A method and system for batch design and evaluation of highly efficient antimicrobial peptides

Abstract

The invention discloses a method and system for batch design and evaluation of efficient antimicrobial peptides, belonging to the technical field of bioinformatics, and comprising: a model training module, which collects antimicrobial peptide related data and constructs an antimicrobial peptide vector database, a conventional corpus and a model training data set, and uses the data set to train the antimicrobial peptide model; a model deployment and application module, which calls the antimicrobial peptide model to sequentially perform sequence generation, structure simulation, and simulation evaluation tasks, batch screens out candidate antimicrobial peptide sequences, performs antimicrobial peptide detection and evaluation on the candidate antimicrobial peptide sequences, updates the antimicrobial peptide vector database according to the evaluation results, and constructs a test vector database for enhanced fine-tuning of the antimicrobial peptide model; an artificial intelligence agent, which dynamically calls the model antimicrobial peptide model according to demand to realize task scheduling and batch process the database, thereby realizing efficient batch design, screening, evaluation and testing of antimicrobial peptides, improving the screening accuracy of antimicrobial peptides, and greatly shortening the batch research and development cycle of antimicrobial peptides.

Description

Technical Field

[0001] The present invention belongs to the technical field of combining bioinformatics, artificial intelligence and antimicrobial peptide research and development, and specifically relates to a method and system for batch design and evaluation of efficient antimicrobial peptides. Background Art

[0002] Studies have shown that the growing problem of antibiotic resistance is a major clinical concern. This antibiotic resistance complicates treatment options, so alternative treatment strategies such as antimicrobial peptides (AMPs) must be explored.

[0003] Antimicrobial peptides are a class of small molecule peptides that are widely present in nature and play an important role in innate immunity in different life forms. They are an important component of the host defense system and play a key role in resisting infections from microorganisms such as bacteria, fungi and viruses. They have a broad-spectrum bactericidal effect and have broad development prospects as new antimicrobial agents. Based on their strong inhibitory or killing ability against bacteria, fungi, parasites, certain viruses and even tumor cells, antimicrobial peptides are no longer limited to the narrow definition of antibacterial peptides, but are more regarded as a class of multifunctional polypeptide antibiotics. Antimicrobial peptides have become a popular candidate for new antimicrobial drugs due to their broad-spectrum antimicrobial properties, low toxicity and low risk of drug resistance.

[0004] The design and optimization of traditional antimicrobial peptides rely on a large number of experimental screenings, which are long and costly. With the development of artificial intelligence, the use of AI technologies such as machine learning and deep learning can accelerate the screening and optimization of antimicrobial peptides, but these methods are mostly limited to a single link. At present, there is still a lack of a relatively complete artificial intelligence-driven antimicrobial peptide design, screening, evaluation and testing system. Therefore, it is of great significance to design an efficient antimicrobial peptide batch design and evaluation method and system to improve R&D efficiency and screening accuracy. Summary of the invention

[0005] In view of the above, the design and optimization of traditional antimicrobial peptides rely on a large number of experimental screenings, long cycles and high costs. The purpose of the present invention is to provide an efficient antimicrobial peptide batch design and evaluation method and system, which enables antimicrobial peptide research and development through antimicrobial peptide models, and uses antimicrobial peptide detection and evaluation to feed back and optimize the antimicrobial peptide model. In this way, through modular design and closed-loop optimization, efficient batch design, screening, evaluation and testing of antimicrobial peptides can be achieved, the screening accuracy of antimicrobial peptides can be improved, and the batch research and development cycle of antimicrobial peptides can be greatly shortened.

[0006] To achieve the above-mentioned purpose of the invention, an embodiment provides an efficient antimicrobial peptide batch design and evaluation system, comprising an artificial intelligence agent, a model training module, and a model deployment and application module, wherein the model training module is used to collect and preprocess antimicrobial peptide related data, construct an antimicrobial peptide vector database and a conventional corpus, and construct a model training data set, and use the data set to train an antimicrobial peptide model;

[0007] The model deployment and application module is used to call the antimicrobial peptide model to perform sequence generation, structure simulation, and simulation evaluation tasks in sequence based on the needs of the artificial intelligence agent, batch screen out candidate antimicrobial peptide sequences that simultaneously meet the antimicrobial activity standards, meet specific structural characteristics, and meet the antimicrobial peptide inline characteristics, and then perform antimicrobial peptide detection and evaluation on the candidate antimicrobial peptide sequences, update the antimicrobial peptide vector database and construct a test vector database based on the evaluation results, and the updated antimicrobial peptide vector database and the test vector database are used for enhanced fine-tuning of the antimicrobial peptide model;

[0008] The artificial intelligence agent serves as a bridge for researchers to interact with the system. It dynamically calls the antimicrobial peptide model to implement task scheduling according to demand, and also dynamically performs batch processing on the database according to demand.

[0009] Preferably, preprocessing of antimicrobial peptide related data includes cleaning to remove missing values, outliers and duplicate values, annotating antimicrobial peptide sequences, physicochemical properties, multi-level structures, and key functional attributes, wherein the key functional attributes include antimicrobial activity and toxicity, and vectorizing the annotated data.

[0010] Preferably, the antimicrobial peptide model comprises a sequence generation model, a structure simulation model, and a simulation evaluation model, wherein the sequence generation model is used to screen candidate antimicrobial peptide sequences that meet the antimicrobial activity standards, the structure simulation model is used to further screen the candidate antimicrobial peptide sequences output by the sequence generation model to screen candidate antimicrobial peptide sequences that meet specific structural characteristics, and the simulation evaluation model is used to further screen the candidate antimicrobial peptide sequences output by the structure simulation model to screen candidate antimicrobial peptide sequences that meet the antimicrobial peptide internal characteristics.

[0011] Preferably, the sequence generation model adopts a Transformer-based large language model.

[0012] Preferably, the structure simulation model can adopt a combination based on Transformer and Stable Diffusion, wherein Transformer is used to embed sequence structure information as guiding information, and Stable Diffusion is used for structure simulation generation.

[0013] Preferably, the simulation evaluation model uses a Transformer-based hybrid expert model for model training.

[0014] Preferably, the antimicrobial activity standard includes basic sequence information, including antimicrobial peptide sequence, physicochemical properties, including amino acid sequence length, net charge, arithmetic mean hydrophobicity, average molecular weight, and theoretical isoelectric point;

[0015] Specific structural characteristics include sequence structure information, which includes secondary structure and tertiary structure information of antimicrobial peptides, as well as structural stability under simulation conditions;

[0016] The internal characteristics of antimicrobial peptides include the core functional evaluation indicators of antimicrobial peptides, which include minimum inhibitory concentration and toxicity, among which toxicity includes cytotoxicity and hemolysis.

[0017] Preferably, when antimicrobial peptide detection and evaluation is performed on candidate antimicrobial peptide sequences, the antimicrobial activity, cytotoxicity and hemolytic activity of the antimicrobial peptides are tested in directed experiments, and antimicrobial peptides with high antimicrobial activity, low cytotoxicity and negative hemolytic activity are screened out based on the results and added to the antimicrobial peptide vector database, while other antimicrobial peptides are recorded in the test vector database.

[0018] To achieve the above-mentioned purpose, the present invention also provides a method for batch design and evaluation of highly effective antimicrobial peptides, comprising the following steps:

[0019] S1, after collecting and preprocessing the antimicrobial peptide related data, construct the antimicrobial peptide vector database and conventional corpus and build the model training data set;

[0020] S2, using the dataset to train the antimicrobial peptide model;

[0021] S3, using the call to the antimicrobial peptide model to perform sequence generation, structure simulation, and simulation evaluation tasks in sequence, completes antimicrobial peptide sequence generation and primary screening, antimicrobial peptide structure simulation and secondary screening, antimicrobial activity and toxicity simulation evaluation;

[0022] S4, conduct experimental evaluation on the results of antimicrobial activity and toxicity simulation evaluation, and then screen antimicrobial peptides in batches. The database is also updated based on the experimental evaluation results to provide feedback and optimize the antimicrobial peptide model.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] 1. Data and model driven: With antimicrobial peptide related data and antimicrobial peptide models as the center, sequence generation, structure simulation and evaluation models work together to achieve automated design and efficient phased screening of antimicrobial peptides;

[0025] 2. Batch parallel testing: With the help of model prediction results, targeted batch tests are carried out to test candidate antimicrobial peptides in parallel, which greatly improves the efficiency of antimicrobial peptide screening and reduces test costs;

[0026] 3. Feedback closed-loop optimization: By feeding back the test verification data to the system, a closed-loop mechanism of continuous optimization is formed, which gradually improves the accuracy and effectiveness of antimicrobial peptide screening and evaluation, providing an efficient and low-cost innovative solution for the research and development of antimicrobial peptides. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0028] Figure 1 It is a schematic diagram of the structure and process of the high-efficiency antimicrobial peptide batch design and evaluation system provided in the embodiment;

[0029] Figure 2 This is a schematic flow chart of a method for batch design and evaluation of highly efficient antimicrobial peptides provided in an embodiment;

[0030] Figure 3 It is a schematic diagram of the structure of the antimicrobial peptide “GIGAVLKVLTTGLPALISWIKRKRQQ” provided in the examples. DETAILED DESCRIPTION

[0031] To make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation methods described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.

[0032] like Figure 1 As shown, the embodiment provides a high-efficiency antimicrobial peptide batch design and evaluation system, which includes an artificial intelligence agent, a model training module, and a model deployment and application module.

[0033] In the embodiment, the artificial intelligence agent is an intelligent body that can autonomously understand, plan and make decisions, and perform complex tasks. It serves as a bridge for researchers to interact with the system of the present invention. It dynamically calls the model according to the needs to realize task scheduling, and also dynamically processes the database in batches according to the needs. Specifically, the existing antimicrobial peptide vector database and the test vector database are added, deleted, modified and checked, and different models are called as needed to perform different prediction tasks and the results are fed back to the researchers. For example, for problems such as sequence generation, antimicrobial peptide characteristics or antimicrobial peptide structure, the artificial intelligence agent can call ChatGPT, AlphaFold and other existing large model applications as needed to save costs. And periodically start model fine-tuning or retraining to optimize the corresponding model, etc.

[0034] In the embodiment, the model training module performs relevant processing on the data and trains an antimicrobial peptide model that meets relevant requirements. Specifically, after collecting and preprocessing antimicrobial peptide related data, an antimicrobial peptide vector database and a conventional corpus are constructed, and a model training data set is constructed, and the antimicrobial peptide model is trained using the data set.

[0035] When collecting antimicrobial peptide related data, antimicrobial peptide related data can be collected from public databases and experimental data. Public related data can be from databases such as Wikipedia, books, journal articles, etc., and experimental data mainly comes from the data of subsequent scientific researchers evaluating and testing the generated candidate antimicrobial peptide sequences. Then the antimicrobial peptide related data collected from outside are preprocessed, including cleaning, removing missing values, outliers and duplicate values ​​to ensure the integrity and consistency of the data; using models or manual methods to annotate antimicrobial peptide sequences, physicochemical properties, multi-level structures, antimicrobial activity and toxicity and other key functional attributes, and vectorizing the annotated data to construct a multidimensional antimicrobial peptide vector database of sequence, physicochemical properties, structure and key functional attributes. At the same time, a conventional corpus is constructed to prepare for the generation of high-quality model training data sets to ensure that the input data of the model is reliable.

[0036] In the embodiment, a key-value storage structure is used to manage antimicrobial peptide related databases, including antimicrobial peptide vector databases, subsequent test vector databases, etc., so as to provide efficient data support for subsequent model training, simulation evaluation and screening. The core reason for using a vector database is that the vector database is more suitable for model training and can assist in retrieval and enhancement generation. The data in the antimicrobial peptide vector database is stored and managed in a key-value manner, and the basic structure format of the data is: {key; value}. Among them, the key is the unique identification sequence of the antimicrobial peptide, which can be the amino acid sequence of the antimicrobial peptide. The value contains multidimensional feature information of the antimicrobial peptide, mainly including three types of information {sequence basic information, sequence structure information, inline characteristics}, which are all stored in the form of vectors, wherein the vectorization method of text data can use the bag of words model (Bag of Words), TF-IDF, word vector (Word2Vec), etc., the vectorization method of image data can use SIFT (scale invariant feature transform), SURF (accelerated robust features), etc., and the vectorization method of audio data can use MFCC (Mel frequency cepstral coefficients), Chroma features, etc.

[0037] Among them, the basic sequence information includes various information of the antimicrobial peptide. Taking the antimicrobial peptide "GIGAVLKVLTTGLPALISWIKRKRQQ" as an example, its antimicrobial peptide sequence is GIGAVLKVLTTGLPALISWIKRKRQQ, amino acid sequence length 26, net charge 5.97, hydrophobicity arithmetic mean 0.27, average molecular weight 2847.43 g / mol, theoretical isoelectric point pH 12.43 and other basic physical and chemical properties, and the structure is as follows: Figure 3 Such information provides a preliminary understanding of the physicochemical properties of antimicrobial peptides and helps generate antimicrobial peptides that meet expectations.

[0038] Among them, sequence structure information includes the secondary structure (such as α helix, β fold, etc.) and tertiary structure information of antimicrobial peptides, as well as structural stability under simulation conditions. Structural information can be used to predict the targeting ability and stability of antimicrobial peptides under physiological conditions. It helps to guide the generation of antimicrobial peptides that meet relevant constraints. It should be noted that the relevant constraints are determined by the structural simulation model according to the existing antimicrobial peptide structural information.

[0039] The inline characteristics include the core functional evaluation indicators of antimicrobial peptides (such as minimum inhibitory concentration (MIC), cytotoxicity, and hemolysis). The specific definitions are as follows:

[0040] Minimum inhibitory concentration (MIC): The minimum effective concentration that inhibits the growth of target bacteria, usually expressed in μg / mL. The lower the MIC value, the higher the antibacterial activity. Cytotoxicity: The toxicity of antimicrobial peptides to host cells, usually expressed by the half-inhibitory concentration (IC50) to normal cells (such as HEK293, HepG2). The higher the IC50 value, the lower the toxicity. Hemolysis: The hemolytic effect of antimicrobial peptides on red blood cells. The lower the hemolysis, the better the biocompatibility.

[0041] There is a certain internal correlation between the minimum inhibitory concentration, cytotoxicity and hemolysis. In order to ensure that the designed antimicrobial peptide meets the expected antibacterial and safety standards, the internal characteristic correlation can be described by the following formula, where MIC represents the minimum inhibitory concentration, CT represents cytotoxicity (usually expressed as IC50 value), and H represents hemolysis. The internal correlation can be expressed as: {high antibacterial activity, low cytotoxicity, negative hemolysis}, and its formula can be expressed as:

[0042]

[0043] in, is the maximum allowable minimum inhibitory concentration value; is the maximum permissible cytotoxicity threshold; H represents hemolysis, where 0 represents negative and 1 represents positive, and the symbol The relationship ensures that the generated antimicrobial peptide has the expected antibacterial effect and that the toxicity and hemolysis to host cells are within a safe range. These inline characteristics are the core indicators for evaluating the functionality and safety of antimicrobial peptides. In the simulation evaluation model, inline characteristic indicators can be used as constraints. When the model outputs the predicted value of the antimicrobial peptide, it ensures the following points:

[0044] 1. Minimum inhibitory concentration (MIC) range: The MIC value of the antimicrobial peptide predicted by the evaluation model must be within the normal activity range of the antimicrobial peptide to ensure the antibacterial effect. (For example, the MIC of the antimicrobial peptide "GIGAVLKVLTTGLPALISWIKRKRQQ" on Staphylococcus aureus ATCC25923 strain is 2 μg / mL)

[0045] 2. Cytotoxicity and hemolysis: The simulation evaluation model excludes antimicrobial peptides that may produce high toxicity or positive hemolysis to host cells through the prediction of cytotoxicity and hemolysis to ensure that they meet the biocompatibility requirements. (For example, the IC50 of the antimicrobial peptide "Cytotoxicity and Hemolysis: The Simulation Evaluation Model predicts cytotoxicity and hemolysis. The IC50 of the antimicrobial peptide on human malignant melanoma A375 cells is 3.38 μ. / mL, and it should show negative hemolysis within the dosage range used).

[0046] After obtaining the antimicrobial peptide vector database and conventional corpus, the antimicrobial peptide database-related data are segmented as needed according to the requirements of different model training corpora, and the segmented data and the conventional corpus are combined with different data ratios to construct different model training data sets, including sequence generation corpus, structure simulation corpus, and simulation evaluation corpus. The core basis for segmentation is the subordinate type of relevant data information. For example, text introducing the sequence characteristics of antimicrobial peptides can be used to construct a sequence generation corpus, content or pictures related to the structure of antimicrobial peptides can be used to construct a structure simulation corpus, and content related to the core functional evaluation indicators of antimicrobial peptides (such as minimum inhibitory concentration (MIC), cytotoxicity, hemolysis) can be used to construct a simulation evaluation corpus. Different corpora are combined with existing large model technology to complete model training of different functions contained in different antimicrobial peptide models.

[0047] In an embodiment, according to different application requirements, the antimicrobial peptide model includes a sequence generation model, a structure simulation model, and a simulation evaluation model, wherein the sequence generation model is used to screen candidate antimicrobial peptide sequences that meet the antimicrobial activity standards, the structure simulation model is used to further screen the candidate antimicrobial peptide sequences output by the sequence generation model, and screen the candidate antimicrobial peptide sequences that meet specific structural characteristics, and the simulation evaluation model is used to further screen the candidate antimicrobial peptide sequences output by the structure simulation model, and screen the candidate antimicrobial peptide sequences that meet the antimicrobial peptide internal characteristics.

[0048] The above three models are trained using the sequence generation corpus, structure simulation corpus, and simulation evaluation corpus. Specifically, the sequence generation model can be trained using a large language model based on Transformer, and the structure simulation model can be trained using a combination of Transformer and Stable Diffusion, where Transformer is used to embed sequence structure information as guidance information and Stable Diffusion is used for structure simulation generation. The simulation evaluation model can be trained using a hybrid expert model based on Transformer.

[0049] In the embodiment, the model deployment and application module applies the trained antimicrobial peptide model to the research and development process of antimicrobial peptides, and organically combines the various models contained in the antimicrobial peptide model with artificial experiments to batch develop antimicrobial peptides with targeted and efficient antimicrobial activity. Specifically, based on the needs of the artificial intelligence agent, the antimicrobial peptide model is called to perform sequence generation, structure simulation, and simulation evaluation tasks in sequence, batch screen out candidate antimicrobial peptide sequences that meet the antimicrobial activity standards, meet specific structural characteristics, and meet the antimicrobial peptide inline characteristics, and then perform antimicrobial peptide detection and evaluation on the candidate antimicrobial peptide sequences, update the antimicrobial peptide vector database and construct the test vector database based on the evaluation results, and the updated antimicrobial peptide vector database and the test vector database are used for enhanced fine-tuning of the antimicrobial peptide model.

[0050] Specifically, on the basis of completing the training of the antimicrobial peptide model, in the process of calling the antimicrobial peptide model for sequence generation and screening, the antimicrobial peptide sequence is first generated and preliminarily screened. Specifically, the intelligent agent calls the sequence generation model to perform the prediction task, and the input prompt words include: the physicochemical properties, structure and function-related information of the target antimicrobial peptide, such as physicochemical property data (such as charge distribution, hydrophobicity, amino acid composition and sequence length, usually set between 10-30 amino acids), known antimicrobial peptide structure and function data (such as effectiveness, toxicity, hemolysis, etc.), and environmental conditions (such as temperature, pH value, salt concentration, etc.). Based on these inputs, candidate antimicrobial peptide sequences A that meet the antimicrobial activity standards are generated in batches (such as the final goal is 100 antimicrobial peptides that meet the requirements), and preliminary screening is carried out based on the specific requirements of the target pathogen (such as targeting Mycoplasma pneumoniae) combined with physicochemical properties. During the first screening, the generated antimicrobial peptide sequence automatically meets the physicochemical property requirements.

[0051] Then, antimicrobial peptide structure simulation and secondary screening are carried out. Specifically, the candidate antimicrobial peptide sequences A (100) obtained from the preliminary screening are input into the structure simulation model in batches to predict their secondary and tertiary structures, and the stability of the antimicrobial peptides under physiological conditions is simulated. The candidate antimicrobial peptide sequences B with stable structures and strong targeting that meet specific structural characteristics are screened out again (the second screening is to screen out candidate antimicrobial peptide sequences with stable structures and strong targeting). If the number of candidate antimicrobial peptide sequences after this screening is less than 100, the artificial intelligence agent will restart the sequence generation model and the structure simulation model to make up the number until the number of candidate antimicrobial peptide sequences B output is 100.

[0052] Finally, the activity and toxicity simulation evaluation is carried out. Specifically, the candidate antimicrobial peptide sequences B (100) obtained in the second screening are batch-inputted into the simulation evaluation model to quickly predict their antimicrobial activity (such as MIC value), toxicity (such as cytotoxicity, hemolysis), test values, etc., to screen out highly toxic and positive hemolytic antimicrobial peptides, and batch-screen out candidate antimicrobial peptide sequences C that meet the antimicrobial peptide inline characteristics (the third screening, screening out candidate antimicrobial peptide sequences that meet the antimicrobial peptide inline characteristics). Similarly, if the number of candidate antimicrobial peptide sequences after this evaluation and screening is less than 100, the artificial intelligence agent will restart the sequence generation model, structure simulation model and simulation evaluation model in order to make up the number until the number of candidate antimicrobial peptide sequences C output is 100.

[0053] The embodiment also performs antimicrobial peptide detection and evaluation. Specifically, the researchers conduct relevant test tests on candidate antimicrobial peptide sequences C (100) in batches according to the prediction results of the simulation evaluation model. According to the test values ​​recommended by the model, the screening range of the test verification is reduced, and the candidate antimicrobial peptide sequences C are subjected to targeted tests for antimicrobial activity (such as MIC value) and toxicity (such as cytotoxicity, hemolysis) in batches. According to the actual test results, antimicrobial peptides D with high antimicrobial activity, low cytotoxicity and negative hemolysis are screened out (the fourth test screening screens out antimicrobial peptides with high antimicrobial activity, low cytotoxicity and negative hemolysis). The number of antimicrobial peptides D is very likely to be less than 100. Similarly, the artificial intelligence agent will call the relevant models in sequence, continuously generate candidate sequences, until manual intervention or preset goals are completed, and the relevant data of the antimicrobial peptide D that meets the requirements will be vectorized and recorded and updated to the antimicrobial peptide vector database, and the relevant data of the candidate antimicrobial peptide E that does not meet the requirements will be vectorized and recorded in the test vector database.

[0054] The antimicrobial activity (such as MIC value) and toxicity (such as cytotoxicity, hemolysis) directional tests are as follows:

[0055] For the MIC value X μg / mL estimated by the simulation evaluation model, the dose range was set to 0.125X, 0.25X, 0.5X, X, 2X, 4X, 8X μg / mL, and a validation test was conducted. 0.1 mL of diluted bacterial solution was added to wells 2-7 of the first row of the 96-well plate, 0.2 mL of drug dilution with an initial concentration of 8X μg / mL was added to the first well, and then 0.1 ml of drug dilution with an initial concentration of 8X μg / mL was added to the second well, mixed and added to the seventh well in turn, 0.2 mL of diluted bacterial solution was added to the eighth well as a positive control, and 0.2 mL of liquid culture medium was added to the ninth well as a blank control. Three parallels were set up respectively. The culture was statically cultured at 37°C for 20 hours, and the bacterial growth was scanned and recorded using a multifunctional microplate reader to calculate the final MIC value of the drug.

[0056] Based on the cytotoxicity value Y μg / mL and hemolysis estimated by the model, the dose range was set to 0.125Y, 0.25Y, 0.5Y, Y, 2Y, 4Y, and 8Y μg / mL, and verification experiments were performed at the same time. On the first day, about 1×10 cells were plated in a 96-well plate. 4 / well, prepare the drug dilution solution on the next day, add it to the 96-well plate at a volume of 100 μL / well, add 0.1 mL of culture medium to the 8th well as a negative control, and set up three parallels. Incubate at 37°C, observe daily, and use MTT to determine cytotoxicity at different time points.

[0057] For the hemolysis value Z estimated by the model, prepare a 4% red blood cell suspension. Take healthy rabbit blood, put it into a conical bottle containing glass beads and shake it for 10 minutes, or stir the blood with a glass rod to remove fibrinogen and make it defibrillated blood. Add about 10 times the amount of 0.9% sodium chloride solution, shake well, centrifuge at 1000-1500 revolutions per minute for 15 minutes, remove the supernatant, and wash the precipitated red blood cells with 0.9% sodium chloride solution for 2-3 times according to the above method until the supernatant is no longer red. The obtained red blood cells are made into a 4% suspension with 0.9% sodium chloride solution for test. Prepare 100 μL / well of sample and positive control solution, and set the sample working solution concentration to 0.125Z, 0.25Z, 0.5Z, Z, 2Z, 4Z, 8Z μg / mL; the positive control (Triton-X-100) content ratio is set to 0.5%, 0.25%, 0.125%, 0.062%, 0.031%, 0.016%, 0.008%. Add 100 μL of red blood cell suspension to each well, and place the 96-well plate in a constant temperature incubator for 1 h. The culture conditions are set to 37°C and 60 rpm. Then observe the hemolysis phenomenon. Observe the hemolysis phenomenon with the naked eye. If the solution in the well is clear red, there is no cell residue at the bottom of the tube or a small amount of red blood cells remain, it indicates that hemolysis has occurred; if all the red blood cells sink and the supernatant is colorless and clear, it indicates that no hemolysis has occurred.

[0058] It should be noted that all the above tests are executed in parallel, and if one feature does not meet the requirements, other related tests are directly stopped to reduce the test cost. After the values ​​obtained from the above tests are compared with the model's estimated values, they are fed back to the model to prepare for subsequent model optimization.

[0059] The relevant data in the updated antimicrobial peptide vector database and the test vector database in the embodiment will be used to fine-tune the relevant antimicrobial peptide model, continuously optimize the model performance, accelerate and ensure the accuracy of subsequent antimicrobial peptide model predictions, and achieve continuous optimization of antimicrobial peptide related models. In the process of antimicrobial peptide detection and evaluation, researchers will feed back the test data to the system, fine-tune the model parameters, and iteratively train the three types of models included in the antimicrobial peptide model as needed, forming a closed loop of "design-screening-testing-optimization".

[0060] By integrating this inline feature with the evaluation model, the screened antimicrobial peptides can better meet the development standards of antimicrobial drugs in terms of activity, safety and stability, thereby providing intelligent support for the subsequent test screening and evaluation process of antimicrobial peptides.

[0061] like Figure 2 As shown, the embodiment also provides a method for batch design and evaluation of highly effective antimicrobial peptides, comprising the following steps:

[0062] S1, after collecting and preprocessing the antimicrobial peptide related data, construct the antimicrobial peptide vector database and conventional corpus and build the model training data set;

[0063] S2, using the dataset to train the antimicrobial peptide model;

[0064] S3, using the call to the antimicrobial peptide model to perform sequence generation, structure simulation, and simulation evaluation tasks in sequence, completes antimicrobial peptide sequence generation and primary screening, antimicrobial peptide structure simulation and secondary screening, antimicrobial activity and toxicity simulation evaluation;

[0065] S4, conduct experimental evaluation on the results of antimicrobial activity and toxicity simulation evaluation, and then screen antimicrobial peptides in batches. The database is also updated based on the experimental evaluation results to provide feedback and optimize the antimicrobial peptide model.

[0066] The above method and system can realize the efficient batch design, screening, evaluation and testing of antimicrobial peptides, improve the screening accuracy of antimicrobial peptides, and significantly shorten the batch research and development cycle of antimicrobial peptides.

[0067] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A high-efficiency antimicrobial peptide batch design and evaluation system, characterized in that: Including artificial intelligence agent, model training module, and model deployment and application module, The model training module is used to collect and preprocess antimicrobial peptide related data, construct an antimicrobial peptide vector database and a conventional corpus, and construct a model training data set, and use the data set to train the antimicrobial peptide model, wherein the preprocessing includes cleaning to remove missing values, outliers and duplicate values, annotating the antimicrobial peptide sequence, physicochemical properties, multi-level structure, and key functional attributes, wherein the key functional attributes include antimicrobial activity and toxicity, and vectorizing the annotated data; The model deployment and application module is used to call the antimicrobial peptide model to perform sequence generation, structure simulation, and simulation evaluation tasks in sequence based on the needs of the artificial intelligence agent, and batch screen out candidate antimicrobial peptide sequences that meet the antimicrobial activity standards, meet specific structural characteristics, and meet the antimicrobial peptide inline characteristics. The antimicrobial peptide model includes a sequence generation model, a structure simulation model, and a simulation evaluation model, wherein the sequence generation model is used to screen candidate antimicrobial peptide sequences that meet the antimicrobial activity standards, the structure simulation model is used to further screen the candidate antimicrobial peptide sequences output by the sequence generation model, and screen the candidate antimicrobial peptide sequences that meet the specific structural characteristics, and the simulation evaluation model is used to further screen the candidate antimicrobial peptide sequences output by the structure simulation model, and screen the candidate antimicrobial peptide sequences that meet the antimicrobial peptide inline characteristics; wherein the sequence generation model adopts a large language model based on Transformer, and the structure simulation model adopts a combination based on Transformer and StableDiffusion, wherein Transformer is used to embed sequence structure information as guidance information, StableDiffusion is used for structure simulation generation, and the simulation evaluation model adopts a hybrid expert model based on Transformer for model training; The antimicrobial activity standard includes basic sequence information, including antimicrobial peptide sequence and physicochemical properties, including amino acid sequence length, net charge, hydrophobicity arithmetic mean, average molecular weight, and theoretical isoelectric point; specific structural characteristics include sequence structure information, including secondary structure and tertiary structure information of antimicrobial peptides, as well as structural stability under simulated conditions; antimicrobial peptide internal characteristics include core functional evaluation indicators of antimicrobial peptides, including minimum inhibitory concentration and toxicity, among which toxicity includes cytotoxicity and hemolysis; Then, the candidate antimicrobial peptide sequences are subjected to antimicrobial peptide detection and evaluation, and the antimicrobial peptide vector database is updated and the test vector database is constructed based on the evaluation results. The updated antimicrobial peptide vector database and the test vector database are used for enhanced fine-tuning of the antimicrobial peptide model. When the candidate antimicrobial peptide sequences are subjected to antimicrobial peptide detection and evaluation, the antimicrobial activity, cytotoxicity and hemolytic activity of the antimicrobial peptides are tested in a directed experiment. According to the results, antimicrobial peptides with high antimicrobial activity, low cytotoxicity and negative hemolytic activity are screened out and added to the antimicrobial peptide vector database, and other antimicrobial peptides are recorded in the test vector database. Researchers interact with the system through the artificial intelligence agent, dynamically call the antimicrobial peptide model according to demand to achieve task scheduling, and also dynamically batch process the database according to demand.

2. A method for batch design and evaluation of highly effective antimicrobial peptides using the system of claim 1, characterized in that: The following steps are involved: S1, after the antimicrobial peptide related data are collected and preprocessed, the antimicrobial peptide vector database and conventional corpus are constructed and the model training data set is constructed. The preprocessing includes cleaning to remove missing values, outliers and duplicate values, annotating the antimicrobial peptide sequence, physicochemical properties, multi-level structure, and key functional attributes, among which the key functional attributes include antimicrobial activity and toxicity, and vectorizing the annotated data; S2, using the data set to train an antimicrobial peptide model, the antimicrobial peptide model includes a sequence generation model, a structure simulation model, and a simulation evaluation model, wherein the sequence generation model is used to screen candidate antimicrobial peptide sequences that meet the antimicrobial activity standard, the structure simulation model is used to further screen the candidate antimicrobial peptide sequences output by the sequence generation model, and screen the candidate antimicrobial peptide sequences that meet specific structural characteristics, and the simulation evaluation model is used to further screen the candidate antimicrobial peptide sequences output by the structure simulation model, and screen the candidate antimicrobial peptide sequences that meet the antimicrobial peptide inline characteristics; S3, researchers interact with the system through the artificial intelligence agent, dynamically call the antimicrobial peptide model according to demand to realize task scheduling, and also dynamically process the database in batches according to demand, call the antimicrobial peptide model to perform sequence generation, structure simulation, and simulation evaluation tasks in sequence, complete antimicrobial peptide sequence generation and primary screening, antimicrobial peptide structure simulation and secondary screening, antimicrobial activity, cytotoxicity and hemolytic simulation evaluation, and screen out candidate antimicrobial peptide sequences that meet the antimicrobial activity standards, meet specific structural characteristics, and meet the antimicrobial peptide inline characteristics; S4, conducts experimental evaluation on the results of simulated evaluation of antimicrobial activity, cytotoxicity and hemolysis, and then screens antimicrobial peptides in batches. The database is also updated based on the experimental evaluation results to provide feedback and optimize the antimicrobial peptide model. Specifically, directed experiments are conducted to test the antimicrobial activity, cytotoxicity and hemolysis of antimicrobial peptides. Based on the results, antimicrobial peptides with high antimicrobial activity, low cytotoxicity and negative hemolysis are screened out and added to the antimicrobial peptide vector database, while other antimicrobial peptides are recorded in the experimental vector database.

Images