A model for predicting high-affinity mutation sites of antibodies and a method for constructing the same and applications thereof

Abstract

The present application relates to a model for predicting high-affinity mutation sites of antibodies and a construction method and application thereof. The construction method comprises: taking the amino acid sequence of a mutant antibody with improved affinity as a training set, the training set comprising the amino acid sequence of the mutant antibody, the mutation site and the amino acid information after mutation, taking the amino acid after mutation as the correct prediction target, training the ESM model using the training set, the training comprising masking the amino acid sequence of the mutation site of the mutant antibody using a masking language model, then predicting the amino acid sequence of the masking site, outputting the probability of the correct prediction target, adjusting the model according to the output result, and obtaining the model for predicting high-affinity mutation sites of antibodies. The present application optimizes the construction of the model for predicting high-affinity mutation sites of antibodies based on the ESM model training, and can be effectively applied to antibody affinity maturation technology to predict effective mutation site combinations.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of biotechnology, and relates to a model for predicting high-affinity mutation sites of antibodies and a construction method and application thereof. BACKGROUND

[0002] Antibodies are an important protein in the immune system, which is used to recognize and neutralize foreign substances such as pathogens, viruses and bacteria. Antibody molecules are usually composed of four polypeptide chains, including two heavy chains and two light chains. These chains form a Y-shaped structure in the molecular structure, allowing antibodies to bind to specific antigens. The role of antibodies in the immune response is to bind to specific antigens through their variable regions, thereby neutralizing or marking these antigens for recognition and elimination by other immune cells. The diversity of antibodies is generated through gene recombination and mutation, which enables the immune system to recognize almost infinite types of antigens. Antibodies have a wide range of applications in medicine and scientific research, including disease diagnosis, vaccine development and treatment, etc. For example, monoclonal antibodies are widely used in cancer treatment, which can specifically target antigens on the surface of cancer cells, thereby inhibiting the growth and spread of cancer cells.

[0003] The affinity of an antibody refers to the binding force between the antibody and the epitope or antigenic determinant of the antigen. In the development and application of antibodies, affinity is an important indicator for measuring their biological activity and clinical value. High-affinity antibodies can significantly reduce the dosage, reduce side effects and save costs. Affinity maturation refers to the process in which antibodies gradually enhance their binding strength to antigens during the immune response. The affinity maturation process of antibodies is achieved through somatic hypermutation, mainly focusing on the CDR region of antibodies. The core idea is to screen antibody variants with higher affinity through random or directed mutation, mainly including single-point saturation mutation at each amino acid site in the CDR region, constructing a single-point saturation mutation plasmid library of the parent antibody, obtaining single mutation points that can improve affinity, and further combining single mutation points to obtain a series of high-affinity mutants. For example, CN115925947A discloses an affinity maturation method and affinity maturation of anti-human PD-L1 single-domain antibody. By combining single-point saturation mutation covering the CDR region with a mammalian system high-throughput expression screening system, high-affinity anti-human PD-L1 single-domain antibody is obtained.

[0004] However, how to combine single-point mutations to obtain effective multi-point mutations is still one of the problems in the field, and blind combination may even lead to a decrease in affinity. Therefore, developing an effective method to predict the effect of combined mutations on antibody affinity is of great significance to the field of antibody development. SUMMARY

[0005] In view of the deficiencies of the prior art and actual needs, the present application provides a model for predicting the affinity of mutant antibodies, a construction method and application thereof, develops a model for predicting high-affinity mutant sites of antibodies, so as to realize rapid and accurate prediction of high-affinity mutant site combinations of antibodies, and promote the development of affinity maturation technology.

[0006] To achieve the above object, the present application adopts the following technical solutions:

[0007] In a first aspect, the present application provides a construction method of a model for predicting high-affinity mutant sites of antibodies, which comprises the following steps:

[0008] The amino acid sequence of the mutant antibody with improved affinity is taken as a training set, the training set comprises the amino acid sequence of the mutant antibody, the mutant site and the amino acid information after mutation, and the amino acid after mutation is taken as a correct prediction target; the training set is used to train an ESM model, the training comprises masking the amino acid sequence of the mutant site of the mutant antibody by using a masking language model, then the masked amino acid sequence is predicted, the probability of the correct prediction target is output, the model is adjusted according to the output result, and a model for predicting high-affinity mutant sites of antibodies is obtained.

[0009] In the present application, the model for predicting high-affinity mutant sites of antibodies is constructed based on the ESM model training and optimization, the obtained model can effectively predict and evaluate the naturalness of different mutant sequences, and then infer the influence of mutation on the affinity of antibodies, so that the model can be effectively applied to the antibody affinity maturation technology to predict effective mutant site combinations.

[0010] Preferably, when the number of masked mutant sites is more than one, the average value of the sum of the probabilities of multiple correct prediction targets is taken as the output result.

[0011] Preferably, the adjustment is performed until the output result is not lower than a preset threshold value, and the threshold value is 0.5.

[0012] In a second aspect, the present application provides a model for predicting high-affinity mutant sites of antibodies, which is constructed by the construction method of the model for predicting high-affinity mutant sites of antibodies according to the first aspect.

[0013] In a third aspect, the present application provides a method for predicting high-affinity mutant site combinations of antibodies, which comprises the following steps:

[0014] The amino acid sequence of the mutant antibody is input into the model for predicting high-affinity mutant sites of antibodies according to the second aspect, a masking language model is used to mask all single mutant sites, then the masked amino acid sequence is predicted, the probability of the correct prediction target is output, the prediction results of multiple mutant sites are summed and averaged to obtain a score of the mutant site combination, and a mutant site combination with a high score is selected as a candidate high-affinity mutant site combination of antibodies.

[0015] In the present application, a method for predicting high-affinity mutation site combinations of an antibody is further developed, based on the constructed prediction model, the influence of different mutation combinations on the affinity of the antibody can be quickly and accurately predicted, and the construction of high-affinity antibodies is guided.

[0016] In the present application, the mutation site combinations can be ranked from high to low based on the scores, and then the mutation site combinations with high scores are selected for affinity test verification, thereby guiding the construction of high-affinity antibodies.

[0017] Preferably, the method further comprises performing single-point saturation mutation on the antibody to be predicted, and testing the affinity to obtain the mutant antibody and single-point mutation information.

[0018] Preferably, the method further comprises the step of performing affinity test on the antibody containing the candidate antibody high-affinity mutation site combination.

[0019] It can be understood that the present application can be used for predicting any antibody.

[0020] Preferably, the antibody to be predicted comprises a CD138 antibody.

[0021] Preferably, the heavy chain variable region amino acid sequence of the antibody to be predicted comprises the sequence shown in SEQ ID NO. 1.

[0022] Preferably, the light chain variable region amino acid sequence of the antibody to be predicted comprises the sequence shown in SEQ ID NO. 2.

[0023] Preferably, the mutation site combination comprises a heavy chain mutation site combination and a light chain mutation site combination, the heavy chain mutation site combination comprises D54F and S57F, and the light chain mutation site combination comprises a combination of L46G, Y50G, I53F and N92Y, or a combination of L46G, Y50G, I53W and N92Y.

[0024] In a fourth aspect, the present application provides a high-affinity antibody, wherein the antibody contains the mutation site combination predicted by the method for predicting high-affinity mutation site combinations of an antibody according to the third aspect.

[0025] Specifically, the present application provides a high-affinity CD138 antibody, wherein the high-affinity CD138 antibody is mutated based on the mutation site combination predicted by the method for predicting high-affinity mutation site combinations of an antibody according to the third aspect, on the basis of the heavy chain variable region with the amino acid sequence shown in SEQ ID NO. 1 and the light chain variable region with the amino acid sequence shown in SEQ ID NO. 2.

[0026] Preferably, the mutation in the heavy chain variable region comprises D54F and S57F.

[0027] Preferably, the mutations in the light chain variable region include a combination of L46G, Y50G, I53F and N92Y, or a combination of L46G, Y50G, I53W and N92Y.

[0028] In a fifth aspect, the present application provides a device for predicting a combination of high-affinity mutation sites of an antibody, the device comprising a single-point mutation testing unit, a mutation combination predicting unit and a verifying unit.

[0029] The single-point mutation testing unit is configured to perform the following steps:

[0030] Performing single-point saturation mutation on the antibody to be predicted, and testing the affinity to obtain a mutated antibody and single-point mutation information;

[0031] The mutation combination predicting unit is configured to perform the following steps:

[0032] Inputting the amino acid sequence of the mutated antibody into the model for predicting high-affinity mutation sites of an antibody according to the second aspect, outputting the probability of predicting the correct target, summing and averaging the prediction results of multiple mutation sites as the score of the mutation site combination, and selecting the mutation site combination with a high score as a candidate high-affinity mutation site combination of an antibody;

[0033] The verifying unit is configured to perform the following steps:

[0034] Testing the affinity of the antibody containing the candidate high-affinity mutation site combination of an antibody.

[0035] In a sixth aspect, the present application provides an electronic device, comprising at least one processor and a memory connected to the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method for predicting a combination of high-affinity mutation sites of an antibody according to the third aspect.

[0036] In a seventh aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method for predicting a combination of high-affinity mutation sites of an antibody according to the third aspect.

[0037] Compared with the prior art, the present application has at least the following beneficial effects:

[0038] In the present application, the model for predicting high-affinity mutation sites of an antibody is constructed based on the ESM model training and optimization, and the obtained model can effectively predict and evaluate the naturalness of different mutation sequences, and further infer the influence of mutation on the affinity of an antibody, which can be effectively applied to antibody affinity maturation technology to predict effective mutation site combinations and guide the construction of high-affinity antibodies. DETAILED DESCRIPTION

[0039] The technical solutions of the present application are further illustrated by the following specific embodiments. However, the following examples are only simple examples of the present application, and do not represent or limit the protection scope of the present application, and the protection scope of the present application is subject to the claims.

[0040] Unless otherwise specified, the specific techniques or conditions in the examples are carried out according to the techniques or conditions described in the literature in the art, or according to the product instructions. Unless otherwise specified, the reagents or instruments used are conventional products that can be purchased through regular channels.

[0041] Transformer is a deep learning model based on self-attention mechanism. The Transformer model mainly consists of an encoder (Encoder) and a decoder (Decoder). The encoder is responsible for encoding the input sequence into a fixed-length vector, and the decoder predicts the next output based on the output of the encoder and the generated partial output sequence. In the encoder, the input sequence is first converted into a vector representation through the word embedding (Word Embedding) layer. Then, these vectors pass through multiple self-attention layers and feedforward neural network layers. The self-attention layer is used to calculate the correlation between different positions in the sequence, and the feedforward neural network layer further transforms the representation of each position. The decoder also contains multiple self-attention layers and feedforward neural network layers, but unlike the encoder, the self-attention layer in the decoder uses a mask mechanism to prevent seeing future information when predicting the current position. In addition, the self-attention layer in the decoder also considers the output of the encoder, realizing the information exchange between the encoder and the decoder. The Transformer model adopts a multi-head attention (Multi-Head Attention) mechanism in the training process, which divides the input sequence into multiple heads and performs self-attention calculation on each head.

[0042] In the field of amino acid sequence analysis, Masked Language Model (MLM) is a pre-training strategy aimed at improving the model's understanding of protein sequences through masking and prediction techniques. Specifically, the model processes part of the original amino acid sequence by masking, and then uses context information to restore the masked amino acids, thereby learning the internal structure and biological semantics of the sequence. In the implementation process, MLM first randomly selects several positions in the amino acid sequence and replaces the amino acids at these positions with specific masking markers. Then, the model is trained to identify and predict the original amino acids at these masked positions. This process not only helps the model capture sequence dependencies between amino acids, but also enables it to understand the role of amino acids in the formation of protein function and structure. In this way, the MLM pre-trained model can gain a deep understanding of the features of amino acid sequences, which is crucial for various bioinformatics tasks. For example, in the fields of protein function prediction, structure prediction, sequence variation analysis, and protein-protein interaction prediction, models pre-trained by MLM show better performance.

[0043] ESM model (Evolutionary Scale Modeling) is a Transformer-based protein language model mainly used for predicting and generating protein structure and function. ESM model uses deep learning technology to treat protein sequences as a language and each amino acid as a character, and uses an autoregressive neural network to learn the statistical rules of this language. Among them, ESM-2 (Evolutionary Scale Modeling 2) is a new MLM language model developed by Facebook AI Research, aiming to understand and predict the structure and function of protein sequences through pre-training and fine-tuning. It is based on the Transformer architecture and uses an 8-layer Transformer encoder with a model size of 15 billion parameters. ESM-2 performs well in predicting protein three-dimensional structure and can predict at atomic resolution, providing a powerful tool for bioinformatics and protein engineering.

[0044] In the present invention, the model for predicting high-affinity antibody mutation sites is constructed based on the ESM model training and optimization, which can effectively predict and evaluate the naturalness of different mutation sequences, and further infer the influence of mutations on antibody affinity. It can be effectively applied to antibody affinity maturation technology to predict effective mutation site combinations and guide the construction of high-affinity antibodies.

[0045] In an embodiment of the present application, a device for predicting a high-affinity mutation site combination of an antibody is provided, and the device comprises a single-point mutation testing unit, a mutation combination prediction unit and a verification unit; the single-point mutation testing unit is configured to perform single-point saturation mutation on a to-be-predicted antibody, and test the affinity to obtain a mutated antibody and single-point mutation information; the mutation combination prediction unit is configured to perform inputting an amino acid sequence of the mutated antibody into a model for predicting a high-affinity mutation site of an antibody, outputting a probability of a correct target, summing and averaging prediction results of a plurality of mutation sites as a score of a mutation site combination, and selecting a mutation site combination with a high score as a candidate high-affinity mutation site combination of an antibody; and the verification unit is configured to perform affinity testing on an antibody containing the candidate high-affinity mutation site combination of an antibody.

[0046] In another embodiment of the present application, an electronic device is provided, and the electronic device comprises at least one processor and a memory connected to the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method for predicting a high-affinity mutation site combination of an antibody.

[0047] In still another embodiment of the present application, a computer-readable storage medium is provided, and the computer-readable storage medium stores a computer program, and the program is executed by a processor to implement the method for predicting a high-affinity mutation site combination of an antibody.

[0048] Embodiment 1

[0049] In this embodiment, a CD138 antibody is taken as an example to predict a mutation site combination.

[0050] The amino acid sequence of a heavy chain of a CD138 antibody is shown in SEQ ID NO. 1, and the amino acid sequence of a light chain is shown in SEQ ID NO. 2, single-point saturation mutation sites of the heavy chain include H_G50W (representing that the 50th site G of the heavy chain is mutated into W, and other mutations can be inferred in the same way), H_I52W, H_D54F, H_D54Y, H_S57W and H_S57F, and mutation sites of the light chain include L_L46G, L_I48F, L_I48G, L_Y49W, L_Y49F, L_Y50G, L_Y50W, L_Y50G, L_S52G, L_S52W, L_I53W, L_I53F, L_N92F, L_N92Y and L_P95G.

[0051] Based on the ESM model, the model is trained and optimized, and the combination of mutation sites is predicted. Taking G50W as an example, the original sequence replaces the 50th site in the heavy chain (H) with mask, and after prediction by the esm model, the target of this site is W (mask (H_G50), W).

[0052] Specifically, the mutant antibody sequence is input into seq_mask (H_G50).

[0053] (1) embedding layer:

[0054] emb = embedding (seq_mask) + position_embedding (seq_mask_index), and emb is obtained.

[0055] (2) encoder:

[0056] Then, after the self-attention layer and the feedforward network layer, the composite state tensor EmbeddingVec1 is multiplied by three weight matrices to obtain Q, K, and V matrices as inputs of the multi-head attention layer. Since the multi-head design can allow each state to build relevance with other states at different times, there are multiple such Q, K, and V, and the calculation formula of the i-th one is as follows:

[0057] Q = emb.Wq;

[0058] K = emb.Wk;

[0059] V = emb.Wv.

[0060] The attention tensor of the i-th output of the multi-head attention layer includes the following calculation formula:

[0061]

[0062] Where dK is the dimension of K.

[0063] Then, the various attention tensors are spliced and mapped to the original input space as the output MultiHead of the multi-head attention layer.

[0064] MultiHead = concat (head1, head2…headh) W;

[0065] AttentionOut = LayerNorm (EmbeddingVec1 + MutiHead1).

[0066] The obtained multi-head attention layer output AttentionOut1 is then input to a feed-forward network layer to obtain FNNOut1, which is composed of a neural network nested with a ReLU function and then nested with a neural network, and is calculated according to the following formula:

[0067] FFNOut = max(0, AttentionOut * W1 + b1) * W2 + b2.

[0068] The feed-forward network layer also contains a residual connection structure, and the final output EncoderVec of the encoder is obtained after the residual connection, and the calculation formula of the encoding tensor EncoderVec is as follows:

[0069] EncoderVec = LayerNorm(AttentionOut1 + FFNOut1).

[0070] (3) Conversion layer:

[0071] The output is mapped into character probability, and the model is trained to learn the probability with a high score, such as H_G50W, so that the model outputs W as much as possible. When predicting, if the combination of H_G50W and H_I52W is scored, the output is the average of the probability of H chain 50 site outputting W and the probability of H chain 52 site outputting W.

[0072] Based on the light chain scoring, the combination with a relatively high score is: L46G, Y50G, I53F and N92Y combination, with a score of 0.5152; or L46G, Y50G, I53W and N92Y combination, with a score of 0.5443; based on the heavy chain scoring, the combination with a relatively high score is: D54F and S57F, with a score of 0.7.

[0073] Example 2

[0074] This example tests the effect of the mutation combination obtained in Example 1 on affinity.

[0075] (1) Preparation of antibodies of specific mutation combinations

[0076] According to the mutation combination, two CD138 antibodies can be designed based on the heavy chain and light chain of SEQ ID NO. 1 and SEQ ID NO. 2. The mutant antibody 1 is a combination of heavy chain D54F and S57F mutation, and a combination of light chain L46G, Y50G, I53F and N92Y mutation; the mutant antibody 2 is a combination of heavy chain D54F and S57F mutation, and a combination of light chain L46G, Y50G, I53W and N92Y mutation.

[0077] 1) Design primers: design and synthesize primers for antibody heavy and light chain genes, and add appropriate restriction enzyme recognition sequences to the 5' end of the primers.

[0078] 2) Gene synthesis: Use designed primers to splice genes by PCR reaction to obtain light and heavy chain DNA sequences of antibodies.

[0079] 3) Enzymatic digestion: Perform enzyme digestion on PCR products and cloning vectors with the same restriction enzymes to produce sticky ends.

[0080] 4) Ligation reaction: Ligate the antibody gene fragments after enzyme digestion with the vector using DNA ligase.

[0081] 5) Transformation: Transform the ligation product into competent E. coli.

[0082] 6) Screening and identification: Screen positive clones by antibiotics, and perform colony PCR and sequence analysis to confirm the correctness of the inserted fragments.

[0083] 7) Expression and purification: Transform the verified correct plasmid into a eukaryotic or prokaryotic expression system, induce expression of antibody protein, and perform protein purification by affinity chromatography and the like.

[0084] (2) Affinity test

[0085] 1) Ligand capture: Dilute the purified antibody with running reagents to an appropriate concentration to fix it on the surface of the PAHC200M chip.

[0086] 2) Analyte dilution: Dilute the analyte with running reagents by 3-fold. Inject the diluted analyte from the lowest concentration to the highest concentration onto the chip surface in turn, and bind and dissociate for corresponding time.

[0087] 3) Baseline equilibration: Perform multiple cycles of solution circulation before running the analyte for baseline equilibration. Both binding and dissociation steps are performed in running reagents.

[0088] 4) Chip regeneration: After the circulation of all analyte concentrations is completed, regenerate the chip with 10 mM Glycine, pH 2.0 for 3 times, 20 s each time, to wash away the ligand and non-dissociated analyte.

[0089] 5) Data collection and analysis: After collecting the data, use the Carterra software to perform kinetic analysis. Use the 1:1 Langmuir binding model or other suitable models to fit the data. Calculate the KD value.

[0090] The affinity results are shown in Table 1, indicating that the combination of the predicted mutation sites of the application can significantly improve the affinity of the antibody.

[0091] Table 1

[0092]

[0093] To sum up, in the application, the model for predicting high-affinity mutation sites of antibodies is constructed based on the ESM model training optimization, the obtained model can effectively predict and evaluate the naturalness of different mutation sequences, and then infer the influence of mutations on the affinity of antibodies, and can be effectively applied to the antibody affinity maturation technology to predict effective mutation site combinations and guide the construction of high-affinity antibodies.

[0094] The applicant declares that the above is only a specific embodiment of the application, but the protection scope of the application is not limited thereto, and those skilled in the art should understand that any changes or replacements within the technical scope disclosed by the application, which can be easily thought of by any person skilled in the art, falls within the protection scope and disclosure scope of the application.

Claims

1. A high affinity CD 138 antibody, characterized in that, The high-affinity CD138 antibody is mutated at a combination of mutation sites predicted by the method for predicting a combination of high-affinity mutation sites of an antibody, based on a heavy chain variable region with an amino acid sequence as shown in SEQ ID NO. 1 and a light chain variable region with an amino acid sequence as shown in SEQ ID NO. 2; the mutations in the heavy chain variable region include D54F and S57F; the mutations in the light chain variable region include a combination of L46G, Y50G, I53F and N92Y, or a combination of L46G, Y50G, I53W and N92Y.

2. The high affinity CD138 antibody according to claim 1, characterized in that The method for predicting a combination of high-affinity mutation sites of an antibody comprises: inputting the amino acid sequence of the mutant antibody into the model for predicting high-affinity mutation sites of an antibody, masking all single mutation sites using a masking language model, then predicting the amino acid sequence of the masked site, outputting the probability of predicting the correct target, adding and averaging the prediction results of multiple mutation sites as the score of the combination of mutation sites, and selecting the combination of mutation sites with a high score as the candidate combination of high-affinity mutation sites of an antibody.

3. The high affinity CD138 antibody according to claim 2, characterized in that The method for predicting a combination of high-affinity mutation sites of an antibody further comprises performing single-point saturation mutation on the antibody to be predicted and testing the affinity to obtain the mutant antibody and single-point mutation information.

4. The high affinity CD138 antibody according to claim 2, characterized in that, The method for constructing the model for predicting high-affinity mutation sites of an antibody comprises: using the amino acid sequence of the mutant antibody with improved affinity as the training set, the training set including the amino acid sequence of the mutant antibody, the mutation site and the amino acid information after mutation, and using the amino acid after mutation as the correct prediction target; training the ESM model using the training set, the training including masking the amino acid sequence of the mutation site of the mutant antibody using a masking language model, then predicting the amino acid sequence of the masked site, outputting the probability of predicting the correct target, adjusting the model according to the output result to obtain the model for predicting high-affinity mutation sites of an antibody; when there is more than one masked mutation site, the average of the sum of the probabilities of predicting the correct target is used as the output result; the adjustment is performed until the output result is not lower than a preset threshold, and the threshold is 0.

5.

5. A device for predicting combinations of antibody high-affinity mutation sites, characterized in that, The device comprises a single-point mutation testing unit, a mutation combination prediction unit and a verification unit; The single-point mutation testing unit is configured to perform the following steps: performing single-point saturation mutation on the antibody to be predicted and testing the affinity to obtain the mutant antibody and single-point mutation information; The mutation combination prediction unit is configured to perform the following steps: inputting the amino acid sequence of the mutant antibody into the model for predicting high-affinity mutation sites of an antibody as claimed in claim 4, outputting the probability of predicting the correct target, adding and averaging the prediction results of multiple mutation sites as the score of the combination of mutation sites, and selecting the combination of mutation sites with a high score as the candidate combination of high-affinity mutation sites of an antibody; The verification unit is configured to perform the following steps: testing the affinity of the antibody containing the candidate combination of high-affinity mutation sites of an antibody.

6. An electronic device, comprising: The electronic device includes at least one processor, and a memory connected to the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method for predicting a combination of high-affinity mutation sites of an antibody as claimed in claim 2.

7. A computer readable storage medium having stored thereon a computer program, characterized in that The program is executed by the processor to implement the method for predicting a combination of high-affinity mutation sites of an antibody as claimed in claim 2.

Images