A method for antibody design targeting a target epitope

Abstract

The application discloses a kind of antibody design methods for targeting target epitope, belong to artificial intelligence technology and computational structural biology technology technical field, it includes: step 1, epitope information acquisition: using python code to obtain epitope site information on target antigen structure;Step 2, mask drive;Step 3, conformation generation;Step 4, fitting verification: the designed antibody is simulated to dock with target antigen in digital space;Step 5, energy efficiency evaluation: the binding energy of the designed antibody and target antigen is calculated using the code.The application re-designs the template antibody sequence by using the target antigen information and the target epitope information, so as to achieve the antibody sequence that can target the specified epitope and has high affinity, and when there is no corresponding targeting antibody at a certain epitope, a large number of candidate targeting antibodies can be designed in a short time to reduce the harm that may be caused by pathogens.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence technology and computational structural biology, and in particular to an antibody design method targeting epitopes. Background Technology

[0002] Antibodies, as immunoglobulins produced by antigen stimulation, possess the biological activity of specifically binding to specific antigenic epitopes and are the core effector molecules of the body's immune response. In recent years, therapeutic antibody drugs have played an increasingly important role in the clinical treatment of tumors, autoimmune diseases, and infectious diseases due to their high targeting and controllable safety profile.

[0003] Current antibody development mainly relies on in vitro screening technology systems, and its standard procedures include:

[0004] Immune library construction: Obtaining diverse antibody sequence groups through animal model immunization or in vitro display techniques;

[0005] High-throughput screening: Multiple rounds of affinity screening are performed using techniques such as enzyme-linked immunosorbent assay (ELISA), surface plasmon resonance (SPR), or flow cytometry.

[0006] Functional validation: In vitro / in vivo validation of candidate antibodies for binding activity, neutralizing efficacy, and cytotoxicity.

[0007] This technical approach has the following core flaws:

[0008] The development cycle is lengthy: it typically takes 6-18 months from antigen immunization to obtaining a preclinical candidate molecule.

[0009] High economic costs: Monoclonal antibody development costs an average of $2 million to $5 million;

[0010] Screening efficiency is limited: the library capacity is constrained by experimental throughput, resulting in a relatively high rate of missed screening of high-affinity antibodies;

[0011] Insufficient controllability of targeting: It is difficult to precisely control the binding specificity of antibodies to specific antigenic epitopes; therefore, we propose an antibody design method that targets the target epitope to solve this problem. Summary of the Invention

[0012] The purpose of this invention is to provide an antibody design method that targets a specific epitope, in order to solve the problems mentioned in the background art.

[0013] To achieve the above objectives, the present invention adopts the following technical solution:

[0014] An antibody design method targeting a specific epitope includes:

[0015] Step 1: Epitope Information Acquisition: Use Python code to obtain epitope site information on the target antigen structure;

[0016] Step 2, Mask-driven: Using epitope site information and target antigen structure, the target template antibody sequence is redesigned based on antibody conserved amino acid alignment analysis and shadow masking mechanism;

[0017] Step 3, Conformation generation: Use protein language models to predict the structure of the redesigned antibody sequence, screen and obtain the most reasonable antibody structure;

[0018] Step 4, mating verification: Simulate the docking of the designed antibody and the target antigen in digital space;

[0019] Step 5, Energy efficiency assessment: Use code to calculate the binding energy between the designed antibody and the target antigen.

[0020] Preferably, the template antibody sequence in step 2 can be an antibody sequence targeting other epitopes of the antigen or an antibody sequence targeting other antigens.

[0021] Preferably, the specific steps of the antibody-based conserved amino acid alignment analysis in step 2 are as follows:

[0022] Step 201: Taking advantage of the relatively conserved amino acid occurrences in the FR region of the antibody sequence, a library is constructed and compared for the FR region of known antibodies.

[0023] Step 202: Define the amino acid and its site that appear most frequently and whose amino acid type accounts for more than 95% of the total as template amino acids;

[0024] Step 203: In actual analysis, the template antibody sequence is aligned with it to obtain the possible structural template of the FR region of the template sequence.

[0025] Preferably, the method for redesigning the target template antibody sequence in step 2 is as follows: a shaded paratope is placed next to a specific epitope, the shaded paratope is roughly combined with the structural template, and the paratope is initially restricted using the structural template and antigen epitope information; the probability of the amino acid that may appear at each site of the paratope is calculated, and the one with the highest probability is selected as the sequence of the paratope.

[0026] Preferably, in step 3, when screening and obtaining the most reasonable antibody structure, the local confidence level and IPTM value of the predicted structure are calculated to evaluate the reasonableness of the predicted structure. Plddt and IPTM > 75 are required as screening criteria. If multiple structures meet the requirements, the structure with the highest plddt and IPTM values ​​is selected as the candidate structure.

[0027] Preferably, in step 4, the specific steps for simulating docking are as follows:

[0028] Step 401: Simulate antigen-antibody docking in space using a rigid docking method.

[0029] Step 402: Determine whether its approximate docking site is an epitope site;

[0030] Step 403: Keep the antigen position unchanged;

[0031] Step 404: Fine-tune the antibody position in space in multiple steps to find the model in which the two combine in space that best matches the natural situation.

[0032] Preferably, in step 4, after each simulated docking process, a relaxation is required to reduce unreasonable structures that may occur during rigid docking and to further simulate the minute deformations that occur during protein docking.

[0033] Preferably, after the two simulated dockings in step 4, the predicted model needs to be further screened. The screening criteria are mainly the bonding surface and the overall energy score. The bonding surface (dSASA) needs to be greater than >1200, and the overall energy score needs to be less than <1000.

[0034] Preferably, step 5 includes the following steps: weighting the predicted antigen-antibody complex structure by summing the energy terms and subtracting the weighted sum of the energy of the antigen and antibody themselves to calculate an approximate value of the binding energy.

[0035] The beneficial effects of this invention are as follows:

[0036] 1. The antibody design method targeting a target epitope described in this invention redesigns the template antibody sequence using target antigen information and target epitope information, thereby enabling the design of antibody sequences that can target a specified epitope and have high affinity. At the same time, it uses spatial structure modeling to predict the structure and simulates and detects in digital space whether the designed antibody can bind to the target epitope and the binding situation under natural conditions. Compared with traditional antibody design schemes, it has the advantages of high efficiency and low cost. It can design a large number of candidate target antibodies in a short time when there is no corresponding target antibody for a certain epitope, in order to reduce the harm that pathogens may cause.

[0037] 2. The antibody design method targeting a target epitope described in this invention calculates and resolves the spatial location of the target epitope and performs comparative analysis with a highly conserved structural template of the antibody's FR region. It innovatively uses a shadow masking mechanism to dynamically constrain and guide the design of the binding site of the template antibody, enabling the newly generated Paratope sequence to accurately match the target epitope. This method eliminates the dependence on natural antibodies for specific targets and can efficiently generate targeted antibody sequences directly based on key antigen epitope information in the absence of readily available antibodies, significantly improving the targeting and feasibility of the design.

[0038] 3. The antibody design method targeting epitopes described in this invention integrates cutting-edge protein language models for high-quality antibody structure prediction and employs rigorous local conformation confidence and overall folding quality assessment indicators to screen the most reasonable structural model. Subsequently, an innovative two-stage simulation docking process is used: first, coarse spatial localization is performed to verify the epitope binding possibility; then, fine-tuning and multiple rounds of energy optimization are performed to simulate real protein binding dynamics and minor conformational adjustments. Finally, the quality and affinity of the final complex are rigorously evaluated through multi-dimensional physical indicators such as binding surface area and comprehensive binding energy intensity, forming a rigorous design-generation-verification closed loop to ensure the structural feasibility and binding efficacy of the antibody design.

[0039] 4. The antibody design method targeting a target epitope described in this invention uses antibody sequences from any source, regardless of whether they target the same antigen or other epitopes, or even completely different antigens. Through conserved FR region backbone template comparison analysis and shadow mask-guided reprogramming, this method can effectively utilize the structural framework information in existing antibody resources and redesign their binding sites to match the target epitope. This high flexibility greatly expands the source of design resources and effectively overcomes the obstacles caused by the lack of template antibodies corresponding to specific epitopes, making it extremely valuable in the development of novel pathogen antibodies and multi-epitope targeting strategies.

[0040] 5. The antibody design method targeting epitopes described in this invention can be efficiently completed in a computing environment, from antigen epitope information extraction, template sequence redesign, structure prediction and screening to simulated docking and binding energy evaluation. Compared with traditional methods that rely on a large number of physical experiments, such as time-consuming and labor-intensive animal immunization or library screening techniques, this method can generate and screen a large number of candidate antibody designs in parallel in a very short time, significantly shortening the research and development cycle, greatly reducing the early research and development costs and resource consumption, and providing an efficient "digital antibody design" response capability for emerging infectious disease outbreaks or rapidly iterating pathogens. Attached Figure Description

[0041] Figure 1This is a schematic diagram illustrating the process of designing targeted antibodies against antigen epitopes according to an embodiment of the present invention;

[0042] Figure 2 The diagram shows the structure of the S-protein of the antigen human respiratory syncytial virus (RSV) in the examples, with the target epitope siteIV marked in dark.

[0043] Figure 3 The antibody structure diagram is redesigned according to the method described in this invention, wherein the H chain structure is marked in dark and the L chain structure is marked in light.

[0044] Figure 4 The predicted complex structure of the S-protein of the target antigen RSV and the designed antibody. Detailed Implementation

[0045] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0046] Reference Figures 1-4 An antibody design method targeting a specific epitope includes:

[0047] Step 1: Epitope Information Acquisition: Use Python code to obtain epitope site information on the target antigen structure.

[0048] Step 2, Mask-driven: Using epitope site information and target antigen structure, the target template antibody sequence is redesigned based on antibody conserved amino acid alignment analysis and shadow masking mechanism.

[0049] Step 3, Conformation generation: Use protein language models to predict the structure of the redesigned antibody sequence, screen and obtain the most reasonable antibody structure.

[0050] Step 4, docking verification: Simulate docking between the designed antibody and the target antigen in digital space.

[0051] Step 5, Energy efficiency assessment: Use code to calculate the binding energy between the designed antibody and the target antigen.

[0052] In this embodiment, the template antibody sequence in step 2 can be an antibody sequence that targets other epitopes of the antigen or an antibody sequence that targets other antigens.

[0053] In this embodiment, the specific steps of the antibody-based conserved amino acid alignment analysis in step 2 are as follows:

[0054] Step 201: Taking advantage of the relatively conserved amino acid occurrences in the FR region of the antibody sequence, a library is constructed and compared for the FR region of known antibodies.

[0055] Step 202: Define the amino acid and its site that appear most frequently and whose amino acid type accounts for more than 95% of the total as template amino acids;

[0056] Step 203: In actual analysis, the template antibody sequence is aligned with it to obtain the possible structural template of the FR region of the template sequence.

[0057] In this embodiment, the method for redesigning the target template antibody sequence in step 2 is as follows: a shaded paratope is placed next to a specific epitope, the shaded paratope is roughly combined with the structural template, and the paratope is initially restricted using the structural template and antigen epitope information; the probability of the amino acid that may appear at each site of the paratope is calculated, and the sequence with the highest probability is selected as the paratope sequence.

[0058] In this embodiment, when screening and obtaining the most reasonable antibody structure in step 3, the local confidence level and IPTM value of the predicted structure are calculated to evaluate the reasonableness of the predicted structure. Plddt and IPTM > 75 are required as screening criteria. If multiple structures meet the requirements, the structure with the highest plddt and IPTM values ​​is selected as the candidate structure.

[0059] In this embodiment, the specific steps for simulating docking in step 4 are as follows:

[0060] Step 401: Simulate antigen-antibody docking in space using a rigid docking method.

[0061] Step 402: Determine whether its approximate docking site is an epitope site;

[0062] Step 403: Keep the antigen position unchanged;

[0063] Step 404: Fine-tune the antibody position in space in multiple steps to find the model in which the two combine in space that best matches the natural situation.

[0064] In this embodiment, in step 4, after each simulated docking process, a relaxation is required to reduce unreasonable structures that may occur during rigid docking and to further simulate the small deformations that occur during protein docking.

[0065] In this embodiment, after the two simulated dockings in step 4, the predicted model needs to be further screened. The screening criteria are mainly the bonding surface and the overall energy score. The bonding surface (dSASA) needs to be greater than >1200, and the overall energy score needs to be less than <1000.

[0066] In this embodiment, step 5 specifically involves the following steps: weighting the predicted antigen-antibody complex structure by summing the energy terms and subtracting the weighted sum of the energy of the antigen and antibody themselves to calculate an approximate value of the binding energy between the two.

[0067] In this embodiment, a targeted antibody was designed for the siteIV epitope on the S-protein of RSV.

[0068] First, the S-protein structure of RSV was searched and downloaded from public protein databases such as PDB. Based on current reports, the site IV on the S-protein was marked (422-471), and the epitope sequence is: CTASNKNRGIIKTFSNGCDYVSNKGMDTVSVGNTLYYVNKQEGKSLYVKG. The antigen structure diagram is shown below. Figure 2 As shown, a Python script is used to save the anatomical position information as a JSON file for later use.

[0069] Based on the antigen information and the JSON file containing the epitope, the original antibody sequence (PDB number: 4JHW) targeting another epitope of RSV - siteIII is as follows: H-chain: QVQLVQSGAEVKKPGSSVMVSCQASGGPLRNYIINWLRQAPGQGPEWMGGIIPVLGTVHYAPKFQGRVTITADESTDTAYIHLISLRSEDTAMYYCATETALVVSTTYLPHYFDNWGQGTLVTVSSASTKGPSVFPLAPSSKSTSGGTAALGCLVKDYFPEPVTVSWNSGALTSGVHTFPAVLQSSGLYSLSSVVTVPSSSLGTQTYICNVNHKPSNTKVDKKVEPKSC;

[0070] L-chain:DIQMTQSPSSLSAAVGDRVTITCQASQDIVNYLNWYQQKPGKAPKLLIYVASNLET GVPSRFSGSGSGTDFSLTISSLQPEDVATYYCQQYDNLPLTFGGGTKVEIKRTVAAPSVFIFPPSDEQ LKSGTASVVCLLNNFYPREAKVQWKVDNALQSGNSQESVTEQDSKDSTYSLSSTLTLSKADYEKHKVYACEVTHQGLSSPVTKSFNRGEC; Based on the conserved amino acid analysis of the antibody, the corresponding conserved template structure of the template antibody sequence was obtained; and all 6 CDR sequences of the template antibody sequence were shaded, with CDR-H3 as the main shaded paratope. The probability of amino acid occurrence at all positions was calculated to obtain the candidate antibody sequence, and further completion processing was performed using a protein language model to ensure the sequence integrity of the designed antibody, as well as its targeting and high affinity for epitopes. The redesigned antibody sequence is: H-chain: QVQLVQSGASEVKKPGSSVMVSCQASGGTFTSTFSTYAINWLRQAPGQGPEWMGGIIPIIPFGTAHYA PKFQQGRVTITADESTDTAYIHLISLRSEDTAMYYCARDRGYGYYYFDYWGQGTLVTVS; TGVPPSRFSGSGSGSGTDFSLTISSLQPEDVATYYCQQYDNYSYTLPLTFGGGTKVEIK.

[0071] The structure prediction tool is run on the server. AlphaFold3 or RosettaFold2 can be selected for structure prediction. Here, the spatial structure of the redesigned antibody sequence is predicted, and plddt, iptm, and ptm are calculated for each predicted structure. These are then integrated and used as a ranking score to evaluate the rationality of the structure. The ranking scores are sorted from high to low, and the top 10 ranking scores are shown in Table 1. The structure with the highest ranking score is selected as the designed antibody structure and saved in a file format that the computer can recognize, such as a pdb file.

[0072] (See antibody structure design) Figure 3 )

[0073] 1 0 0.846464 1 1 0.849866 1 2 0.840173 1 3 0.854403 23 0 0.846972 23 1 0.847134 23 2 0.839615 23 3 0.851606 23 4 0.844674

[0074] Table 1

[0075] Using the predicted antibody structure, we first calculate the possible docking modes and situations of the antibody and antigen under natural conditions using rigid docking. Here, we set the antibody as the receptor and the antigen as the ligand. That is, while keeping the antibody's posture unchanged, we change the position and orientation of the antigen in space to observe its binding model with the antibody and determine whether the designed antibody can dock to the corresponding antigen epitope. After confirming that the antibody can dock to the corresponding antigen epitope, we relax the corresponding model to reduce potential unreasonable structures caused by rigid docking. After relaxing, in order to further find the antigen-antibody docking state that best matches the natural situation, we perform multiple fine-tuning steps such as rotating and slightly translating the antibody around the binding site while keeping the antigen position unchanged, and observe the differences in the binding interface and energy as a result. At the same time, we screen these models based on the binding surface (dSASA) and the overall energy score, and also screen multiple models based on the rationality of the binding interface. The screened results are relaxed to simulate the small deformation of the protein itself during the docking process for binding, and further reduce unreasonable conformations caused by rigid docking. (The final antibody-antigen binding diagram is shown below) Figure 4 )

[0076] Finally, we calculated the binding energy of the relaxed results and performed a final analysis of the binding interface. To compare the performance of the designed antibody against the natural antibody, we also calculated the binding energy of the natural antibody (PDB number: 8WSQ) targeting the RSV S-protein at siteIV. The calculated binding energies of the designed antibody and the natural antibody 8WSQ are shown in Table 2.

[0077] Design antibodies 51.239 8WSQ 48.878

[0078] Table 2

[0079] In summary, the antibody design method of this invention can efficiently generate antibody sequences and structures with high affinity and targeting specific antigenic epitopes in digital space using only computers. This provides important support for subsequent antibody drug development, significantly reducing experimental costs and eliminating the limitations of wet experimental conditions. It allows for the exploration of all possible sequences and conformations during the generation process, and enables the artificial control of antibody generation and design goals through parameter setting. This improves the efficiency and quality of antibody engineering, reduces trial-and-error costs in wet experimental processes, increases the efficiency of antibody generation and expression, and provides more drug development possibilities for antibody molecules.

[0080] The above provides a detailed description of an antibody design method targeting a specific epitope provided by this invention. Specific embodiments have been used to illustrate the principles and implementation methods of this invention. These embodiments are merely illustrative and are intended to aid in understanding the method and core concepts of this invention. It should be noted that those skilled in the art can make various improvements and modifications to this invention without departing from its principles, and these improvements and modifications also fall within the scope of protection of the claims of this invention.

Claims

1. A method for designing antibodies targeting a specific epitope, characterized in that, include: Step 1: Epitope Information Acquisition: Use Python code to obtain epitope site information on the target antigen structure; Step 2, Mask-driven: Using epitope site information and target antigen structure, the target template antibody sequence is redesigned based on antibody conserved amino acid alignment analysis and shadow masking mechanism; Step 3, Conformation generation: Use protein language models to predict the structure of the redesigned antibody sequence, screen and obtain the most reasonable antibody structure; Step 4, mating verification: Simulate the docking of the designed antibody and the target antigen in digital space; Step 5, Energy Efficiency Assessment: Calculate the binding energy between the designed antibody and the target antigen using code; The template antibody sequence mentioned in step 2 is an antibody sequence that targets other epitopes of the antigen or an antibody sequence that targets other antigens; The specific steps of the antibody-based conserved amino acid alignment analysis described in step 2 are as follows: Step 201: Taking advantage of the relatively conserved amino acid occurrences in the FR region of the antibody sequence, a library is constructed and compared for the FR region of known antibodies. Step 202: Define the amino acid and its site that appear most frequently and whose amino acid type accounts for more than 95% of the total as template amino acids; Step 203: In actual analysis, the template antibody sequence is aligned with it to obtain the possible structural template of the FR region of the template sequence; The method for redesigning the target template antibody sequence described in step 2 is as follows: a shaded paratope is placed next to a specific epitope, the shaded paratope is bound to the structural template, and the paratope is initially restricted using the structural template and antigen epitope information; the probability of the amino acid appearing at each site of the paratope is calculated, and the sequence with the highest probability is selected as the paratope sequence. In step 3, when screening and obtaining the most reasonable antibody structure, the local confidence level and IPTM value of the predicted structure are calculated to evaluate the reasonableness of the predicted structure. The screening criteria are plddt>75 and IPTM>75 respectively. If multiple structures meet the requirements at the same time, the structure with the highest plddt and IPTM values ​​is selected as the candidate structure. In step 4, the specific steps for simulating docking are as follows: Step 401: Simulate antigen-antibody docking in space using a rigid docking method. Step 402: Determine whether its approximate docking site is an epitope site; Step 403: Keep the antigen position unchanged; Step 404: Fine-tune the antibody position in space in multiple steps to find the model in which the two combine in space that best matches the natural situation; In step 4, after each simulated docking process, a relaxation is required to reduce unreasonable structures that may occur during rigid docking and to further simulate the small deformations that occur during protein docking. After the two simulated dockings in step 4 are completed, the predicted model needs to be further screened. The screening criteria are mainly the bonding surface and the overall energy score. The bonding surface score needs to be >1200, and the overall energy score needs to be <-1000. In step 5, the specific steps are as follows: the predicted antigen-antibody complex structure is weighted and summed for each energy term, and the difference is calculated with the weighted sum of the energy of the antigen and antibody themselves to calculate the approximate value of the binding energy of the two.

Images