Automatic screening and optimizing method and system for bispecific immune receptors

Abstract

The invention provides an automatic screening and optimizing method and system for bispecific immune receptors, and belongs to the field of immunotherapy. The method comprises the following steps: acquiring a to-be-screened bispecific immune receptor sequence, a first target antigen sequence and a second target antigen sequence, and presetting a mutation domain of a bispecific immune receptor; performing parallel conformation prediction on the bispecific immune receptor sequence and the two target antigens based on the structure prediction model, and screening out an optimal conformation based on confidence; generating candidate objects carrying different mutation combinations through a sequence generation model based on the mutation domain of the bispecific immune receptor in the optimal conformation; constructing a multi-target scoring function, and carrying out at least two rounds of screening on the candidate objects: screening a first candidate set according to stability and interface indexes in the first round, and carrying out parallel docking simulation calculation on the first candidate set and the two target antigens in the second round to calculate docking energy and interface residue annotations, and updating the scoring function based on a calculation result to obtain an optimized bispecific immune receptor, thereby realizing automatic screening and optimization of the bispecific immune receptor.

Description

Technical Field

[0001] This invention relates to the field of immunotherapy technology, and in particular to an automated method and system for screening and optimizing bispecific immune receptors. Background Technology

[0002] In tumor immunotherapy, T-cell receptor (TCR) and chimeric antigen receptor (CAR) technologies are core approaches, showing significant efficacy in the treatment of hematologic malignancies. Bispecific chimeric T-cell receptor (Bi-ChTCR), as an advanced technology, introduces different antigen-binding domains (such as scFv) into the TCR α chain (TRAC) and β chain (TRBC), respectively. It can simultaneously recognize two tumor-associated antigens and bind to the endogenous CD3 complex to achieve TCR-like signal transduction. Theoretically, it can reduce the risk of single-target escape, lower the T-cell activation threshold and the probability of functional exhaustion, and modulate the signal intensity. It has shown enhanced cytotoxicity in some tumor models, providing a new direction for the treatment of solid tumors.

[0003] Despite the outstanding potential of Bi-ChTCR and the technical support provided by AI tools such as AlphaFold2 / 3 and FoldX for its structural optimization, solid tumor treatment still faces the bottleneck of traditional immune receptor technology, and the design of Bi-ChTCR itself also has problems, making it difficult to fully exert its effects.

[0004] Existing technologies have significant drawbacks: First, traditional single-target receptors are prone to failure due to tumor antigen heterogeneity. Although Bi-ChTCRs can recognize dual targets, their design relies on manual experience in splicing, often resulting in problems such as chain entanglement and excessive tension in the linker region, leading to conformational instability and affecting membrane expression and signal transduction. Second, receptor optimization relies on manual adjustment and experimental screening, which is inefficient, costly, and cannot cover a large-scale mutation space. Third, there is a lack of integrated tools. AI structure prediction and functional verification are disconnected, and there is no closed-loop system that integrates interface energy scoring, mutation selection, and visualization, making it difficult to achieve automated iterative optimization. This severely restricts the application of Bi-ChTCRs in the treatment of solid tumors. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes an automated screening and optimization method and system for bispecific immune receptors, which improves screening efficiency and accuracy while ensuring balanced binding of the two targets, providing strong support for the research and development of immunotherapy for solid tumors.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an automated screening and optimization method for bispecific immune receptors, comprising: Obtain the sequence of the bispecific immune receptor to be screened, and the sequences of the first and second target antigens, and predetermine the mutation domain of the bispecific immune receptor; The bispecific immune receptor sequence and two target antigens were predicted in parallel using a structural prediction model, and the optimal conformation was selected based on confidence level. Based on the preset mutation domain of the bispecific immune receptor in the optimal conformation, candidate mutation combinations are generated in the mutation domain by a sequence generation model. After filtering out invalid sequences, multiple bispecific immune receptors carrying different mutation combinations are obtained as candidate objects. A multi-objective scoring function is constructed to screen candidates in two rounds. Specifically, the first round screens the first candidate set based on stability and interface indicators. The second round performs parallel docking simulations on the first candidate set and two target antigens, calculates docking energy and interface residue annotations, updates the scoring function based on the calculation results, and obtains the optimized bispecific immune receptor sequence and indicator report for the second candidate set, thus completing the automated screening.

[0007] Secondly, the present invention provides an automated screening and optimization system for bispecific immune receptors, comprising: An initialization unit is used to acquire the sequence of the bispecific immune receptor to be screened, and the sequences of the first and second target antigens, and to preset the mutation domain of the bispecific immune receptor. The conformation prediction unit is used to perform parallel conformation prediction of the bispecific immune receptor sequence and two target antigens based on the structure prediction model, and to select the optimal conformation based on the confidence level. The mutation unit is used to generate candidate mutation combinations within the mutation domain based on a preset mutation domain of the bispecific immune receptor in the optimal conformation, using a sequence generation model. After filtering out invalid sequences, multiple bispecific immune receptors carrying different mutation combinations are obtained as candidate objects. The screening unit is used to construct a multi-objective scoring function and perform two rounds of screening for candidate objects. Specifically, the first round screens the first candidate set based on stability and interface indicators. The second round performs parallel docking simulations on the first candidate set and two target antigens, calculates docking energy and interface residue annotations, updates the scoring function based on the calculation results, and obtains the optimized bispecific immune receptor sequence and indicator report of the second candidate set, thus completing the automated screening.

[0008] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the automated screening and optimization method for bispecific immune receptors described in the first aspect.

[0009] Fourthly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the automated screening and optimization method for bispecific immune receptors described in the first aspect.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention pre-defines mutation domains and combines structural prediction models with parallel conformation prediction to accurately pinpoint the optimization range, avoiding chain entanglement and conformational instability issues caused by human experience, thus ensuring the expression potential on the receptor membrane. Simultaneously, a sequence generation model generates candidate mutation combinations in batches, coupled with two rounds of screening. The first round uses stability and interface indicators for initial screening, while the second round uses parallel docking simulation to update the scores for further screening, covering a large-scale mutation space and replacing manual adjustments, which significantly shortens the design cycle. Furthermore, the multi-objective scoring function is integrated with automated processes to achieve linkage between AI prediction and functional verification. The output results can be directly connected to experimental systems, reducing manual data processing and providing reliable technical support for the efficient development of bispecific immune receptors for solid tumor treatment.

[0011] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0012] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute a limitation thereof.

[0013] Figure 1 The main flowchart of an automated screening and optimization method for bispecific immune receptors provided in an embodiment of the present invention; Figure 2 A flowchart illustrating an automated screening and optimization method for bispecific immune receptors provided in an embodiment of the present invention; Figure 3 This is a simplified schematic diagram of the BiChTCR structure provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of surface expression analysis provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the NFAT activation function test provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of an in vitro killing experiment provided in an embodiment of the present invention; Figure 7 A schematic diagram illustrating the success rate of an embodiment of the present invention; Figure 8 A schematic diagram illustrating the reduction rate of manual workload provided in an embodiment of the present invention; Figure 9 This is a schematic diagram showing the reduction rate distribution for each task provided in an embodiment of the present invention. Detailed Implementation

[0014] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0015] Explanation of technical terms 1. Bispecific immune receptor (dual-target immune receptor): A receptor construct with two independent or functionally linked binding domains (targeting different targets A and B) for simultaneous or selective recognition of two antigens. It can be a bispecific TCR, CAR, antibody-derived construct, etc., and this invention does not limit the specific scaffold.

[0016] 2. Binding domain / binding arm: The functional fragment that enables binding to the antigen. For antibody types, this corresponds to VH / VL (heavy chain / light chain variable region); for TCR types, it corresponds to TRα / TRβ variable region.

[0017] 3. CDR / Frame Region (FR): Complementarity-determining loops (CDR1–CDR3) and structural scaffold regions (FR1–FR4) within the antibody variable region.

[0018] 4. IMGT / Kabat numbering: Two universal site numbering systems; in this invention, IMGT is preferred for the TCR chain and Kabat is preferred for the antibody variable region, which facilitates cross-sequence alignment and reproducibility.

[0019] 5. Linker peptides ):Depend on A flexible connection sequence formed by repeating the unit n times is used to adjust the spacing and relative orientation between two domains; n is a configurable parameter.

[0020] 6. ΔΔG (change in binding free energy): ΔΔG < 0 indicates that the receptor-antigen binding is more stable / tighter after mutation. The more negative the value, the better the binding.

[0021] 7. BSA (Buried Surface Area): The surface area within the interface where the receptor and antigen are in contact and "buried" when they bind. A larger BSA usually means a more extensive and stable binding interface.

[0022] Example 1 like Figure 1 As shown in the figure, this embodiment discloses an automated screening and optimization method for bispecific immune receptors, including the following steps: S1: Obtain the sequence of the bispecific immune receptor to be screened, and the sequences of the first and second target antigens, and pre-define the mutation domain of the bispecific immune receptor; S2: Based on the structural prediction model, the bispecific immune receptor sequence and the two target antigens are predicted in parallel, and the optimal conformation is selected based on the confidence level. S3: Based on the preset mutation domain of the bispecific immune receptor in the optimal conformation, candidate mutation combinations are generated in the mutation domain through a sequence generation model. After filtering out invalid sequences, multiple bispecific immune receptors carrying different mutation combinations are obtained as candidate objects. S4: Construct a multi-objective scoring function to perform two rounds of screening for candidate objects. Specifically, in the first round, the first candidate set is screened based on stability and interface indicators. In the second round, parallel docking simulations are performed on the first candidate set and two target antigens to calculate docking energy and interface residue annotations. Based on the calculation results, the scoring function is updated and the optimized bispecific immune receptor sequence and indicator report of the second candidate set are obtained, thus completing the automated screening.

[0023] Next, combined Figure 2 This embodiment provides a detailed description of an automated screening and optimization method for bispecific immune receptors.

[0024] In S1, firstly, basic data is collected and filtering conditions are preset.

[0025] Obtain the bispecific immune receptor (BiChTCR) to be screened, as well as the first and second target antigen sequences. Extract the BiChTCR α and β chain sequences and the target antigen sequence / structure file.

[0026] Simultaneously, preset mutation regions (such as α chain CDR3), optimization targets (stability priority / affinity priority) and the number of mutations are defined.

[0027] like Figure 3 The diagram illustrates the structural composition of a Bi-ChTCR (bispecific chimeric T-cell receptor) with scFv-A and scFv-B as recognition modules. Each scFv module consists of a pair of light chain variable regions (VL) and heavy chain variable regions (VH), which are connected to the constant regions of the TCR α chain (Cα) and β chain (Cβ), respectively. The transmembrane region utilizes the CD28™ region, and the intracellular signaling structure includes a 4-1BB co-stimulatory domain and a CD3ζ activation domain. The figure also illustrates the ε, δ, and γ subunits of the CD3 complex and the IgG4 hinge structure. This structure enables dual antigen recognition and highly efficient T-cell activation, making it suitable for research on dual-target solid tumor immunotherapy.

[0028] The preset mutation domain can be set in at least one CDR (antigen binding critical region) of the α chain and / or β chain, as well as optional frame regions, linker peptides or transmembrane regions.

[0029] By pre-setting mutation domains and limiting their locations, the drawbacks of traditional manual design can be avoided from the outset. Specifically, focusing the mutation domain on at least one CDR on the α / β chain allows for precise optimization of the core binding site between the receptor and antigen, significantly reducing the search space, improving efficiency and quality, and ensuring enhanced affinity. Incorporating optional framework regions, linker peptides, and transmembrane regions allows for simultaneous adjustment of receptor structural stability and membrane localization capabilities, avoiding conformational problems such as chain entanglement and excessive linker tension. This limitation not only prevents random mutations from damaging the receptor's functional structure but also ensures that mutations revolve around core requirements, laying the foundation for subsequent batch generation of effective candidate mutation combinations and improving screening efficiency and accuracy.

[0030] In S2, structural prediction is performed. First, the reference sequences of the α and β chains of BiChTCR are obtained, as well as the sequence or three-dimensional structural data of the two target antigens that need to be specifically recognized in the design goal of BiChTCR (for example, IL13Rα2 and B7-H3).

[0031] Secondly, based on the mechanism of action of bispecific immune receptors, it is determined that BiChTCR needs to form a complex system that specifically binds to two target antigens. That is, two types of complexes to be predicted are virtually set up: "BiChTCR-target antigen A" and "BiChTCR-target antigen B". The sequence / structure data of BiChTCR is associated and matched with the sequence / structure data of the corresponding target antigens to construct the initial input dataset for each type of complex.

[0032] Finally, the datasets of the two types of complexes were imported into the structure prediction model, and the BiChTCR-antigen complex structure was predicted in parallel. The conformation and confidence scores were output, and the optimal conformation was automatically selected.

[0033] Because the two target antigens differ significantly in molecular structure and amino acid sequence, the binding interface and interaction mechanisms between BiChTCR and different target antigens will vary when constructing BiChTCR-target antigen A and BiChTCR-target antigen B complexes, resulting in a variety of possible conformations. When predicting these two types of complexes, the structure prediction model generates corresponding conformation sets and confidence scores based on their respective binding characteristics, and then selects the optimal conformation for each.

[0034] The structure prediction model can be selected by those skilled in the art based on the actual situation, and can be an existing protein structure prediction tool, such as AlphaFold, AlphaFold2 / 3, ESMFold, or RoseTTAFold. Its input is the amino acid sequence of the bispecific immune receptor and the first and second target antigens. After the model runs, it outputs a predicted three-dimensional conformation file (such as PDB or mmCIF format) and generates a prediction confidence score (such as pLDDT, pTM, or ipTM values) to characterize the reliability of the predicted structure.

[0035] In this embodiment, the bispecific immune receptor is input into the model in parallel with the first and second target antigens to obtain the predicted conformations of the two complexes. These are then screened based on the confidence scores, with the optimal conformation satisfying a preset threshold condition serving as the structural basis for subsequent mutation generation and multi-target screening. For example, the preset threshold condition is set to pLDDT ≥ 70 or pTM ≥ 0.7.

[0036] In S3, for BiChTCR in the optimal conformation, candidate mutation combinations are generated in batches (≥50 per batch) based on a preset mutation domain using a sequence generation model, and invalid sequences are filtered out according to rule constraints to obtain candidate mutants.

[0037] Specifically, the predetermined mutation domain position is determined in the obtained optimal three-dimensional conformation, such as a variable fragment located in the complementarity determination region (CDR) or frame region of the α or β chain. The amino acid sequence corresponding to the optimal three-dimensional conformation is then provided as input to the sequence generation model. This model can be a sequence generation tool based on a protein language model, such as ProtGPT2, ESM2, or other variational autoencoders, generative adversarial networks, etc. Its principle is to learn the statistical laws of amino acid sequences through large-scale training data and generate multiple possible mutation combinations according to probability distribution within the defined mutation domain.

[0038] The generation process can incorporate various constraints, such as amino acid physicochemical property preferences, secondary structure preservation requirements, or solvent accessibility limitations, and the generation diversity can be controlled through methods such as temperature sampling or Top-k sampling, thereby obtaining a series of candidate mutant sequences. It should be understood that the constraints can be set by those skilled in the art according to the actual situation.

[0039] The final output is a list of amino acid sequences containing multiple candidate mutation combinations, typically stored in FASTA format, and used as input for subsequent energy calculations and multi-target screening.

[0040] In this embodiment, by efficiently generating diverse mutation combinations that conform to probability distribution within a defined mutation domain, and by combining constraints such as amino acid physicochemical properties to filter out invalid sequences, it is possible to provide rich and valuable candidate mutants for subsequent energy calculations and multi-target screening, thereby improving the efficiency and success rate of bispecific immune receptor optimization.

[0041] In S4, automated screening is performed through energy and interface analysis.

[0042] A multi-objective scoring function is constructed to perform two rounds of screening for candidate objects. The first round uses multi-objective scoring based on stability and interface indicators such as ΔΔG / BSA, while the second round introduces docking energy and interface annotation to update the scoring function F, ensuring the balance between the two targets.

[0043] The multi-objective scoring function includes indicators such as the free energy difference before and after mutation, the buried surface area of ​​the binding interface, molecular docking energy, the number of interfacial contact residues, and the number of interfacial hydrogen bonds.

[0044] In the first round of screening, when ΔΔG and BSA are used as indicators, for example, it is preferred that ΔΔG ≤ -0.5~-1.5 kcal / mol and BSA improves by ≥3%~10% relative to the baseline to select the first candidate set.

[0045] It should be understood that the steps for calculating ΔΔG / BSA are feasible for those skilled in the art.

[0046] In the second round of screening, the first candidate set and the dual-target antigen were docked in parallel using a docking simulation model to obtain docking energy and interface annotations (including the number of interface contact residues, the number of interface hydrogen bonds, etc.). The docking simulation model can be ClusPro, HADDOCK, or ZDOCK, and calculating the docking energy and interface annotations is something that those skilled in the art can do.

[0047] Based on the obtained docking energy and interface annotations, the scoring function F for the second round is updated to include docking energy, interface-related indicators, etc. The objects in the first candidate set are re-scored and screened to obtain the second candidate set, which contains the optimized bispecific immune receptor sequence. At the same time, the corresponding indicator report is generated to complete the automated screening.

[0048] Outputs the optimal mutant combination and standardized results package, presented as a structural superposition diagram and index report, and exported in machine-readable formats such as FASTA, PDB / mmCIF, CSV, JSON, and PDF, supporting integration with LIMS.

[0049] For example, under a given computing environment, the average time for a single task can be controlled within 2 hours (reported mean ± SD), and it supports the parallel execution of ≥10 tasks (example, not a limitation), and generates data files that can be directly used in the LIMS system.

[0050] This embodiment proposes a two-round screening process. The first round focuses on stability and interface indicators, selecting a first candidate set that meets the basic performance requirements from a batch of candidate mutation combinations. This quickly eliminates invalid mutations with unstable structures and poor interface binding ability, significantly narrowing the screening range, reducing subsequent computational costs, and avoiding false positives caused by traditional single-indicator screening. The second round introduces parallel docking simulation, performing docking analysis on the first candidate set with two target antigens respectively. It supplements the calculation of docking energy and interface annotation and updates the scoring function. This not only further verifies the binding compatibility between candidate mutants and dual-target antigens, but also ensures balanced binding affinity between the two targets through multi-target scoring, effectively reducing the risk of immune escape caused by single-target loss.

[0051] In this embodiment, firstly, by pre-setting core mutation domains such as α / β chain CDRs and framework regions, and combining models such as AlphaFold with parallel conformation prediction and screening of optimal conformations, the optimization range is precisely locked, avoiding manual design defects such as chain entanglement and excessive tension in the linker region from the source, thus ensuring the expression potential on the receptor membrane. Secondly, candidate mutation combinations are generated in batches based on sequence generation models, and a two-round screening is achieved with a multi-objective scoring function. The first round uses indicators such as ΔΔG and BSA to initially screen stable candidates, and the second round uses parallel docking simulation to supplement energy and interface annotation for further screening, covering a large-scale mutation space, replacing inefficient manual adjustments, and significantly shortening the design cycle. Finally, an integrated closed loop is achieved from sequence input to structure prediction, mutation design, energy and interface analysis, visualization and report output, without manual intervention. This opens up a linkage channel between AI prediction and functional verification, and the output results can be directly connected to the experimental system, providing reliable technical support for the efficient development of bispecific immune receptors for solid tumor treatment.

[0052] Example 1 1. Experimental setup Construction: Reference sequence Ref and representative optimized sequences Rep-1, Rep-2, Rep-3, Rep-4.

[0053] Target cell groups: A+ (IL13Rα2⁺ / B7-H3⁻), B+ (IL13Rα2⁻ / B7-H3⁺), A+B+ (double positive), A–B– (double negative control).

[0054] Replication: Biological replication n=3 (donor / batch), per conditional technical replication k=3; statistics are based on the donor mean in the analysis.

[0055] Endpoints: NFAT reporter (GFP / luciferase), in vitro killing (E:T=1:1 and 10:1, 24 / 48h).

[0056] 2. Scope Design 1) Representative sequence and design scope Four representative candidates (Rep-1 / Rep-2 / Rep-3 / Rep-4) were selected, and the mutations were restricted to the confined domain (at least one CDR on the α / β chain, with optional framework / linker peptide / transmembrane region) to verify the effect of "two-round screening and dual-target balance".

[0057] Reference receptor: BiChTCR-Ref (humanized α / β chain).

[0058] Target antigens: B7-H3 (CD276, the dominant human 4Ig subtype): The extracellular region consists of two pairs of IgV-IgC domains and is a popular clinical tumor immune checkpoint target (4Ig is the dominant human isotype; a 2Ig isotype also exists). IL13Rα2 (human UniProt: Q14627): The extracellular region, topologically ranging from amino acids 27 to 343, consists of one S-type Ig domain + two FNIII domains.

[0059] Documentation and Version Control: Antigen ECD Source: B7-H3 (UniProt Q5ZPR3, 4Ig background information), IL13Rα2 (UniProt Q14627); PDB / AlphaFold prediction files for docking are recorded with MD5 checksums.

[0060] 2) Candidate List (Mutations and Basic Information) Table 1. List of representative sequences (IMGT / Kabat numbers)

[0061] Note: The numbering system is consistent with the reference sequence; the length of the linker peptide and the hydrophobic window are recorded as adjustable parameters.

[0062] 3. Calculation of evaluation and results of two rounds of screening Using a multi-objective scoring function F (the index set includes the difference in free energy before and after mutation, the buried surface area of ​​the binding interface, the molecular docking energy, the number of interface contact residues, and the number of interface hydrogen bonds, corresponding to {-ΔΔG, BSA, N_contact, Hbond, -E_dock}, normalized to [0,1]; example weights of 0.30 / 0.30 / 0.15 / 0.10 / 0.15; the equilibrium constraint uses harmonic mean or max-min), a first round of (stability / interface) screening is performed. Then, the shortlisted candidates are docked with two target antigens in parallel, and a second round of screening is performed after reinjecting E_dock / interface annotations.

[0063] Table 2. Calculation Indicators and Equilibrium Score of Representative Sequences

[0064] Compliance criteria (configurable): Example thresholds: τ_ΔΔG≤-0.5; τ_BSA≥+3%; lower docking energy is preferred; if it enters the Top-K and min{Score_A,Score_B}≥τ_score (e.g., 0.65), it is considered to meet the criteria.

[0065] Note: Balanced distribution (harmony): It can suppress the imbalance of "one high and one low"; the minimum score directly reflects the performance of the weakest link.

[0066] 4. Construction and Expression The optimized sequences of Rep-1 / Rep-2 / Rep-3 / Rep-4 were cloned into a unified expression backbone (with consistent signal peptide / transmembrane / intracellular domains), and cell expression and surface presentation were detected under standard conditions.

[0067] Quality control: Sequencing alignment consistency ≥ 99.9%; expression positivity rate (flow cytometry); Output: Standardized result sets in FASTA / PDB / CSV / JSON / PDF format, facilitating LIMS integration and archiving.

[0068] 5. In vitro functional verification (1) Detection of surface expression level Flow cytometry (FACS) analysis using anti-TCRα / β and anti-CD3 antibodies showed that the Rep-2 conformation successfully restored TCR / CD3 complex surface expression, with a positive expression rate of 69.98%, significantly superior to other conformations. Figure 4 As shown.

[0069] (2) Functional screening: NFAT activation under antigen stimulation An NFAT-GFP reporter system was constructed and co-transfected into Jurakat cells. Changes in GFP expression were observed after co-culturing with target cells. The results showed that BiChTCR-2 stimulation significantly enhanced NFAT activation, suggesting that NFAT can trigger downstream signaling pathways after binding to antigens and possesses functional activity. Figure 5 As shown.

[0070] (3) In vitro killing experiment The in vitro killing rates (%) of Ref and four representative candidates (Rep-1 to Rep-4) are shown in four target cell groups: AB- (double negative control), A+ (IL13Rα2⁺), B+ (B7-H3⁺), and A+B+ (double positive). The bars represent the biological replicate mean (n=3, k=3 technical replicates per group), and the error bars represent the standard deviation; grayscale indicates different constructs (see legend). In the example data, Rep-2 showed the highest performance under A+, B+, and A+B+ conditions, while AB- was close to background. Figure 6 As shown.

[0071] (4) Success rate and labor reduction rate Under the conditions of a representative environment and a sample size of n=20: Success rate (criteria: entering Top-K and ΔΔG / BSA / E_dock meeting the criteria) expected 75% (15 / 20), Wilson 95% CI: 57%–88%.

[0072] Labor reduction rate (by task pairing) The expected mean is 60%, with a 95% CI of 52%–68%; the median is approximately 61%. Figure 7 As shown.

[0073] The bars represent the success rate (%), defined as the percentage of candidates that meet the preset criteria (entering the Top-K and achieving ΔΔG / BSA / E_dock); the error bar is the Wilson 95% confidence interval. Example: k=15, n=20.

[0074] Statistical method: The 95% CI uses Wilson interval (two-sided, significance level 0.05). For small samples or extreme proportions, Clopper–Pearson interval can be added.

[0075] The column represents the average (%) of the reduction rate in manual workload, such as... Figure 8 As shown. The error bars represent the 95% confidence interval; the upper label is "mean [lower limit, upper limit] (n=sample size)". The reduction rate is defined as... .

[0076] The reduction rate distribution for each task, such as Figure 9As shown: The box plot displays the median, interquartile range, and mean; the overlaid jitter points represent the observations (%) for each individual task. This is used to illustrate dispersion and outliers, and is corroborated by the mean ± CI in Figure Xa. Statistical methods: Summary and Intervals: Reports the mean, standard deviation (SD), and median; the 95% confidence interval for the mean is a t-interval. .

[0077] In the B7-H3 / IL13Rα2 dual-target task, Rep-2 achieved the highest balanced score (0.74) after the second round (parallel docking and reinjection), followed by Rep-1; Rep-3 was used as an alternative. This embodiment illustrates that the present invention can shorten the screening cycle and output candidates that can be directly used for construction and validation while maintaining the balance between the two targets.

[0078] Example 2 This embodiment provides an automated screening and optimization system for bispecific immune receptors, including: An initialization unit is used to acquire the sequence of the bispecific immune receptor to be screened, and the sequences of the first and second target antigens, and to preset the mutation domain of the bispecific immune receptor. The conformation prediction unit is used to perform parallel conformation prediction of the bispecific immune receptor sequence and two target antigens based on the structure prediction model, and to select the optimal conformation based on the confidence level. The mutation unit is used to generate candidate mutation combinations within the mutation domain based on a preset mutation domain of the bispecific immune receptor in the optimal conformation, using a sequence generation model. After filtering out invalid sequences, multiple bispecific immune receptors carrying different mutation combinations are obtained as candidate objects. The screening unit is used to construct a multi-objective scoring function and perform two rounds of screening for candidate objects. Specifically, the first round screens the first candidate set based on stability and interface indicators. The second round performs parallel docking simulations on the first candidate set and two target antigens, calculates docking energy and interface residue annotations, updates the scoring function based on the calculation results, and obtains the optimized bispecific immune receptor sequence and indicator report of the second candidate set, thus completing the automated screening.

[0079] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the automated screening and optimization method for bispecific immune receptors as described in Embodiment 1 above.

[0080] Example 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the automated screening and optimization method for bispecific immune receptors as described in Embodiment 1 above.

[0081] The steps or modules involved in Embodiments 2 to 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0082] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An automated screening and optimization method for bispecific immune receptors, characterized in that, include: Obtain the sequence of the bispecific immune receptor to be screened, and the sequences of the first and second target antigens, and predetermine the mutation domain of the bispecific immune receptor; The bispecific immune receptor sequence and two target antigens were predicted in parallel using a structural prediction model, and the optimal conformation was selected based on confidence level. Based on the preset mutation domain of the bispecific immune receptor in the optimal conformation, candidate mutation combinations are generated in the mutation domain by a sequence generation model. After filtering out invalid sequences, multiple bispecific immune receptors carrying different mutation combinations are obtained as candidate objects. A multi-objective scoring function is constructed to screen candidates in two rounds. Specifically, the first round screens the first candidate set based on stability and interface indicators. The second round performs parallel docking simulations on the first candidate set and two target antigens, calculates docking energy and interface residue annotations, updates the scoring function based on the calculation results, and obtains the optimized bispecific immune receptor sequence and indicator report for the second candidate set, thus completing the automated screening.

2. The automated screening and optimization method for bispecific immune receptors as described in claim 1, characterized in that, The mutation domain of the predefined bispecific immune receptor includes at least one CDR, framework region, linker peptide, or transmembrane region of the α / β chain.

3. The automated screening and optimization method for bispecific immune receptors as described in claim 1, characterized in that, The method of performing parallel conformation prediction of the bispecific immune receptor sequence and two target antigens based on the structure prediction model, and selecting the optimal conformation based on the confidence score, specifically includes: inputting the combination of the bispecific immune receptor sequence and the first and second target antigen sequences into the structure prediction model to generate the predicted conformations and confidence scores of the two complexes; and selecting the optimal predicted conformation based on the confidence scores.

4. The automated screening and optimization method for bispecific immune receptors as described in claim 1, characterized in that, The multi-objective scoring function includes indicators such as the free energy difference before and after mutation, the buried surface area of ​​the binding interface, molecular docking energy, the number of interfacial contact residues, and the number of interfacial hydrogen bonds.

5. The automated screening and optimization method for bispecific immune receptors as described in claim 1, characterized in that, The docking simulation includes two stages: global search and local refinement, specifically including: In the global search phase, based on the predicted three-dimensional conformations of the bispecific immune receptor and two target antigens, the rigid body docking method is used to sample the multiple binding orientations of the two in space. Based on the scoring criteria of shape complementarity, hydrophobicity or electrostatic energy, all docking conformations are initially screened to obtain multiple candidate complex structures that meet the preset potential binding probability threshold. In the local refinement stage, the candidate complex structure is adjusted at the atomic level using molecular force field optimization or short-range molecular dynamics simulation to eliminate spatial conflicts and enhance interface stability, thereby obtaining the refined docking complex structure and its energy score.

6. An automated screening and optimization system for bispecific immune receptors, characterized in that, include: An initialization unit is used to acquire the sequence of the bispecific immune receptor to be screened, and the sequences of the first and second target antigens, and to preset the mutation domain of the bispecific immune receptor. The conformation prediction unit is used to perform parallel conformation prediction of the bispecific immune receptor sequence and two target antigens based on the structure prediction model, and to select the optimal conformation based on the confidence level. The mutation unit is used to generate candidate mutation combinations within the mutation domain based on a preset mutation domain of the bispecific immune receptor in the optimal conformation, using a sequence generation model. After filtering out invalid sequences, multiple bispecific immune receptors carrying different mutation combinations are obtained as candidate objects. The screening unit is used to construct a multi-objective scoring function and perform two rounds of screening for candidate objects. Specifically, the first round screens the first candidate set based on stability and interface indicators. The second round performs parallel docking simulations on the first candidate set and two target antigens, calculates docking energy and interface residue annotations, updates the scoring function based on the calculation results, and obtains the optimized bispecific immune receptor sequence and indicator report of the second candidate set, thus completing the automated screening.

7. The automated screening and optimization system for bispecific immune receptors as described in claim 6, characterized in that, The mutation domain of the predefined bispecific immune receptor includes at least one CDR, framework region, linker peptide, or transmembrane region of the α / β chain.

8. The automated screening and optimization system for bispecific immune receptors as described in claim 6, characterized in that, The method of performing parallel conformation prediction of the bispecific immune receptor sequence and two target antigens based on the structure prediction model, and selecting the optimal conformation based on the confidence score, specifically includes: inputting the combination of the bispecific immune receptor sequence and the first and second target antigen sequences into the structure prediction model to generate the predicted conformations and confidence scores of the two complexes; and selecting the optimal predicted conformation based on the confidence scores.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the automated screening and optimization method for bispecific immune receptors as described in any one of claims 1-5.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the automated screening and optimization method for bispecific immune receptors as described in any one of claims 1-5.

Images