension of the feature through a linear transformation layer and a transpose operation, and output the distance and direction matrices respectively.

4. A method for predicting double-branched structural features of antibodies according to claim 3, characterized in that: S13 medium CA , d CB , d NO , θ is projected to 37, and φ is projected to 18.

5. The method for predicting double-branched structural features of antibodies according to claim 2, characterized in that: For the inter-residue direction matrix, using the pair feature as input, a structural symmetry state deduction network guided by dilated convolution is constructed for prediction, including the following steps: S21, add the pair features and their transposed matrix to get the average, and get the symmetric matrix as input; S22, through multiple convolution modules, including convolution layer, instance normalization layer, ELU activation function layer, the features of multi-layer convolution output are added to the original input and activated by ELU to realize residual connection; S23. After a convolutional layer and a linear transformation layer, the dimension is adjusted. After the transpose operation, the output is a dihedral angle ω direction matrix with a shape of 1,nres,nres,37, where nres is the number of amino acid residues in the antibody sequence.

6. A method for predicting double-branched structural features of antibodies according to claim 5, characterized in that: The convolutional layers in S22 use convolution kernels with increasing sizes and dilation rates in order to perform dilation convolution to expand the receptive field.

7. The method for predicting double-branched structural features of antibodies according to claim 1, characterized in that: In step three, the RoseTTAFold2 network is used to extract features of antibody sequences during model training.

8. A method for predicting double-branched structural features of antibodies according to claim 7, characterized in that: In step 3, the activation functions used in model training are exponential linear unit ELU and rectified linear unit ReLU, the dropout is set to 0.1 to avoid overfitting, the optimizer is Adam, the training is 10 cycles, the learning rate is set to 0.001, and the learning rate scheduler MultiStepLR is used to reduce the learning rate by 0.5 times in the 2nd, 3rd, 4th, 6th, and 8th cycles.

Images