A protein characterization learning method and device, computer equipment and storage medium
By integrating multimodal protein information through the MASSA framework and utilizing graph networks and memory enhancement mechanisms, this study addresses the issues of insufficient generalization ability and multimodal information fusion in protein representation learning for small-sample tasks. It achieves the effectiveness of cross-task transfer learning and improves the accuracy and efficiency of the model.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENZHEN INST OF ADVANCED TECH CHINESE ACAD OF SCI
- Filing Date
- 2025-01-22
- Publication Date
- 2026-07-24
AI Technical Summary
Existing protein characterization learning methods have insufficient generalization ability in small sample tasks, inadequate fusion of multimodal information, poor cross-task transfer learning performance, and serious data scarcity problems.
By integrating protein sequence, structure, and function annotation information through the MASSA framework, multimodal feature extraction and graph network construction are employed. Combined with memory enhancement mechanisms and transfer learning frameworks, multimodal information is dynamically fused and existing knowledge is transferred. The model is trained using similarity sharing and inter-task similarity in the graph network.
It improves the model's accuracy and generalization ability in few-shot learning tasks, solves the shortcomings of multimodal information fusion and cross-task transfer, alleviates the problem of data scarcity, and improves the efficiency and accuracy of protein characterization learning.
Smart Images
Figure CN122455104A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the fields of biology and artificial intelligence technology, and specifically relates to a protein characterization learning method, device, computer equipment, and storage medium. Background Technology
[0002] With the rapid development of biology, medicine, and drug development, protein characterization learning has become one of the core tasks in life science research. The function and structure of proteins are closely related, and this information often requires complex experimental procedures to obtain. In recent years, with the development of deep learning technology, protein characterization learning methods have been gradually incorporated into the latest algorithms in computer science to help solve the challenges of large-scale data analysis and prediction tasks.
[0003] In existing technologies, protein characterization learning methods mainly include the following directions:
[0004] (1) Protein sequence-based learning models employ sequence-based protein language models for protein representation learning. For example, DeepFRI (Deep Functional Representations for Proteins) predicts protein function using a deep learning model based on protein sequences, and the ESM (Evolutionary Scale Modeling) series of language models, such as ESM1b and ESM2, are also pre-trained models based on protein sequences. Although these models can capture complex patterns in protein sequences, they fail to fully consider protein structure and other modal information, ignore the complementarity of structural and functional annotation information, and have limited performance in small sample tasks.
[0005] (2) Protein learning based on graph neural networks (GNNs): Since the structure of proteins can be naturally represented as a graph, where amino acid residues can be used as nodes and chemical relationships between proteins can be represented as edges, graph networks have shown great potential in tasks such as protein function prediction and structural analysis. For example, some graph neural network methods in DeepChem have been used to predict protein interactions. However, this method is usually used for structural modality data, and it still suffers from insufficient generalization ability when dealing with small sample learning tasks.
[0006] In summary, current protein characterization learning still faces many challenges, especially in the following aspects:
[0007] (1) Data scarcity: Although protein datasets are constantly accumulating, multimodal data such as protein sequences, structures and functional annotations usually lack sufficient annotation information, and the acquisition process is complex and costly. Many datasets have problems such as small sample size and imbalanced samples, which makes traditional machine learning methods perform poorly in these tasks.
[0008] (2) Multimodal information fusion: Protein characterization not only depends on its sequence information, but also includes structural information, functional annotation, etc. These information belong to different modalities. However, traditional protein characterization learning methods mostly focus on information of a single modality and ignore the complementarity of information of different modalities. How to effectively integrate these modalities and fully explore the various features of proteins is a technical problem that needs to be solved.
[0009] (3) Cross-task transfer learning: In protein-related tasks, especially in few-shot learning tasks, effectively transferring knowledge learned from one task to another is an important way to improve the generalization ability of the model. However, traditional methods have certain limitations in few-shot and cross-task transfer, especially when the amount of data for the new task is insufficient to support traditional training methods. Summary of the Invention
[0010] This application provides a protein characterization learning method, apparatus, computer device, and storage medium, which aims to at least partially solve one of the aforementioned technical problems in the prior art.
[0011] To address the above problems, this application provides the following technical solution:
[0012] A protein characterization learning method, comprising:
[0013] Collect protein sequence, structure, and function annotation information to construct a multimodal feature dataset;
[0014] Multimodal features were extracted from the multimodal feature dataset using the MASSA framework to obtain multimodal protein characterization.
[0015] The protein multimodal representation is used to construct a graph network with protein tasks as nodes and distances between tasks as edges, and the graph network is used to store the representations and historical models of existing protein tasks.
[0016] The multimodal protein representation is used to train a task model. During the training of a new protein task, a memory enhancement mechanism is used to retrieve similar nodes of the new protein task from the graph network. The historical models of existing protein tasks corresponding to the similar nodes are obtained. The model parameters of the historical models are transferred through a transfer learning framework. The model is then trained for the new protein task. The protein representation learning prediction results are output through the trained task model.
[0017] The technical solution adopted in this application embodiment further includes: after collecting protein sequence, structure, and functional annotation information, it further includes:
[0018] The protein sequence, structure, and function annotation information is cleaned and standardized.
[0019] The technical solution adopted in this application embodiment also includes: the multimodal feature extraction of the multimodal feature dataset using the MASSA framework, specifically:
[0020] The ESM2 protein language model was used to extract features from protein sequences and obtain sequence characterization.
[0021] Graph convolutional networks are used to extract structural information features and capture the spatial topological relationships between protein residues;
[0022] Feature extraction of functional annotation information is performed based on functional annotation graphs, and graph neural networks are used to model hierarchical relationships of functions.
[0023] The technical solution adopted in this application embodiment further includes: extracting multimodal features from the multimodal feature dataset using the MASSA framework to obtain protein multimodal characterization, which further includes:
[0024] The sequence representation, spatial topological relationship, and functional hierarchical association are fused using a deep learning framework to obtain a protein multimodal representation.
[0025] The technical solution adopted in this application embodiment further includes: constructing a graph network with protein tasks as nodes and inter-task distances as edges using the protein multimodal representation, and storing existing protein task knowledge and representations through the graph network, specifically:
[0026] The graph network structure between different protein tasks is constructed using the protein multimodal representation and its similarity matrix, represented as: G = (V, E), where VVV are task nodes, each task node represents a protein task, and EEE are the edge weights between different tasks, where the edge weights represent the similarity between different protein tasks.
[0027] The technical solution adopted in this application embodiment further includes: constructing a graph network with protein tasks as nodes and inter-task distances as edges using the protein multimodal representation, and storing existing protein task knowledge and representations through the graph network, further including:
[0028] A dynamic memory module is set in the graph network to store the knowledge and representation of existing protein tasks.
[0029] The technical solution adopted in this application embodiment further includes: transferring the model parameters of the historical model through a transfer learning framework to train the model for the new protein task, specifically as follows:
[0030] After fine-tuning the transfer model parameters in conjunction with the new protein task, the model is trained for the new protein task.
[0031] By iteratively optimizing the attention weights of the model, a well-trained task model is generated.
[0032] Another technical solution adopted in this application embodiment is: a protein characterization learning device, comprising:
[0033] Information collection module: used to collect protein sequence, structure and function annotation information to construct a multimodal feature dataset;
[0034] Feature extraction module: used to extract multimodal features from the multimodal feature dataset using the MASSA framework to obtain multimodal protein characterization;
[0035] Graph network construction module: used to construct a graph network with protein tasks as nodes and distances between tasks as edges using the protein multimodal representation, and to store the representations and historical models of existing protein tasks through the graph network;
[0036] The memory enhancement and knowledge transfer module is used to train a task model using the protein multimodal representation. During the training of a new protein task, a memory enhancement mechanism is used to retrieve similar nodes of the new protein task from the graph network, obtain the historical models of existing protein tasks corresponding to the similar nodes, transfer the model parameters of the historical models through a transfer learning framework, train the model for the new protein task, and output the protein representation learning prediction results through the trained task model.
[0037] Another technical solution adopted in this application embodiment is: a computer device, the computer device including a processor and a memory coupled to the processor, wherein,
[0038] The memory stores program instructions for implementing the protein characterization learning method;
[0039] The processor is used to execute the program instructions stored in the memory to control the protein characterization learning method.
[0040] Another technical solution adopted in this application embodiment is: a storage medium storing processor-executable program instructions, the program instructions being used to execute the protein characterization learning method.
[0041] Compared with the prior art, the beneficial effects of the embodiments of this application are as follows: The protein representation learning method, device, computer equipment and storage medium of the embodiments of this application integrate multimodal information such as protein sequence, structure and function annotation through the MASSA framework, and achieve efficient fusion of multimodal information by adopting a dynamic weighting strategy based on attention mechanism. By combining multimodal features and the memory enhancement mechanism of graph network, the memory enhancement mechanism is used to dynamically retrieve relevant task experience in graph network, share existing knowledge and representations by utilizing the similarity between tasks, and transfer existing knowledge and representations by using the transfer learning framework. This effectively alleviates the problem of data scarcity, solves the shortcomings of protein representation learning in small sample learning, multimodal information fusion and cross-task transfer ability, and improves the accuracy and generalization ability of the model in small sample learning tasks. Attached Figure Description
[0042] Figure 1 This is a flowchart of the protein characterization learning method according to an embodiment of this application;
[0043] Figure 2 This is a schematic diagram of the protein characterization learning device according to an embodiment of this application;
[0044] Figure 3 This is a schematic diagram of the computer device structure according to an embodiment of this application;
[0045] Figure 4 This is a schematic diagram of the structure of the storage medium according to an embodiment of this application. Detailed Implementation
[0046] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0047] The terms "first," "second," and "third" in this application are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationships and movements between components in a specific orientation (as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or computer device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or computer devices.
[0048] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.
[0049] Specifically, please refer to Figure 1 This is a flowchart of a protein characterization learning method according to an embodiment of this application. The protein characterization learning method according to an embodiment of this application includes the following steps:
[0050] S100: Collect multimodal information such as different protein sequences, structures and functions (Gene Ontology, abbreviated as GO), and construct a multimodal feature dataset after preprocessing the multimodal information;
[0051] In this step, the preprocessing of multimodal information specifically involves cleaning and standardizing multimodal information such as different protein sequences, structural and functional annotations, and then using the cleaned and standardized multimodal information to construct a multimodal feature dataset.
[0052] S110: The MASSA (Multimodal Protein Representation Framework) framework is used to extract multimodal features from the multimodal feature dataset, obtaining initial feature vectors for protein sequence, structure and function annotations respectively. The initial feature vectors of protein sequence, structure and function annotations are then fused to generate a multimodal protein representation.
[0053] In this step, the multimodal feature extraction of the multimodal feature dataset using the MASSA multimodal framework specifically includes:
[0054] S111: Use the ESM2 (Evolutionary Scale Modeling) protein language model to extract features from protein sequences and obtain sequence representations;
[0055] S112: Graph Convolutional Network (GCN) is used to extract structural information features and capture the spatial topological relationships between protein residues;
[0056] S113: Extract features from functional annotation information based on GO graphs and use graph neural networks to model hierarchical functional relationships;
[0057] S114: Using deep learning frameworks such as PyTorch, the extracted sequence representations, spatial topological relationships, and functional hierarchical associations are fused to obtain multimodal protein representations.
[0058] S120: Construct a graph network based on protein multimodal representation, with protein tasks as nodes and distances between tasks as edges, and store the representations and historical models of existing protein tasks through the graph network.
[0059] In this step, the graph network construction method includes: constructing a graph network structure between different protein tasks using the multimodal representations of different protein tasks and their similarity matrices, represented as: G = (V, E), where VVV are task nodes, each task node represents a protein task, and protein tasks include but are not limited to functional prediction tasks or interaction analysis tasks; EEE are the edge weights between different tasks, and the edge weights represent the similarity between different protein tasks. In this application embodiment, OTDD (Optimal Transport Domain Distance) is used to measure the edge weights.
[0060] Furthermore, in this embodiment, a dynamic memory module is set in the constructed graph network to store the representations and historical models of existing protein tasks. For new protein tasks, the representations and association information between similar tasks are retrieved from the graph network through a memory enhancement mechanism. Transfer learning is then performed in combination with the historical models corresponding to similar tasks to provide a reference for new protein tasks, thereby achieving excellent performance and significantly improving the accuracy of model predictions. At the same time, when dealing with complex biological tasks, training time and computational costs can be reduced.
[0061] S130: Train the task model through protein multimodal representation, and during the training of the new protein task, use the memory enhancement mechanism to retrieve similar nodes of the new protein task from the graph network, obtain the historical model of the existing protein task corresponding to the similar node, transfer the model parameters of the historical model through the transfer learning framework, fine-tune the transferred model parameters in combination with the new protein task, and then train the model for the new protein task.
[0062] In this step, by combining memory enhancement mechanisms with graph networks, the graph network is used to reveal potential connections and similarities between different protein tasks. During the training of new tasks, the representations and historical models of existing protein tasks are dynamically retrieved from the memory modules of the graph network. The parameters of the historical models are transferred from existing protein tasks through a transfer learning framework, and fine-tuning is performed on new protein tasks using the transfer learning framework. This optimizes the transfer learning strategy, solves the overfitting problem in small sample data, improves the efficiency of transfer learning and the performance of the model in small sample learning tasks, supports efficient model fine-tuning across tasks, and reduces the dependence on target task data.
[0063] S140: Optimize the attention weights of the model through multiple iterations to generate a trained task model;
[0064] S150: Outputs protein representation learning prediction results through a trained task model;
[0065] In this step, the trained task model is applicable to multiple bioinformatics fields such as protein function prediction, interaction analysis, and drug development, and can generate more comprehensive and accurate protein characterizations.
[0066] Based on the above, the protein representation learning method of this application integrates multimodal information such as protein sequence, structure, and function annotations through the MASSA framework, and adopts a dynamic weighting strategy based on attention mechanism to achieve efficient fusion of multimodal information. By combining multimodal features and the memory enhancement mechanism of graph networks, the method dynamically retrieves relevant task experience in the graph network using the memory enhancement mechanism, shares existing knowledge and representations using the similarity between tasks, and transfers existing knowledge and representations using a transfer learning framework. This effectively alleviates the problem of data scarcity, solves the shortcomings of protein representation learning in few-shot learning, multimodal information fusion, and cross-task transfer capabilities, and improves the accuracy and generalization ability of the model in few-shot learning tasks.
[0067] Please see Figure 2 This is a schematic diagram of the protein characterization learning device according to an embodiment of this application. The protein characterization learning method device 40 according to an embodiment of this application includes:
[0068] Information collection module 41: Used to collect protein sequence, structure and function annotation information to construct a multimodal feature dataset;
[0069] Feature extraction module 42: used to extract multimodal features from the multimodal feature dataset using the MASSA framework to obtain protein multimodal characterization;
[0070] Graph network construction module 43: used to construct a graph network with protein tasks as nodes and distances between tasks as edges using the protein multimodal representation, and to store the existing protein task representations and historical models through the graph network;
[0071] Memory enhancement and knowledge transfer module 44: Used to train the task model using the protein multimodal representation, and during the training process of the new protein task, use the memory enhancement mechanism to retrieve similar nodes of the new protein task from the graph network, obtain the historical model of the existing protein task corresponding to the similar node, transfer the model parameters of the historical model through the transfer learning framework, train the model for the new protein task, and output the protein representation learning prediction result through the trained task model.
[0072] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.
[0073] The apparatus provided in this application can be applied to the foregoing method embodiments. For details, please refer to the description of the above method embodiments, which will not be repeated here.
[0074] Please see Figure 3This is a schematic diagram of a computer device structure according to an embodiment of this application. The computer device 50 includes:
[0075] Memory 51 storing executable program instructions;
[0076] Processor 52 connected to memory 51;
[0077] The processor 52 is used to call the executable program instructions stored in the memory 51 and perform the following steps: collect protein sequence, structure and functional annotation information to construct a multimodal feature dataset; use the MASSA framework to extract multimodal features from the multimodal feature dataset to obtain protein multimodal representations; use the protein multimodal representations to construct a graph network with protein tasks as nodes and inter-task distances as edges, and store the representations and historical models of existing protein tasks through the graph network; use the protein multimodal representations to train a task model, and during the training of a new protein task, use a memory enhancement mechanism to retrieve similar nodes of the new protein task from the graph network, obtain the historical models of existing protein tasks corresponding to the similar nodes, transfer the model parameters of the historical models through a transfer learning framework, train the model for the new protein task, and output the protein representation learning prediction results through the trained task model.
[0078] The processor 52 can also be referred to as a CPU (Central Processing Unit). The processor 52 may be an integrated circuit chip with signal processing capabilities. The processor 52 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor.
[0079] Please see Figure 4This is a schematic diagram of the storage medium in an embodiment of this application. The storage medium in this embodiment stores program instructions 61 capable of implementing the following steps: collecting protein sequence, structural, and functional annotation information to construct a multimodal feature dataset; extracting multimodal features from the multimodal feature dataset using the MASSA framework to obtain protein multimodal representations; constructing a graph network with protein tasks as nodes and inter-task distances as edges using the protein multimodal representations, and storing existing protein task representations and historical models through the graph network; training a task model using the protein multimodal representations, and during the training of a new protein task, using a memory enhancement mechanism to retrieve similar nodes of the new protein task from the graph network, obtaining the historical models of existing protein tasks corresponding to the similar nodes, transferring the model parameters of the historical models through a transfer learning framework, training the model for the new protein task, and outputting protein representation learning prediction results through the trained task model. The program instructions 61 can be stored in the aforementioned storage medium in the form of a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network computer device, etc.) or processor to execute all or part of the steps of the methods in various embodiments of this application. The aforementioned storage media include: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program instructions, or terminal computer devices such as computers, servers, mobile phones, and tablets. Servers can be independent servers or cloud servers providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.
[0080] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, apparatuses, or units, and may be electrical, mechanical, or other forms.
[0081] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.
Claims
1. A protein characterization learning method, characterized in that, include: Collect protein sequence, structure, and function annotation information to construct a multimodal feature dataset; Multimodal features were extracted from the multimodal feature dataset using the MASSA framework to obtain multimodal protein characterization. The protein multimodal representation is used to construct a graph network with protein tasks as nodes and distances between tasks as edges, and the graph network is used to store the representations and historical models of existing protein tasks. The multimodal protein representation is used to train a task model. During the training of a new protein task, a memory enhancement mechanism is used to retrieve similar nodes of the new protein task from the graph network. The historical models of existing protein tasks corresponding to the similar nodes are obtained. The model parameters of the historical models are transferred through a transfer learning framework. The model is then trained for the new protein task. The protein representation learning prediction results are output through the trained task model.
2. The protein characterization learning method according to claim 1, characterized in that, After collecting protein sequence, structural, and functional annotation information, the process also includes: The protein sequence, structure, and function annotation information is cleaned and standardized.
3. The protein characterization learning method according to claim 2, characterized in that, The process of extracting multimodal features from the multimodal feature dataset using the MASSA framework specifically involves: The ESM2 protein language model was used to extract features from protein sequences and obtain sequence characterization. Graph convolutional networks are used to extract structural information features and capture the spatial topological relationships between protein residues; Feature extraction of functional annotation information is performed based on functional annotation graphs, and graph neural networks are used to model hierarchical relationships of functions.
4. The protein characterization learning method according to claim 3, characterized in that, The step of using the MASSA framework to extract multimodal features from the multimodal feature dataset to obtain a protein multimodal characterization also includes: The sequence representation, spatial topological relationship, and functional hierarchical association are fused using a deep learning framework to obtain a protein multimodal representation.
5. The protein characterization learning method according to any one of claims 1 to 4, characterized in that, The process involves constructing a graph network using the protein multimodal representation, with protein tasks as nodes and distances between tasks as edges, and storing existing knowledge and representations of protein tasks through this graph network. Specifically: The graph network structure between different protein tasks is constructed using the protein multimodal representation and its similarity matrix, represented as: G = (V, E), where VVV are task nodes, each task node represents a protein task, and EEE are the edge weights between different tasks, where the edge weights represent the similarity between different protein tasks.
6. The protein characterization learning method according to claim 5, characterized in that, The method of constructing a graph network with protein tasks as nodes and distances between tasks as edges using the protein multimodal representation, and storing existing knowledge and representations of protein tasks through the graph network, further includes: A dynamic memory module is set in the graph network to store the knowledge and representation of existing protein tasks.
7. The protein characterization learning method according to claim 6, characterized in that, The step of transferring the model parameters of the historical model using a transfer learning framework to train the model for the new protein task specifically involves: After fine-tuning the model parameters for the transfer based on the new protein task, the model is trained for the new protein task. By iteratively optimizing the attention weights of the model, a well-trained task model is generated.
8. A protein characterization learning device, characterized in that, include: Information collection module: used to collect protein sequence, structure and function annotation information to construct a multimodal feature dataset; Feature extraction module: used to extract multimodal features from the multimodal feature dataset using the MASSA framework to obtain multimodal protein characterization; Graph network construction module: used to construct a graph network with protein tasks as nodes and distances between tasks as edges using the protein multimodal representation, and to store the representations and historical models of existing protein tasks through the graph network; The memory enhancement and knowledge transfer module is used to train a task model using the protein multimodal representation. During the training of a new protein task, a memory enhancement mechanism is used to retrieve similar nodes of the new protein task from the graph network, obtain the historical models of existing protein tasks corresponding to the similar nodes, transfer the model parameters of the historical models through a transfer learning framework, train the model for the new protein task, and output the protein representation learning prediction results through the trained task model.
9. A computer device, characterized in that, The computer device includes a processor and a memory coupled to the processor, wherein, The memory stores program instructions for implementing the protein characterization learning method according to any one of claims 1-7; The processor is used to execute the program instructions stored in the memory to control the protein characterization learning method.
10. A storage medium, characterized in that, The device stores processor-executable program instructions for performing the protein characterization learning method according to any one of claims 1 to 7.