Protein-molecule characterization interactive prediction method, device, equipment and storage medium
By combining cross-modal feature fusion and a teacher-student model architecture with knowledge distillation techniques, the accuracy and efficiency of protein-molecule interaction prediction are improved, overcoming the limitations of existing models in terms of computational resources and data representation, and achieving efficient protein-molecule interaction prediction.
Patent Information
- Application Number
- CN202510096713.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-22
- Publication Date
- 2026-07-24
Smart Images
Figure CN122455082A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the fields of biomedicine and deep learning technology, and specifically relates to a method, device, equipment and storage medium for predicting protein-molecule characterization interactions. Background Technology
[0002] In recent years, the rapid development of deep learning technology has brought tremendous progress to research in protein characterization, molecular characterization, and related biomedical fields. However, as the scale of models increases, especially when using complex deep neural networks (DNNs) for tasks such as protein structure prediction or molecular graph embedding, training time and computational resource consumption have become significant bottlenecks. In these tasks, models typically need to handle millions or even billions of parameters and must be trained on massive datasets. This not only requires high-performance hardware resources but also increases the time cost of training and inference. Simultaneously, excessive computational and time consumption during training, especially when dealing with large-scale protein and molecular data, limits the practical application and widespread adoption of these models. Furthermore, most current protein-drug interaction (PDI) prediction methods can only handle single-modality data (such as protein sequences or molecular structures). In practical applications, proteins and molecules are intertwined, and the interaction relationships and representation effectiveness between different modalities have not been fully explored, resulting in limited predictive capabilities of the models. Meanwhile, existing protein and molecular characterization models fail to take into account the complex interactions between proteins and molecules in complex tasks such as cross-modal learning and multi-task learning, and cannot fully capture key information in protein sequences and molecular structures, resulting in insufficient expressive power of feature extraction. Summary of the Invention
[0003] This application provides a method, apparatus, device, and storage medium for predicting protein-molecular characterization interactions, which aims to at least partially solve one of the aforementioned technical problems in the prior art.
[0004] To address the above problems, this application provides the following technical solution:
[0005] A method for predicting protein-molecule characterization interactions, comprising:
[0006] Protein sequence data and molecular structure data are acquired separately, and feature extraction is performed on the protein sequence data and molecular structure data to obtain protein features and molecular structure features.
[0007] A Transformer-based protein molecule cross-fusion module is used to perform cross-modal feature fusion of the protein features and molecular structure features, and outputs the fused high-dimensional feature representation.
[0008] Based on the fused high-dimensional feature representation, a teacher-student model architecture and knowledge distillation technique are used to train and optimize the model, resulting in a well-trained prediction model.
[0009] The trained prediction model is used to predict the strength of protein-molecule interactions.
[0010] The technical solution adopted in this application embodiment further includes: the separate acquisition of protein sequence data and molecular structure data, specifically as follows:
[0011] The protein sequence data is in text format, and the molecular structure data is a molecular diagram representation or SMILES sequence, including atoms and bonds.
[0012] The technical solution adopted in this application embodiment further includes: extracting features from the protein sequence data and molecular structure data to obtain protein features and molecular structure features, including:
[0013] The protein sequence data is encoded using either letter-based sequence encoding or amino acid-based attribute encoding, and a pre-trained SaProt model is used to extract global and local features of the protein sequence data. The penultimate layer of the SaProt model outputs a protein feature vector of fixed dimensions.
[0014] The technical solution adopted in this application embodiment further includes: the step of extracting features from the protein sequence data and molecular structure data to obtain protein features and molecular structure features further includes:
[0015] Based on molecular graph embedding technology, atoms and bonds in the molecular structure data are represented as feature vectors. Graph convolutional neural networks are used to extract topological features and functional group information from the molecular structure data, and after multi-layer graph attention processing, low-dimensional molecular structure features are output.
[0016] The technical solution adopted in this application embodiment further includes: using a Transformer-based protein molecule cross-fusion module to perform cross-modal feature fusion of the protein features and molecular structure features, and outputting the fused high-dimensional feature representation, specifically:
[0017] The protein molecule cross-fusion module adopts a multi-layer decoder architecture. After projecting the protein features to the molecular feature dimension through a linear layer, it uses an attention mechanism to learn the interaction between the protein features and molecular structure features. The Transformer decoder extracts cross-modal interaction features and outputs the fused high-dimensional feature representation.
[0018] The technical solution adopted in this application embodiment further includes: based on the fused high-dimensional feature representation, the model training and optimization are performed using a teacher-student model architecture and knowledge distillation technology, specifically as follows:
[0019] During training, the Dropout mechanism is introduced in the feature extraction and cross-modal feature fusion process, and KL divergence is used as the distillation loss function. In the final regression layer, the protein features and molecular features are weighted and fused, and the interactive prediction results of the model are optimized by soft maxing weighting.
[0020] The technical solution adopted in this application embodiment further includes: predicting the protein-molecule interaction strength using the trained prediction model, specifically as follows:
[0021] The high-dimensional feature representation is reduced in dimensionality and predicted using a fully connected layer.
[0022] The protein features are spliced with the molecular features;
[0023] After processing the spliced features through a two-layer fully connected network, the final protein-molecular characterization interaction prediction result is output, which is a single regression value.
[0024] Another technical solution adopted in this application embodiment is: a protein-molecular characterization interaction prediction device, comprising:
[0025] Feature extraction module: used to acquire protein sequence data and molecular structure data respectively, and to extract features from the protein sequence data and molecular structure data to obtain protein features and molecular structure features;
[0026] Feature fusion module: Used to perform cross-modal feature fusion of the protein features and molecular structure features using a Transformer-based protein molecule cross-fusion module, and output the fused high-dimensional feature representation;
[0027] Model training module: used to train and optimize the model based on the fused high-dimensional feature representation, using a teacher-student model architecture and knowledge distillation technology, to obtain a trained prediction model;
[0028] PDI prediction module: used to predict the strength of protein-molecule interactions using the trained prediction model.
[0029] Another technical solution adopted in this application embodiment is: a device, the device including a processor and a memory coupled to the processor, wherein,
[0030] The memory stores program instructions for implementing the protein-molecule characterization interaction prediction method.
[0031] The processor is used to execute the program instructions stored in the memory to control the protein-molecular characterization interaction prediction method.
[0032] 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-molecule characterization interaction prediction method.
[0033] Compared to existing technologies, the beneficial effects of the embodiments of this application are as follows: The protein-molecular characterization interaction prediction method, apparatus, device, and storage medium of the embodiments of this application, by combining cross-modal feature fusion and a teacher-student model architecture, and by combining protein and molecular characterization models with multi-layer Transformer structures for feature extraction, can significantly improve the ability to capture key information in protein and molecular structures. Through deep fusion of protein and molecular features via cross-modal feature fusion, the complex interaction relationships between the two can be fully captured, overcoming the limitation of traditional methods that can only process single-modal data, and improving the prediction accuracy of PDI tasks. Through knowledge distillation technology, the teacher model guides the student model in learning, enabling the teacher and student models to work collaboratively on features at different levels, achieving a better understanding of complex biological interactions, improving the performance of the student model in protein-molecular interaction tasks, not only improving model performance but also accelerating training and reducing the risk of overfitting. The embodiments of this application can learn more meaningful features from complex protein-molecule interactions, further improving the accuracy of PDI tasks. While achieving efficient and accurate PDI prediction, it avoids the high computational cost required for large-scale model training in traditional methods, overcomes the limitations of traditional methods in terms of computational resources, efficiency and data representation, and improves the adaptability and transferability of the model in different tasks, which has important application value. Attached Figure Description
[0034] Figure 1 This is a flowchart of the protein-molecule characterization interaction prediction method according to an embodiment of this application;
[0035] Figure 2 This is a schematic diagram of the protein-molecular characterization interaction prediction device according to an embodiment of this application;
[0036] Figure 3 This is a schematic diagram of the device structure according to an embodiment of this application;
[0037] Figure 4 This is a schematic diagram of the structure of the storage medium according to an embodiment of this application. Detailed Implementation
[0038] 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.
[0039] The terms "first," "second," and "third" in this application are 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 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 devices.
[0040] 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.
[0041] Specifically, please refer to Figure 1 This is a flowchart of a protein-molecular characterization interaction prediction method according to an embodiment of this application. The protein-molecular characterization interaction prediction method according to an embodiment of this application includes the following steps:
[0042] S100: Acquire protein sequence data and molecular structure data respectively;
[0043] In this step, the protein sequence data is in text format, the molecular structure data is SNP (Single Nucleotide Polymorphism) sites or chemical structure data, and the molecular structure data is a molecular graph representation or SMI LES sequence, including atoms and bonds.
[0044] S110: After preprocessing the protein sequence data and molecular structure data, feature extraction is performed on the protein sequence data and molecular structure data respectively to obtain high-dimensional protein features and low-dimensional molecular structure features.
[0045] In this step, the preprocessing of protein sequence data and molecular structure data specifically includes:
[0046] S111: Protein sequence data is encoded using encoding methods such as letter-based sequence encoding or amino acid-based attribute encoding, and global and local features of the protein sequence data are extracted using a pre-trained SaProt model. A fixed-dimensional protein feature vector is output through the penultimate layer of the SaProt model. In this embodiment, by converting the protein sequence data into a high-dimensional protein feature vector, its biochemical properties and structural features can be better captured.
[0047] S112: Based on molecular graph embedding technology, atoms and bonds in molecular structure data are represented as feature vectors. Topological features and functional group information in molecular structure data are extracted using methods such as graph convolutional neural networks (GCN). After multi-layer graph attention processing, low-dimensional molecular structure features are output.
[0048] S120: Employs a Transformer-based protein-molecule cross-fusion module to perform cross-modal feature fusion of protein features and molecular structure features, captures the interaction information between proteins and molecules, and outputs the fused high-dimensional feature representation.
[0049] In this step, after obtaining protein features and molecular structure features, a Transformer-based protein-molecule cross-fusion module is used to perform cross-modal feature fusion and capture the correlation information between the two. Specifically, the protein-molecule cross-fusion module adopts a multi-layer decoder architecture. After projecting protein features to the molecular feature dimension through a linear layer, a self-attention mechanism is used to learn the interaction between protein features and molecular structure features. The Transformer decoder extracts cross-modal interaction features and outputs the fused high-dimensional feature representation.
[0050] It is understood that the embodiments of this application adopt a protein molecular cross-fusion module based on Transformer, which can fully explore the potential interaction relationship between proteins and molecular structures. During the information transmission process, it ensures that protein features and molecular features are effectively fused and enhanced, improves the model's understanding ability, enhances the model's learning ability for cross-modal interactions, and improves the prediction accuracy of PDI tasks.
[0051] S130: Based on the fused high-dimensional feature representation, a teacher-student model architecture and knowledge distillation technology are used for model training and optimization to obtain a well-trained prediction model;
[0052] In this step, to optimize the model training process, this embodiment employs a teacher-student model architecture and knowledge distillation techniques to improve training efficiency and reduce computational resource consumption. The teacher-student model architecture, as a knowledge distillation method, transfers the knowledge of a larger, well-trained "teacher" model to a smaller, simpler "student" model, enabling the student model to achieve performance close to the teacher model with fewer parameters and less computational overhead. Specifically, the teacher model is pre-trained on large-scale data, enabling it to learn more complex feature representations. Through knowledge distillation, the teacher model's logits and other prediction distributions are passed to the student model as targets. The student model can learn the complex patterns acquired by the teacher model during training, thus achieving efficient inference and training on a smaller scale. When dealing with tasks such as protein characterization and molecular characterization, the teacher-student model architecture can effectively reduce the computational complexity of the student model while maintaining high prediction accuracy. It reduces the parameter size and computational complexity of the student model and avoids the need for a complete update of the teacher model during each training session, thereby significantly reducing the number of parameters and computational complexity. It can be trained with less computing resources while maintaining high accuracy.
[0053] Furthermore, during model training, this embodiment employs a standard supervised learning loss function, while also introducing feature-based Triplet Loss and knowledge distillation loss to further enhance the model's learning ability and generalization performance. During training, to improve the model's generalization ability and reduce the risk of overfitting, this embodiment adopts the following optimization strategies: Dropout is introduced during feature extraction and cross-modal feature fusion to prevent overfitting during training; a teacher-student model architecture is used, employing KL divergence as the distillation loss function to ensure the student model effectively learns the knowledge from the teacher model, reducing computational overhead; in the final regression layer, protein features and molecular features are weighted and fused, and soft maxing is used to further optimize the final interactive prediction results.
[0054] It is understood that this invention introduces a strategy combining Triplet Loss and knowledge distillation. Triplet Loss guides the model to optimize the distance relationship between features, leading the model to better focus on representative feature regions during feature learning, thereby enhancing the model's representation ability in complex tasks, improving feature representativeness and robustness, and avoiding overfitting during learning. The combined use of knowledge distillation and Triplet Loss improves the model's understanding and modeling ability of complex interactions. By using a teacher-student model and a cross-modal feature fusion strategy, it can not only effectively learn the representation of a single task, but also share and transfer knowledge across multiple tasks through knowledge distillation and joint training. Multi-task learning enhances the model's generalization ability, sharing features across multiple tasks and optimizing knowledge transfer between tasks. Through pre-training of the teacher model and fine-tuning of the student model, a highly flexible and adaptable framework is provided, enabling rapid adaptation to multiple different tasks, improving the model's adaptability and transferability across different tasks. The modular design ensures the model's flexibility and generalization ability, and it can be extended to other PDI tasks.
[0055] S140: Predict the strength of protein-molecule interactions using a trained prediction model;
[0056] In this step, the prediction model specifically includes the following methods for predicting protein-molecule interaction strength:
[0057] S141: Dimensionality reduction and prediction of high-dimensional feature representations using fully connected layers (FC):
[0058] S142: Splicing protein features with molecular features;
[0059] S143: After processing the spliced features through a two-layer fully connected network, the final protein-molecule characterization interaction prediction result (PD I prediction score) is output.
[0060] Among them, the regression prediction result is a single regression value, which represents the strength of the interaction between proteins and molecules. This prediction result can be used for drug development or interaction analysis, promoting the research and development of new drugs and the understanding of complex biological interactions.
[0061] Based on the above, the protein-molecular representation interaction prediction method of this application, by combining cross-modal feature fusion and a teacher-student model architecture, and by combining protein and molecular representation models with multi-layer Transformer structures for feature extraction, can significantly improve the ability to capture key information in protein and molecular structures. Through cross-modal feature fusion, protein and molecular features are deeply fused, fully capturing the complex interaction relationships between them, overcoming the limitation of traditional methods that can only process single-modal data, and improving the prediction accuracy of PDI tasks. Through knowledge distillation technology, the teacher model guides the student model to learn, enabling the teacher and student models to work collaboratively on features at different levels, achieving a better understanding of complex biological interactions, improving the performance of the student model in protein-molecular interaction tasks, not only improving model performance but also accelerating training and reducing the risk of overfitting. This application embodiment can learn more meaningful features from complex protein-molecular interactions, further improving the accuracy of PDI tasks. While achieving efficient and accurate PDI prediction, it avoids the high computational cost required for large-scale model training in traditional methods, overcoming the limitations of traditional methods in terms of computational resources, efficiency, and data representation, and improving the model's adaptability and transferability in different tasks, thus possessing significant application value.
[0062] Please see Figure 2 This is a schematic diagram of the protein-molecular characterization interaction prediction device according to an embodiment of this application. The protein-molecular characterization interaction prediction method device 40 according to an embodiment of this application includes:
[0063] Feature extraction module 41: used to acquire protein sequence data and molecular structure data respectively, and to extract features from the protein sequence data and molecular structure data to obtain protein features and molecular structure features;
[0064] Feature fusion module 42: Used to perform cross-modal feature fusion of the protein features and molecular structure features using a Transformer-based protein molecule cross-fusion module, and output the fused high-dimensional feature representation;
[0065] Model training module 43: Used to train and optimize the model based on the fused high-dimensional feature representation, using a teacher-student model architecture and knowledge distillation technology, to obtain a trained prediction model;
[0066] PDI prediction module 44: used to predict the strength of protein-molecule interactions using the trained prediction model.
[0067] 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.
[0068] 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.
[0069] Please see Figure 3 This is a schematic diagram of the device structure according to an embodiment of this application. The device 50 includes:
[0070] Memory 51 storing executable program instructions;
[0071] Processor 52 connected to memory 51;
[0072] The processor 52 is used to call the executable program instructions stored in the memory 51 and perform the following steps: acquire protein sequence data and molecular structure data respectively, and extract features from the protein sequence data and molecular structure data to obtain protein features and molecular structure features; use a Transformer-based protein-molecule cross-fusion module to perform cross-modal feature fusion on the protein features and molecular structure features, and output the fused high-dimensional feature representation; based on the fused high-dimensional feature representation, use a teacher-student model architecture and knowledge distillation technology to train and optimize the model to obtain a trained prediction model; and use the trained prediction model to predict the strength of protein-molecule interactions.
[0073] 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.
[0074] 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 performing the following steps: acquiring protein sequence data and molecular structure data respectively, and extracting features from the protein sequence data and molecular structure data to obtain protein features and molecular structure features; using a Transformer-based protein-molecule cross-fusion module to perform cross-modal feature fusion of the protein features and molecular structure features, and outputting a fused high-dimensional feature representation; based on the fused high-dimensional feature representation, using a teacher-student model architecture and knowledge distillation technology to train and optimize the model, obtaining a trained prediction model; and predicting the strength of protein-molecule interactions using the trained prediction 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 device (which may be a personal computer, server, or network 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 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.
[0075] 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.
[0076] 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 method for predicting protein-molecular characterization interactions, characterized in that, include: Protein sequence data and molecular structure data are acquired separately, and feature extraction is performed on the protein sequence data and molecular structure data to obtain protein features and molecular structure features. A Transformer-based protein molecule cross-fusion module is used to perform cross-modal feature fusion of the protein features and molecular structure features, and outputs the fused high-dimensional feature representation. Based on the fused high-dimensional feature representation, a teacher-student model architecture and knowledge distillation technique are used to train and optimize the model, resulting in a well-trained prediction model. The trained prediction model is used to predict the strength of protein-molecule interactions.
2. The protein-molecular characterization interaction prediction method according to claim 1, characterized in that, The acquisition of protein sequence data and molecular structure data, respectively, specifically involves: The protein sequence data is in text format, and the molecular structure data is a molecular diagram representation or SMILES sequence, including atoms and bonds.
3. The protein-molecular characterization interaction prediction method according to claim 2, characterized in that, The step of extracting features from the protein sequence data and molecular structure data to obtain protein features and molecular structure features includes: The protein sequence data is encoded using either letter-based sequence encoding or amino acid-based attribute encoding, and a pre-trained SaProt model is used to extract global and local features of the protein sequence data. The penultimate layer of the SaProt model outputs a protein feature vector of fixed dimensions.
4. The protein-molecular characterization interaction prediction method according to claim 2, characterized in that, The step of extracting features from the protein sequence data and molecular structure data to obtain protein features and molecular structure features further includes: Based on molecular graph embedding technology, atoms and bonds in the molecular structure data are represented as feature vectors. Graph convolutional neural networks are used to extract topological features and functional group information from the molecular structure data, and after multi-layer graph attention processing, low-dimensional molecular structure features are output.
5. The protein-molecular characterization interaction prediction method according to any one of claims 1 to 4, characterized in that, The process employs a Transformer-based protein molecule cross-fusion module to perform cross-modal feature fusion of the protein features and molecular structure features, and outputs a fused high-dimensional feature representation, specifically as follows: The protein molecule cross-fusion module adopts a multi-layer decoder architecture. After projecting the protein features to the molecular feature dimension through a linear layer, it uses an attention mechanism to learn the interaction between the protein features and molecular structure features. The Transformer decoder extracts cross-modal interaction features and outputs the fused high-dimensional feature representation.
6. The protein-molecular characterization interaction prediction method according to claim 5, characterized in that, The high-dimensional feature representation based on the fusion is used for model training and optimization using a teacher-student model architecture and knowledge distillation techniques, specifically as follows: During training, the Dropout mechanism is introduced in the feature extraction and cross-modal feature fusion process, and KL divergence is used as the distillation loss function. In the final regression layer, the protein features and molecular features are weighted and fused, and the interactive prediction results of the model are optimized by soft maxing weighting.
7. The protein-molecular characterization interaction prediction method according to claim 6, characterized in that, The prediction of protein-molecule interaction strength using the trained prediction model specifically involves: The high-dimensional feature representation is reduced in dimensionality and predicted using a fully connected layer. The protein features are spliced with the molecular features; After processing the spliced features through a two-layer fully connected network, the final protein-molecular characterization interaction prediction result is output, which is a single regression value.
8. A protein-molecular characterization interaction prediction device, characterized in that, include: Feature extraction module: used to acquire protein sequence data and molecular structure data respectively, and to extract features from the protein sequence data and molecular structure data to obtain protein features and molecular structure features; Feature fusion module: Used to perform cross-modal feature fusion of the protein features and molecular structure features using a Transformer-based protein molecule cross-fusion module, and output the fused high-dimensional feature representation; Model training module: used to train and optimize the model based on the fused high-dimensional feature representation, using a teacher-student model architecture and knowledge distillation technology, to obtain a trained prediction model; PDI prediction module: used to predict the strength of protein-molecule interactions using the trained prediction model.
9. A device, characterized in that, The device includes a processor and a memory coupled to the processor, wherein, The memory stores program instructions for implementing the protein-molecule characterization interaction prediction 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-molecular characterization interaction prediction method.
10. A storage medium, characterized in that, The device stores processor-executable program instructions for performing the protein-molecular characterization interaction prediction method according to any one of claims 1 to 7.