Deep learning-based antimicrobial peptide series analysis method, device, equipment and medium

Abstract

This application relates to the field of bioinformatics. It provides a method, apparatus, equipment, and medium for analyzing antimicrobial peptide sequences based on deep learning. The method includes: extracting multi-scale structural features from the target antimicrobial peptide sequence using the ESM3 protein language model based on target application scenario parameters and Universal production system parameters, generating a structural feature vector; generating an initial peptide sequence from the structural feature vector; enhancing the bioactivity and adapting the initial peptide sequence to a mold using a multi-stage optimization network, generating candidate antimicrobial peptide sequences; and performing safety modifications on the candidate antimicrobial peptide sequences using a bio-inspired optimizer, generating an antimicrobial peptide design scheme and mold parameters. This approach aims to alleviate the fragmentation of multiple steps, reduce cross-platform development costs, shorten performance verification cycles, and thereby improve the efficiency of antimicrobial peptide transformation from design to industrialization.

Description

Technical Field

[0001] This invention relates to the field of bioinformatics, and in particular to methods, apparatus, equipment and media for analyzing antimicrobial peptide series based on deep learning. Background Technology

[0002] With the deep integration of bioinformatics and artificial intelligence, the intelligent design of antimicrobial peptides (AMPs) has become a core research direction in the fields of anti-infective therapy and biomanufacturing. Due to their broad-spectrum antibacterial activity and low risk of drug resistance, antimicrobial peptides have shown great potential in new drug development, functional modification of medical devices, and other applications.

[0003] However, in traditional antimicrobial peptide design, independent computational models lead to the fragmentation of feature extraction, sequence generation, and production adaptation processes. The lack of a unified optimization logic for bioactivity and process parameters results in a severe disconnect between the designed sequence and production requirements. Faced with diverse biomanufacturing platforms such as electrospinning, 3D printing, and hydrogel molding, traditional technologies require the development of dedicated sequences for each platform. This leads to redundant R&D resource waste due to the "one platform, one solution" model, and extremely low industrialization efficiency due to the lack of a universal adaptation mechanism. Furthermore, when relying on wet experiments such as inhibition zone tests and animal experiments to verify performance, the feedback cycle can be as long as several weeks or even months, making it impossible to guide iteration in real time and respond quickly to clinical needs. Summary of the Invention

[0004] Therefore, it is necessary to provide deep learning-based methods, devices, equipment, and media for analyzing antimicrobial peptides to address the aforementioned technical issues, thereby alleviating the fragmentation of multiple processes, reducing cross-platform development costs, shortening performance verification cycles, and ultimately improving the conversion efficiency of antimicrobial peptides from design to industrialization.

[0005] In a first aspect, this application provides a deep learning-based method for analyzing antimicrobial peptide series, the method comprising:

[0006] Based on the target application scenario parameters and Universal production system parameters, the target antimicrobial peptide sequence is subjected to multi-scale structural feature extraction processing using the ESM3 protein language model to generate a structural feature vector.

[0007] The structural feature vectors were processed to generate antimicrobial peptide sequences, resulting in initial peptide sequences.

[0008] The initial peptide sequence was bioactively enhanced and molded by a multi-stage optimization network to generate candidate antimicrobial peptide sequences.

[0009] Based on a bio-inspired optimizer, candidate antimicrobial peptide sequences are safely modified to generate antimicrobial peptide design schemes and mold forming parameters.

[0010] In one embodiment, a multi-stage optimization network is used to enhance the bioactivity of the initial peptide sequence and perform mold-forming adaptation to generate candidate antimicrobial peptide sequences, including:

[0011] The initial peptide sequence is subjected to bioactivity enhancement treatment through the first optimization layer to generate a bioactivity-enhanced sequence.

[0012] Based on the mold forming parameters in the Universal production system, the mold structure is adapted by molecular simulation driven by the second optimization layer to obtain the mold adaptation sequence.

[0013] Synergistic optimization of bioactivity-enhancing sequences and mold-fitting sequences was performed to generate candidate antimicrobial peptide sequences.

[0014] In one embodiment, based on the mold forming parameters in the Universal production system, a second optimization layer is used to perform molecular simulation-driven mold structure adaptation processing on the bioactivity enhancement sequence to obtain a mold adaptation sequence, including:

[0015] Based on the mold cavity geometry parameters in the mold forming parameters, the bioactivity enhancement sequence is spatially conformationally matched through molecular docking simulation to obtain the conformationally matched sequence.

[0016] Based on the mold material properties in the mold forming parameters, the conformation matching sequence is optimized for thermal stability through molecular dynamics simulation to generate a thermally stable sequence.

[0017] The demolding performance of the thermally stable sequence was optimized to obtain the demolding optimized sequence;

[0018] The molding shrinkage rate is controlled by processing the demolding optimization sequence to generate a mold adaptation sequence.

[0019] In one embodiment, the bioactivity-enhancing sequence and the mold-fitting sequence are synergistically optimized to generate candidate antimicrobial peptide sequences, including:

[0020] A balanced sequence is generated by performing multi-objective weight allocation on bioactivity enhancement sequences and mold adaptation sequences using reinforcement learning algorithms.

[0021] When an activity-forming conflict is detected in the equilibrium sequence, a priority dynamic adjustment mechanism is activated to resolve the conflict in the equilibrium sequence, resulting in a resolved sequence.

[0022] Based on real-time simulation feedback, the digested sequence is optimized for both bioactivity and mold forming compatibility to generate candidate antimicrobial peptide sequences.

[0023] In one embodiment, a bio-inspired optimizer is used to safely modify candidate antimicrobial peptide sequences to generate antimicrobial peptide design schemes and mold forming parameters, including:

[0024] Degradation risk analysis of candidate antimicrobial peptide sequences is performed using a pre-defined safety assessment unit to identify risk sites.

[0025] Using the following formula, based on risk sites, a trained biocompatibility optimizer is used to target and modify candidate antimicrobial sequences to obtain safety-enhanced sequences:

[0026]

[0027] in, This represents the optimized security enhancement sequence. This indicates the total number of identified risk sites. Indicates the first The weighting coefficients of each risk site, This represents a hazard assessment function for risk sites. Indicates the first One risk site, Indicates candidate antibacterial sequences, The regularization parameter representing the biocompatibility constraint. Represents the biocompatibility loss function;

[0028] The safety enhancement sequence was subjected to multi-dimensional production feasibility verification to obtain the antimicrobial peptide design scheme and mold forming parameters.

[0029] In one embodiment, the safety enhancement sequence undergoes multi-dimensional production feasibility verification to obtain the antimicrobial peptide design scheme and mold forming parameters, including:

[0030] The surface adsorption behavior of the safety enhancement sequence is simulated by a preset material-biological interface simulation unit to generate interface stability data.

[0031] Based on a pre-set injection molding process database, thermo-mechanical coupling simulation is performed on the safety enhancement sequence to generate a set of optimized process parameters;

[0032] Defect risk analysis is performed on interface stability data and process parameter optimization sets to identify production defect types.

[0033] Based on the type of production defect, the safety enhancement sequence is structurally fine-tuned using an iterative optimization algorithm to obtain an optimized sequence.

[0034] The optimized sequence was subjected to full-process injection molding simulation to generate antimicrobial peptide design schemes and mold forming parameters.

[0035] In one embodiment, the structural feature vector is subjected to antimicrobial peptide sequence generation processing to obtain an initial peptide sequence, including:

[0036] Conditional diffusion generation is performed on the structural feature vectors to obtain the peptide sequence framework;

[0037] The peptide sequence framework is enhanced with antibacterial activity based on the preset ByteNet temporal network to obtain the initial peptide sequence. The conditional diffusion generation process includes noise scheduling control based on ESM3 structural feature vectors.

[0038] Secondly, this application also provides a deep learning-based antimicrobial peptide series analysis device, which includes:

[0039] The feature extraction module is used to perform multi-scale structural feature extraction on the target antimicrobial peptide sequence based on the target application scenario parameters and Universal production system parameters, and generate structural feature vectors through the ESM3 protein language model.

[0040] The peptide sequence generation module is used to process the structural feature vector to generate antimicrobial peptide sequences, thus obtaining the initial peptide sequence.

[0041] The dual-modal optimization module is used to enhance the bioactivity of the initial peptide sequence and adapt it to the mold through a multi-stage optimization network to generate candidate antimicrobial peptide sequences.

[0042] The safety-parameter coordination module is used to safely modify candidate antimicrobial peptide sequences based on a bio-inspired optimizer, generating antimicrobial peptide design schemes and mold forming parameters.

[0043] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods in the first aspect of this application.

[0044] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods in the first aspect of this application.

[0045] The antimicrobial peptide series analysis method, device, equipment, and media provided in this application integrate target application scenario parameters with universal production system parameters, and use the ESM3 protein language model to achieve multi-scale structural feature extraction, laying a unified feature foundation for subsequent sequence design. This solves the problem of the separation between feature extraction, sequence generation, and production adaptation in traditional design. In the sequence generation and optimization process, based on universal production system parameters and a multi-stage optimization network, bioactivity enhancement and mold forming adaptation are processed collaboratively, eliminating the need to develop sequences separately for different production platforms and effectively reducing cross-platform development costs. At the same time, digital processing such as molecular simulation-driven adaptation through a multi-stage optimization network and safe modification by a bio-inspired optimizer replaces the traditional wet experiment-based verification method, significantly shortening the performance verification cycle. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart of a deep learning-based antimicrobial peptide series analysis method in one embodiment of the present invention;

[0048] Figure 2 The following is a flowchart of the mold forming parameters in a Universal production system based on one embodiment of the present invention: the mold structure adaptation process driven by molecular simulation of the bioactivity enhancement sequence is performed through the second optimization layer to obtain the mold adaptation sequence.

[0049] Figure 3 This is a structural diagram of an antimicrobial peptide series analysis device based on deep learning, according to one embodiment of the present invention. Detailed Implementation

[0050] To make the above-mentioned objects, features, and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0051] First, the application scenarios of the embodiments of this application are described. In the embodiments of this application, a deep learning-based method, apparatus, equipment, and medium for analyzing antimicrobial peptide series are provided, applicable to scenarios such as antimicrobial peptide drug development, functional modification of medical devices, and production of anti-infective biological agents.

[0052] For example, in the development of novel antimicrobial drugs, it is necessary to design antimicrobial peptide sequences that combine high antimicrobial activity with production feasibility for specific infection types; in the functionalization of medical dressings, it is necessary to design antimicrobial peptide molecules that can be adapted to spinning or 3D printing processes to improve the anti-infection performance of dressings; in large-scale biomanufacturing scenarios, it is necessary to quickly generate antimicrobial peptide production solutions that meet the requirements of different production platforms, etc.

[0053] The antimicrobial peptide series analysis method, device, equipment and medium based on deep learning provided in this application embodiment can also be applied to other application scenarios. This is only an example and does not limit the specific application scenarios.

[0054] like Figure 1 As shown, this application provides a deep learning-based method for analyzing antimicrobial peptide series, the method comprising:

[0055] S101: Based on the target application scenario parameters and Universal production system parameters, the target antimicrobial peptide sequence is processed by multi-scale structural feature extraction using the ESM3 protein language model to generate a structural feature vector.

[0056] For example, during multi-scale structural feature extraction, the target application scenario parameters and Universal production system parameters are integrated as the basic input conditions for feature extraction. Then, the ESM3 protein language model is invoked. Based on its deep understanding of protein sequences, this model performs multi-scale analysis of the target antimicrobial peptide sequence, from local details at the amino acid residue level to the overall structure at the peptide chain spatial conformation level. During this process, the aforementioned input parameters guide and constrain the direction and focus of feature extraction, ensuring that the extracted features both meet the application scenario requirements and adapt to the characteristics of the production system.

[0057] Through the aforementioned series of targeted analyses and processing, a structural feature vector is generated that can comprehensively characterize the multi-scale structural information of the target antimicrobial peptide sequence. The target application scenario parameters include scenario-specific requirements such as anti-infection type and targeted microbial characteristics. Universal production system parameters include common production standards and equipment parameters for various manufacturing platforms.

[0058] S102: Perform antimicrobial peptide sequence generation processing on the structural feature vector to obtain the initial peptide sequence.

[0059] For example, when generating antimicrobial peptide sequences from structural feature vectors, the multi-scale structural information contained in the structural feature vectors is used as the core basis, combined with the basic functional requirements of the target application scenario, to employ a sequence generation mechanism. This mechanism constructs an initial sequence generation framework by analyzing key information such as the arrangement rules and spatial conformational relationships of amino acid residues in the feature vectors, ensuring that the generated sequence backbone matches the structural characteristics represented by the feature vectors.

[0060] Subsequently, based on the potential constraints on the basic physicochemical properties of the sequence in the Universal production system parameters, the amino acid combinations within the framework were initially screened and adjusted, eliminating sequence fragments that clearly did not meet the basic production conditions. Through the above series of generation processes based on feature analysis and constraint adaptation, an initial peptide sequence with basic structural integrity and preliminary functional potential was obtained.

[0061] S103: The initial peptide sequence is enhanced with bioactivity and adapted to the mold by a multi-stage optimization network to generate candidate antimicrobial peptide sequences.

[0062] For example, when processing the initial peptide sequence through a multi-stage optimization network, the first-stage optimization process targets the core bioactivity indicators of the initial peptide sequence, such as antimicrobial spectrum coverage, antimicrobial efficacy, and anti-drug resistance potential, and combines them with the microbial characteristics of the target application scenario to perform targeted enhancement adjustments, thereby screening out sequence fragments that meet the bioactivity standards.

[0063] Next, the second stage of optimization begins. Based on the mold cavity size, material properties, and molding process requirements in the Universal production system parameters, the molecular conformational adaptation of the bioactive-enhanced sequence is adjusted, including optimizing the spatial matching degree between the sequence and the mold structure and improving the stability of the sequence in the molding environment.

[0064] By optimizing the network's synergistic regulation mechanism in a multi-stage manner, the potential constraints between bioactivity enhancement and mold forming adaptation are balanced, eliminating the influence of conflicting parameters between the two, and generating candidate antimicrobial peptide sequences that combine high bioactivity and good production adaptability.

[0065] S104: Based on a bio-inspired optimizer, candidate antimicrobial peptide sequences are safely modified to generate antimicrobial peptide design schemes and mold forming parameters.

[0066] For example, when performing safety modification on candidate antimicrobial peptide sequences based on a bio-inspired optimizer, potential risk sites in the candidate antimicrobial peptide sequences are systematically identified and evaluated according to the safety assessment criteria built into the bio-inspired optimizer. These potential risk sites include sequence fragments that may induce immunogenicity, cytotoxicity, or abnormal degradation.

[0067] Based on the assessment results, a targeted modification mechanism was used to adjust risk sites in accordance with biocompatibility requirements, while preserving the bioactivity and production adaptability of the sequence core. During the modification process, the process constraints of the Universal production system were simultaneously considered, and the modified sequence characteristics were correlated and optimized with key parameters for mold forming. These key parameters included cavity matching and material compatibility.

[0068] Through the above series of safety verifications and parameter adaptation processes, an antimicrobial peptide design scheme and matching mold forming parameters that take into account safety, functionality, and production feasibility are generated.

[0069] This application provides a deep learning-based method for analyzing antimicrobial peptide series. By integrating target application scenario parameters with Universal production system parameters, and leveraging the ESM3 protein language model, it achieves multi-scale structural feature extraction, laying a unified feature foundation for subsequent sequence design. This solves the problem of the fragmented process between feature extraction, sequence generation, and production adaptation in traditional design. During sequence generation and optimization, based on Universal production system parameters and a multi-stage optimization network, bioactivity enhancement and mold forming adaptation are processed collaboratively, eliminating the need to develop separate sequences for different production platforms and effectively reducing cross-platform development costs. Simultaneously, through digital processing such as molecular simulation-driven adaptation via a multi-stage optimization network and safe modification by a bio-inspired optimizer, it replaces the traditional wet experiment-dependent verification method, significantly shortening the performance verification cycle.

[0070] In one embodiment, a multi-stage optimization network is used to enhance the bioactivity of the initial peptide sequence and perform mold-forming adaptation to generate candidate antimicrobial peptide sequences, including:

[0071] The initial peptide sequence is subjected to bioactivity enhancement treatment through the first optimization layer to generate a bioactivity-enhanced sequence.

[0072] Based on the mold forming parameters in the Universal production system, the mold structure is adapted by molecular simulation driven by the second optimization layer to obtain the mold adaptation sequence.

[0073] Synergistic optimization of bioactivity-enhancing sequences and mold-fitting sequences was performed to generate candidate antimicrobial peptide sequences.

[0074] For example, using the first optimization layer, based on the initial peptide sequence, and combined with the microbial infection characteristics and antimicrobial efficacy standards corresponding to the target application scenario, targeted sequence structure adjustment and enhancement screening are performed on core bioactivity indicators such as the antimicrobial activity intensity, antimicrobial spectrum coverage, and antimicrobial resistance potential of the sequence. Fragments with insufficient activity are eliminated and high-efficiency sequence units are retained to generate bioactively enhanced sequences.

[0075] Next, the second optimization layer is invoked. Based on the preset mold forming parameters in the Universal production system, molecular simulation techniques such as molecular docking simulation and molecular dynamics simulation are used to adjust the spatial conformation of the bioactivity enhancement sequence to optimize its spatial fit with the mold cavity, its stability in the mold material environment, and its demolding performance after molding, thus obtaining the mold-fit sequence. The preset mold forming parameters include the geometric dimensions of the mold cavity, the surface characteristics of the cavity, and the physicochemical properties of the mold material.

[0076] To address the functional characteristics of bioactivity-enhancing sequences and the production compatibility characteristics of mold-fitting sequences, a collaborative optimization mechanism is used to balance the core indicators of both through multi-objective weight allocation. This approach specifically resolves the existing activity-molding conflict and verifies the compatibility of the two characteristics through real-time simulation feedback, thereby generating candidate antimicrobial peptide sequences that possess both high bioactivity and good mold-fitting properties.

[0077] like Figure 2 As shown, based on the mold forming parameters in the Universal production system, the mold structure is adapted through molecular simulation driven by the second optimization layer to obtain the mold adaptation sequence, which includes:

[0078] S201: Based on the mold cavity geometry parameters in the mold forming parameters, the bioactivity enhancement sequence is spatially conformationally matched by molecular docking simulation to obtain the conformationally matched sequence;

[0079] S202: Based on the mold material properties in the mold forming parameters, the conformation matching sequence is optimized for thermal stability through molecular dynamics simulation to generate a thermally stable sequence;

[0080] S203: Optimize the demolding performance of the thermally stable sequence to obtain the demolding optimized sequence;

[0081] S204: Perform molding shrinkage control processing on the demolding optimization sequence to generate a mold adaptation sequence.

[0082] For example, based on the geometric parameters of the mold cavity in the mold forming parameters, molecular docking simulation technology is used to adjust the spatial folding state and conformational features such as the orientation of amino acid side chains of the bioactive enhancement sequence, so that the spatial structure of the sequence matches the geometry of the cavity, generating a conformationally matched sequence. The geometric parameters of the mold cavity include the three-dimensional dimensions of the cavity, surface curvature, and the distribution of internal protrusions and depressions.

[0083] Subsequently, based on the characteristics of the mold material in the mold forming parameters, molecular dynamics simulations were used to simulate the molecular motion state of the sequence under the temperature environment of the mold material. This targeted enhancement of hydrogen bonding and hydrophobic interactions within the sequence improved its structural stability within the molding temperature fluctuation range, resulting in a thermally stable sequence. The characteristics of the mold material included high-temperature threshold, thermal conductivity, and chemical stability.

[0084] For thermally stable sequences, the intermolecular forces between them and the mold cavity surface are analyzed. By adjusting the amino acid composition at the end of the sequence or the side chain groups, the adsorption strength between the sequence and the cavity surface is reduced, thereby reducing the risk of structural damage during demolding and generating an optimized demolding sequence. The intermolecular forces on the mold cavity surface include van der Waals forces and electrostatic attraction.

[0085] By combining the shrinkage characteristics of the mold material and the cooling rate in the molding process, the molecular packing density, elastic modulus and other parameters of the demolding optimization sequence are finely adjusted so that the volume shrinkage of the sequence during the molding and curing process is adapted to the reserved space of the mold cavity, thus generating a mold-adapted sequence.

[0086] In one embodiment, the bioactivity-enhancing sequence and the mold-fitting sequence are synergistically optimized to generate candidate antimicrobial peptide sequences, including:

[0087] A balanced sequence is generated by performing multi-objective weight allocation on bioactivity enhancement sequences and mold adaptation sequences using reinforcement learning algorithms.

[0088] When an activity-forming conflict is detected in the equilibrium sequence, a priority dynamic adjustment mechanism is activated to resolve the conflict in the equilibrium sequence, resulting in a resolved sequence.

[0089] Based on real-time simulation feedback, the digested sequence is optimized for both bioactivity and mold forming compatibility to generate candidate antimicrobial peptide sequences.

[0090] For example, by using reinforcement learning algorithms, and combining the functional requirements priority of the target application scenario with the process constraints of the universal production system, multi-objective weight allocation is performed on the core activity indicators of the bioactivity enhancement sequence and the key molding parameters of the mold adaptation sequence. Through dynamic learning, the influence weights of each indicator are adjusted to form a balanced and coordinated sequence that retains high activity and good moldability. The core activity indicators of the bioactivity enhancement sequence include antibacterial efficacy and antibacterial spectrum range. The key molding parameters of the mold adaptation sequence include spatial matching degree and thermal stability.

[0091] When real-time monitoring reveals conflicts between bioactivity indicators and mold-forming requirements in the equilibrium sequence, a priority dynamic adjustment mechanism is immediately activated. Target weights are reallocated based on the urgency of the scenario, and the conflicting sites are targeted for adjustment, resulting in a resolved sequence. Conflicts between bioactivity indicators and mold-forming requirements include abnormal molding shrinkage caused by highly active structures and mold conformations that weaken antibacterial efficacy. The urgency of the scenario includes prioritizing activity in clinical emergency settings and focusing on molding stability in large-scale production settings. Targeted adjustments to conflicting sites include appropriately reducing non-core activity indicators to meet molding requirements or optimizing molding parameters to be compatible with key active structures.

[0092] Based on real-time feedback data from molecular simulation and production process simulation, the digested sequence is subjected to dual compatibility verification of bioactivity and mold forming. By fine-tuning the amino acid composition or conformation of the sequence, it is ensured that it meets the antibacterial efficacy standards and is compatible with the mold cavity structure, material properties and forming process, thus generating candidate antimicrobial peptide sequences with both high bioactivity and production feasibility.

[0093] In one embodiment, a bio-inspired optimizer is used to safely modify candidate antimicrobial peptide sequences to generate antimicrobial peptide design schemes and mold forming parameters, including:

[0094] Degradation risk analysis of candidate antimicrobial peptide sequences is performed using a pre-defined safety assessment unit to identify risk sites.

[0095] Using the following formula, based on risk sites, a trained biocompatibility optimizer is used to target and modify candidate antimicrobial sequences to obtain safety-enhanced sequences:

[0096]

[0097] in, This represents the optimized security enhancement sequence. This indicates the total number of identified risk sites. Indicates the first The weighting coefficients of each risk site, This represents a hazard assessment function for risk sites. Indicates the first One risk site, Indicates candidate antibacterial sequences, The regularization parameter representing the biocompatibility constraint. Represents the biocompatibility loss function;

[0098] The safety enhancement sequence was subjected to multi-dimensional production feasibility verification to obtain the antimicrobial peptide design scheme and mold forming parameters.

[0099] For example, using a pre-defined safety assessment unit, the molecular structure of candidate antimicrobial peptide sequences is systematically scanned according to biosafety standards. The analysis focuses on fragments that may lead to premature sequence degradation, trigger immune responses, or produce cytotoxicity, identifying risk sites. These biosafety standards include degradation rate thresholds and toxicity risk levels.

[0100] Next, a trained biocompatibility optimizer is invoked, using identified risk sites as core targets. Combining the hazard level of each risk site with biocompatibility requirements, the amino acid composition or chemical bond structure of risk sites in the candidate antimicrobial peptide sequence is directionally adjusted. This reduces safety risks while preserving the core function of the sequence, generating a safety-enhanced sequence. Biocompatibility requirements include compatibility with human tissues and blood.

[0101] Multi-dimensional production feasibility verification of the safety enhancement sequence was conducted, including chemical compatibility testing with mold materials in the Universal production system, structural stability testing under molding process conditions, and cost and efficiency assessment in the production process. Based on the verification results, the matching degree between the sequence structure and production parameters was optimized to generate an antimicrobial peptide design scheme and matching mold molding parameters that take into account safety, functionality and production feasibility.

[0102] In one embodiment, the safety enhancement sequence undergoes multi-dimensional production feasibility verification to obtain the antimicrobial peptide design scheme and mold forming parameters, including:

[0103] The surface adsorption behavior of the safety enhancement sequence is simulated by a preset material-biological interface simulation unit to generate interface stability data.

[0104] Based on a pre-set injection molding process database, thermo-mechanical coupling simulation is performed on the safety enhancement sequence to generate a set of optimized process parameters;

[0105] Defect risk analysis is performed on interface stability data and process parameter optimization sets to identify production defect types.

[0106] Based on the type of production defect, the safety enhancement sequence is structurally fine-tuned using an iterative optimization algorithm to obtain an optimized sequence.

[0107] The optimized sequence was subjected to full-process injection molding simulation to generate antimicrobial peptide design schemes and mold forming parameters.

[0108] For example, a preset material-biological interface simulation unit is used to simulate the interaction between the safety reinforcement sequence and the surface of the mold material used in production, analyze the characteristics of the sequence on the material surface such as adsorption strength, distribution uniformity and binding stability, and generate interface stability data that reflects the interface adaptation state of the two.

[0109] Then, the preset injection molding process database is called. Based on the process information such as temperature gradient, pressure range, and cooling rate contained in the database, thermo-mechanical coupling simulation is performed on the safety reinforcement sequence. The simulation simulates the effect of temperature changes and mechanical action on the structure of the sequence during injection molding, and then generates an optimized set of process parameters that are adapted to the characteristics of the sequence.

[0110] By combining the potential adsorption anomalies in the interface stability data with the potential fluctuation factors in the process parameter optimization set, a systematic analysis is conducted on the defect risks that may occur during the production process, such as sequence structure damage, incomplete molding, and adhesion to the mold, and the types of production defects are identified.

[0111] Based on the identified production defect types, an iterative optimization algorithm is used to make targeted fine adjustments to the amino acid sequence, molecular chain length, or side chain groups of the safety enhancement sequence in order to improve its stability and adaptability in production, resulting in an optimized sequence.

[0112] The optimized sequence was subjected to full-process injection molding simulation, including the complete production process from raw material injection, cavity filling to cooling and solidification, and demolding. The performance of the sequence in the actual production process was verified, and an antimicrobial peptide design scheme and matching mold forming parameters that meet production requirements were generated.

[0113] In one embodiment, the structural feature vector is subjected to antimicrobial peptide sequence generation processing to obtain an initial peptide sequence, including:

[0114] Conditional diffusion generation is performed on the structural feature vectors to obtain the peptide sequence framework;

[0115] The peptide sequence framework is enhanced with antibacterial activity based on the preset ByteNet temporal network to obtain the initial peptide sequence. The conditional diffusion generation process includes noise scheduling control based on ESM3 structural feature vectors.

[0116] For example, using the structural feature vector extracted by the ESM3 protein language model as the core constraint, the intensity and distribution of random interference during sequence generation are adjusted through a preset noise scheduling control mechanism. The sequence gradually evolves from an initial state with high noise to a low noise state, so that the generated sequence fragments continuously conform to the multi-scale information such as the amino acid residue connection rules and secondary structure tendencies contained in the structural feature vector, thereby constructing a peptide sequence framework with a relatively complete peptide backbone and conforming to the basic structural features.

[0117] Subsequently, the pre-defined ByteNet temporal network was used. Based on the learning results of antibacterial activity-related sequence patterns, the network was used to target and strengthen key regions in the peptide sequence framework that are related to antibacterial activity. This included optimizing the amino acid arrangement to enhance the potential for interaction with microbial targets and adjusting the length of local sequences to adapt to the antibacterial mechanism, while maintaining the integrity of the core structure of the framework.

[0118] Through the aforementioned conditional generation and activity-directed enhancement processes, an initial peptide sequence with both a reasonable basic structure and preliminary antibacterial potential was obtained. The antibacterial activity-related sequence patterns include the amino acid composition of specific functional domains and peptide chain folding tendencies.

[0119] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0120] In one embodiment, such as Figure 3 As shown, this application also provides a deep learning-based antimicrobial peptide series analysis device 300, which includes:

[0121] The feature extraction module 301 is used to perform multi-scale structural feature extraction on the target antimicrobial peptide sequence based on the target application scenario parameters and Universal production system parameters, and generate a structural feature vector by using the ESM3 protein language model.

[0122] The peptide sequence generation module 302 is used to perform antimicrobial peptide sequence generation processing on the structural feature vector to obtain the initial peptide sequence.

[0123] The dual-modal optimization module 303 is used to perform bioactivity enhancement and mold-forming adaptation on the initial peptide sequence through a multi-stage optimization network to generate candidate antimicrobial peptide sequences.

[0124] The safety-parameter coordination module 304 is used to safely modify candidate antimicrobial peptide sequences based on a bio-inspired optimizer, and generate antimicrobial peptide design schemes and mold forming parameters.

[0125] Specifically, the feature extraction module 301 integrates the target application scenario parameters and Universal production system parameters, and uses the ESM3 protein language model to perform multi-scale analysis on the target antimicrobial peptide sequence, including information such as the arrangement rules of amino acid residues, local secondary structure units, and overall spatial conformation dynamics. Through systematic extraction and integration, a structural feature vector that can comprehensively reflect the sequence structural characteristics is generated.

[0126] The peptide sequence generation module 302 takes the above-mentioned structural feature vector as the core input, and transforms the structural information in the feature vector into a sequence framework with a basic peptide backbone through the sequence generation mechanism. It also makes preliminary adjustments based on the basic requirements of antibacterial function to generate the initial peptide sequence.

[0127] The dual-modal optimization module 303 uses a multi-stage optimization network to target and enhance the bioactivity indicators such as the antibacterial efficacy and spectrum of action of the initial peptide sequence. Then, it optimizes the molding adaptation based on the mold cavity parameters and material physicochemical properties of the production system. By synergistically balancing the core requirements of both, it generates candidate antibacterial peptide sequences that take into account both activity and production adaptability.

[0128] The safety-parameter coordination module 304 uses a bio-inspired optimizer to identify and modify potential risk sites in candidate antimicrobial peptide sequences, generating safety-enhancing sequences. At the same time, it matches and optimizes these sequences with the molding parameters of the production mold to generate antimicrobial peptide design schemes and mold molding parameters.

[0129] The dual-modal optimization module 303 is also used for:

[0130] The initial peptide sequence is subjected to bioactivity enhancement treatment through the first optimization layer to generate a bioactivity-enhanced sequence.

[0131] Based on the mold forming parameters in the Universal production system, the mold structure is adapted by molecular simulation driven by the second optimization layer to obtain the mold adaptation sequence.

[0132] Synergistic optimization of bioactivity-enhancing sequences and mold-fitting sequences was performed to generate candidate antimicrobial peptide sequences.

[0133] The dual-modal optimization module 303 is also used for:

[0134] Based on the mold cavity geometry parameters in the mold forming parameters, the bioactivity enhancement sequence is spatially conformationally matched through molecular docking simulation to obtain the conformationally matched sequence.

[0135] Based on the mold material properties in the mold forming parameters, the conformation matching sequence is optimized for thermal stability through molecular dynamics simulation to generate a thermally stable sequence.

[0136] The demolding performance of the thermally stable sequence was optimized to obtain the demolding optimized sequence;

[0137] The molding shrinkage rate is controlled by processing the demolding optimization sequence to generate a mold adaptation sequence.

[0138] The dual-modal optimization module 303 is also used for:

[0139] A balanced sequence is generated by performing multi-objective weight allocation on bioactivity enhancement sequences and mold adaptation sequences using reinforcement learning algorithms.

[0140] When an activity-forming conflict is detected in the equilibrium sequence, a priority dynamic adjustment mechanism is activated to resolve the conflict in the equilibrium sequence, resulting in a resolved sequence.

[0141] Based on real-time simulation feedback, the digested sequence is optimized for both bioactivity and mold forming compatibility to generate candidate antimicrobial peptide sequences.

[0142] The safety-parameter coordination module 304 is also used for:

[0143] Degradation risk analysis of candidate antimicrobial peptide sequences is performed using a pre-defined safety assessment unit to identify risk sites.

[0144] Using the following formula, based on risk sites, a trained biocompatibility optimizer is used to target and modify candidate antimicrobial sequences to obtain safety-enhanced sequences:

[0145]

[0146] in, This represents the optimized security enhancement sequence. This indicates the total number of identified risk sites. Indicates the first The weighting coefficients of each risk site, This represents a hazard assessment function for risk sites. Indicates the first One risk site, Indicates candidate antibacterial sequences, The regularization parameter representing the biocompatibility constraint. Represents the biocompatibility loss function;

[0147] The safety enhancement sequence was subjected to multi-dimensional production feasibility verification to obtain the antimicrobial peptide design scheme and mold forming parameters.

[0148] The safety-parameter coordination module 304 is also used for:

[0149] The surface adsorption behavior of the safety enhancement sequence is simulated by a preset material-biological interface simulation unit to generate interface stability data.

[0150] Based on a pre-set injection molding process database, thermo-mechanical coupling simulation is performed on the safety enhancement sequence to generate a set of optimized process parameters;

[0151] Defect risk analysis is performed on interface stability data and process parameter optimization sets to identify production defect types.

[0152] Based on the type of production defect, the safety enhancement sequence is structurally fine-tuned using an iterative optimization algorithm to obtain an optimized sequence.

[0153] The optimized sequence was subjected to full-process injection molding simulation to generate antimicrobial peptide design schemes and mold forming parameters.

[0154] The peptide sequence generation module 302 is also used for:

[0155] Conditional diffusion generation is performed on the structural feature vectors to obtain the peptide sequence framework;

[0156] The peptide sequence framework is enhanced with antibacterial activity based on the preset ByteNet temporal network to obtain the initial peptide sequence. The conditional diffusion generation process includes noise scheduling control based on ESM3 structural feature vectors.

[0157] In one embodiment, during the diffusion generation stage, the structural feature vector, process parameters, and safety priors are concatenated into a ternary condition vector. Based on the estimation of site-level uncertainties using a protein language model, an uncertainty-driven noise scheduler is constructed to adaptively scale the noise intensity at each time step. This allows high-uncertainty sites to receive stronger constraints in the early stages while low-uncertainty sites are refined in the later stages, thereby improving the synergistic convergence speed of structural correctness and process feasibility, and reducing the probability of significant corrections in the later stages of generation.

[0158] The denoiser adopts a ByteNet structure and sets two sets of conditional FiLM parameters: active pathway embedding and molding pathway embedding. The system uses learnable Lagrange multipliers to jointly optimize targets such as "antibacterial efficacy / spectrum width" and "spatial matching / thermal stability / shrinkage rate / demolding performance". When an active-molding conflict is detected, a hierarchical reinforcement learner is triggered to adjust the target weights and repair strategies in real time according to the application scenario (e.g., emergency application or mass production transformation), so as to achieve online conflict resolution and steady-state convergence.

[0159] A constraint decoder is introduced during the generation process: alphabetical restrictions, toxicity motif masking, and regular syntax are used to pre-filter candidate fragments; targeted modifications (including but not limited to amino acid substitution and local reparameterization) are performed on identified risk sites, and regularization is performed with biocompatibility loss; combined with process simulation feedback, design schemes and molding parameters that meet safety and molding constraints are output, and site-level interpretation indicators (attention attribution and mutation sensitivity) are given to support review and production verification.

[0160] The artificial intelligence portion of this invention introduces a ternary conditional diffusion generation framework and a hierarchical multi-objective learning mechanism, achieving unified optimization along three main lines: structure, process, and safety.

[0161] (1) The multi-scale structural representation of protein language model, production process parameters and safety priors are used to form a ternary condition vector, which pre-process and safety constraints are applied in the generation stage to reduce rework in the subsequent stage.

[0162] (2) Introduce active / molding dual-channel gating and FiLM conditionalization in the ByteNet denoiser, and use learnable Lagrange multipliers to make an end-to-end trade-off between antibacterial efficacy and molding shrinkage / thermal stability targets;

[0163] (3) Combining hierarchical reinforcement learning with molecular / process simulation closed loop, the target weights are adjusted online and targeted repair is carried out on conflict sites;

[0164] (4) Constraint decoding and directional modification are used to achieve compliance upon generation and output site-level interpretation signals to support safety traceability and process alignment.

[0165] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0166] In one embodiment, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0167] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0168] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A deep learning-based method for analyzing antimicrobial peptide series, characterized in that, The method includes: Based on the target application scenario parameters and Universal production system parameters, the target antimicrobial peptide sequence is subjected to multi-scale structural feature extraction processing using the ESM3 protein language model to generate a structural feature vector. The structural feature vector is subjected to antimicrobial peptide sequence generation processing to obtain an initial peptide sequence; The initial peptide sequence is subjected to bioactivity enhancement and mold-forming adaptation processing through a multi-stage optimization network to generate candidate antimicrobial peptide sequences. The candidate antimicrobial peptide sequences are safely modified using a bio-inspired optimizer to generate antimicrobial peptide design schemes and mold forming parameters.

2. The method for analyzing antimicrobial peptide series based on deep learning according to claim 1, characterized in that, The process of enhancing the bioactivity of the initial peptide sequence and adapting it to a mold through a multi-stage optimization network to generate candidate antimicrobial peptide sequences includes: The initial peptide sequence is subjected to bioactivity enhancement treatment through a first optimization layer to generate a bioactivity-enhanced sequence. Based on the mold forming parameters in the Universal production system, the bioactivity enhancement sequence is subjected to molecular simulation-driven mold structure adaptation processing through the second optimization layer to obtain the mold adaptation sequence. The bioactivity-enhancing sequence and the mold-fitting sequence are synergistically optimized to generate the candidate antimicrobial peptide sequence.

3. The method for analyzing antimicrobial peptide series based on deep learning according to claim 2, characterized in that, The mold forming parameters in the Universal production system are optimized through a second optimization layer using molecular simulation-driven mold structure adaptation of the bioactivity enhancement sequence to obtain a mold adaptation sequence, including: Based on the mold cavity geometry parameters in the mold forming parameters, the bioactivity enhancement sequence is spatially conformationally matched using molecular docking simulation to obtain a conformationally matched sequence. Based on the mold material properties in the mold forming parameters, the conformation matching sequence is optimized for thermal stability through molecular dynamics simulation to generate a thermally stable sequence. The thermally stable sequence is subjected to demolding performance optimization processing to obtain the demolding optimized sequence; The molding shrinkage rate is controlled by processing the demolding optimization sequence to generate the mold adaptation sequence.

4. The method for analyzing antimicrobial peptide series based on deep learning according to claim 2, characterized in that, The step of synergistically optimizing the bioactivity-enhancing sequence and the mold-fitting sequence to generate the candidate antimicrobial peptide sequence includes: A balanced sequence is generated by performing multi-objective weight allocation processing on the bioactivity enhancement sequence and the mold adaptation sequence using a reinforcement learning algorithm. When an activity-forming conflict is detected in the equilibrium sequence, a priority dynamic adjustment mechanism is activated to resolve the conflict in the equilibrium sequence, resulting in a resolved sequence. Based on real-time simulation feedback, the digested sequence is optimized for both bioactivity and mold forming compatibility to generate the candidate antimicrobial peptide sequence.

5. The method for analyzing antimicrobial peptide series based on deep learning according to claim 1, characterized in that, The bio-inspired optimizer-based safe modification of the candidate antimicrobial peptide sequence to generate antimicrobial peptide design schemes and mold forming parameters includes: The candidate antimicrobial peptide sequences are subjected to degradation risk analysis using a preset safety assessment unit to identify risk sites. Using the following formula, based on the risk sites, the candidate antimicrobial sequences are targeted for modification using a trained biocompatibility optimizer to obtain safety-enhanced sequences: in, This represents the optimized security enhancement sequence. This indicates the total number of identified risk sites. Indicates the first The weighting coefficients of each risk site, This represents a hazard assessment function for risk sites. Indicates the first One risk site, Indicates candidate antibacterial sequences, The regularization parameter representing the biocompatibility constraint. Represents the biocompatibility loss function; The safety enhancement sequence was subjected to multi-dimensional production feasibility verification to obtain the antimicrobial peptide design scheme and mold forming parameters.

6. The method for analyzing antimicrobial peptide series based on deep learning according to claim 5, characterized in that, The process of performing multi-dimensional production feasibility verification on the safety enhancement sequence to obtain the antimicrobial peptide design scheme and mold forming parameters includes: The surface adsorption behavior of the safety enhancement sequence is simulated by a preset material-biological interface simulation unit to generate interface stability data. The safety enhancement sequence is subjected to thermo-mechanical coupling simulation based on a preset injection molding process database to generate a set of optimized process parameters; Defect risk analysis is performed on the interface stability data and the optimized set of process parameters to identify the types of production defects. Based on the aforementioned production defect type, the safety enhancement sequence is structurally fine-tuned using an iterative optimization algorithm to obtain an optimized sequence. The optimized sequence is subjected to a full-process injection molding simulation to generate the antimicrobial peptide design scheme and mold forming parameters.

7. The method for analyzing antimicrobial peptide series based on deep learning according to claim 1, characterized in that, The process of generating an antimicrobial peptide sequence from the structural feature vector to obtain an initial peptide sequence includes: Conditional diffusion generation is performed on the structural feature vectors to obtain the peptide sequence framework; The peptide sequence framework is subjected to antibacterial activity enhancement processing based on a preset ByteNet temporal network to obtain the initial peptide sequence. The conditional diffusion generation process includes noise scheduling control based on ESM3 structural feature vectors.

8. A deep learning-based antimicrobial peptide series analysis device, characterized in that, The device includes: The feature extraction module is used to perform multi-scale structural feature extraction on the target antimicrobial peptide sequence based on the target application scenario parameters and Universal production system parameters, and generate structural feature vectors through the ESM3 protein language model. A peptide sequence generation module is used to perform antimicrobial peptide sequence generation processing on the structural feature vector to obtain an initial peptide sequence; A dual-modal optimization module is used to perform bioactivity enhancement and mold-forming adaptation on the initial peptide sequence through a multi-stage optimization network to generate candidate antimicrobial peptide sequences. The safety-parameter coordination module is used to safely modify the candidate antimicrobial peptide sequence based on a bio-inspired optimizer, and generate antimicrobial peptide design schemes and mold forming parameters.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the deep learning-based antimicrobial peptide series analysis method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the deep learning-based antimicrobial peptide series analysis method as described in any one of claims 1 to 7.

Images