Multi-modal feature fusion RNA binding protein recognition method and system

Abstract

The invention relates to the field of biological artificial intelligence crossing, and provides a multi-modal feature fusion RNA binding protein recognition method and system. The method comprises the following steps: performing internal enhancement on protein sequence characteristics, protein physicochemical property characteristics, protein structure characteristics and protein RNA interaction network characteristics respectively by adopting a self-attention mechanism; obtaining protein sequence enhancement characteristics, protein physicochemical property enhancement characteristics, protein structure enhancement characteristics and protein RNA interaction network enhancement characteristics; fusing the protein sequence enhancement feature, the protein physicochemical property enhancement feature, the protein structure enhancement feature and the protein RNA interaction network enhancement feature by adopting a cross attention mechanism to obtain a fusion feature; based on the fusion features, obtaining a combination specificity prediction result; wherein the coefficient of the self-attention mechanism and the coefficient of the cross-attention mechanism are subjected to regularization processing added with information bottleneck constraints, so that a loss function is constructed.

Description

TECHNICAL FIELD

[0001] The present application relates to the cross field of bioinformatics and artificial intelligence, and in particular to a multi-modal feature fusion RNA binding protein identification method and system. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute the prior art.

[0003] In recent years, with the rapid development of high-throughput sequencing (High-throughput Sequencing) technology and artificial intelligence algorithm, the bioinformatics field has accumulated a large amount of structured and unstructured multi-source data, providing an unprecedented opportunity for modeling the interaction relationship between RNA and protein. RNA binding proteins (RNA-binding proteins, RBP) play an important role in post-transcriptional regulation, and accurate prediction of their binding specificity is of great significance for understanding gene expression regulation mechanisms, revealing disease mechanisms and discovering new drug targets.

[0004] In existing research, although many computational methods based on machine learning or deep learning have been proposed for predicting RBP binding sites or specificity, most of these methods still use single modal information as input, such as RNA sequence, secondary structure, physical and chemical properties, or modification information. Although such methods have certain prediction ability, they have obvious shortcomings in capturing potential synergies and high-order semantic associations between different modalities, which limits the further improvement of prediction performance. SUMMARY

[0005] In order to solve the problems of low prediction accuracy of RNA binding protein (RBP) specificity and insufficient utilization of multi-modal information in the background art, the present application provides a multi-modal feature fusion RNA binding protein identification method and system, which designs a sparse attention method based on regularization Lq. By introducing sparsity constraints, the attention on redundant or irrelevant fragments is effectively suppressed. In the RNA binding protein (RBP) identification task, the introduction of this method helps to fuse the multiple biological features of RNA, thereby improving the prediction accuracy and generalization ability of binding specificity.

[0006] In order to achieve the above purpose, the present application adopts the following technical solutions: The first aspect of the present application provides a multi-modal feature fusion RNA binding protein identification method.

[0007] A multi-modal feature fusion RNA binding protein identification method, comprising: The protein sequence information, the protein physicochemical property information, the protein structure information, and the protein RNA interaction network feature are extracted to obtain protein sequence features, protein physicochemical property features, protein structure features, and protein RNA interaction network features. The protein sequence features, the protein physicochemical property features, the protein structure features, and the protein RNA interaction network features are respectively internally enhanced by using a self-attention mechanism to obtain protein sequence enhanced features, protein physicochemical property enhanced features, protein structure enhanced features, and protein RNA interaction network enhanced features. The protein sequence enhanced features, the protein physicochemical property enhanced features, the protein structure enhanced features, and the protein RNA interaction network enhanced features are fused by using a cross-attention mechanism to obtain fusion features. Based on the fusion features, a binding specificity prediction result is obtained. The coefficients of the self-attention mechanism and the coefficients of the cross-attention mechanism are subjected to regularization processing of adding information bottleneck constraints to construct a loss function and optimize the self-attention mechanism and the cross-attention mechanism.

[0008] Further, before the feature extraction, the protein sequence information, the protein physicochemical property information, the protein structure information, and the protein RNA interaction network are preprocessed, and the preprocessing includes standardization and embedding coding.

[0009] Further, the feature extraction includes: Based on the protein sequence information, a protein large language model ESM3 is used to perform high-dimensional context modeling on the original amino acid sequence to generate a high-dimensional embedding vector with rich structure and evolution information as a pre-training sequence feature representation of the protein. Based on the generated high-dimensional embedding vector, a BiLSTM model is used for sequence feature learning to extract the context dynamic representation of the protein sequence, that is, the protein sequence features. Based on the protein physicochemical property information, a DNN model is used to obtain the protein physicochemical property features. Based on the protein structure information, a GAT model is used to obtain the protein structure features. Based on the protein RNA interaction network, a GraphSAGE model is used to obtain the protein RNA interaction network features.

[0010] Further, the process of adopting the self-attention mechanism comprises: internally enhancing the protein sequence feature, the protein physicochemical property feature, the protein structure feature and the protein RNA interaction network feature to capture long-distance dependence and context information in the features, to obtain a protein sequence enhanced feature, a protein physicochemical property enhanced feature, a protein structure enhanced feature and a protein RNA interaction network enhanced feature.

[0011] Further, the loss function is:

[0012] wherein, denotes the loss function, denotes an original task loss function, , respectively, are regularization coefficients of the self-attention mechanism and the cross-attention mechanism, , respectively, are a matrix of the regularization coefficient of the self-attention mechanism and a matrix of the regularization coefficient of the cross-attention mechanism, and q denotes a regularization index, denotes a bottleneck constraint of the self-attention mechanism, denotes a bottleneck constraint of the cross-attention mechanism, , is a variable parameter, denotes an input, denotes a predicted target.

[0013] Further, the method comprises: inputting the fusion feature into a full connection layer for linear transformation and non-linear activation function processing to obtain a binding specificity prediction result.

[0014] The second aspect of the application provides a multi-modal feature fusion RNA binding protein recognition system.

[0015] A multi-modal feature fusion RNA binding protein recognition system comprises: A feature extraction module is configured to: extract protein sequence information, protein physicochemical property information, protein structure information and protein RNA interaction network features to obtain protein sequence features, protein physicochemical property features, protein structure features and protein RNA interaction network features. A self-attention mechanism module is configured to: respectively adopt a self-attention mechanism to internally enhance the protein sequence feature, the protein physicochemical property feature, the protein structure feature and the protein RNA interaction network feature to obtain a protein sequence enhanced feature, a protein physicochemical property enhanced feature, a protein structure enhanced feature and a protein RNA interaction network enhanced feature. a cross-attention mechanism module configured to: employ a cross-attention mechanism to fuse the protein sequence enhanced feature, the protein physicochemical property enhanced feature, the protein structure enhanced feature, and the protein RNA interaction network enhanced feature to obtain a fused feature; a prediction module configured to: obtain a binding specificity prediction result based on the fused feature; wherein the coefficients of the self-attention mechanism and the coefficients of the cross-attention mechanism are both subjected to regularization processing of adding an information bottleneck constraint to construct a loss function and optimize the self-attention mechanism and the cross-attention mechanism.

[0016] A third aspect of the present application provides a computer device, comprising: a processor adapted to execute a computer program; a computer readable storage medium, the computer readable storage medium storing a computer program, the computer program being executed by the processor to implement the steps in the multi-modal feature fusion-based RNA binding protein recognition method according to the first aspect.

[0017] A fourth aspect of the present application provides a computer readable storage medium storing a computer program, the computer program being adapted to be loaded and executed by a processor to implement the steps in the multi-modal feature fusion-based RNA binding protein recognition method according to the first aspect.

[0018] A fifth aspect of the present application provides a computer program product or a computer program.

[0019] The present application provides a computer program product or a computer program, which comprises computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to cause the computer device to perform the steps in the multi-modal feature fusion-based RNA binding protein recognition method according to the first aspect.

[0020] Compared with the prior art, the present application has the following advantages: The present application provides a multi-modal feature fusion-based RNA binding protein recognition method and system, which is based on multiple biological modalities of RNA (including sequence, structure, physicochemical property, modification information, etc.), extracts feature representation of each modality through a deep learning network, and realizes bidirectional high-order interaction between modalities by means of a self-attention mechanism and a cross-attention mechanism, thereby significantly improving the recognition ability of RBP specificity.

[0021] The application not only significantly improves the accuracy and robustness of RBP binding specificity recognition, but also can be used for analyzing the relative contribution of each modal feature to the final prediction, providing a computational basis for biological explanation of RNA-protein interaction mechanism, and has good theoretical value and practical application prospect. BRIEF DESCRIPTION OF DRAWINGS

[0022] The accompanying drawings, which form a part of this specification, are included to provide a further understanding of the application and are incorporated by reference herein. The embodiments illustrated in the drawings are presented by way of example in explaining the present application and are not intended to limit the present application.

[0023] Figure 1 is a flowchart of the multi-modal feature fusion RNA binding protein recognition method according to an embodiment of the present application; Figure 2 is a flowchart of another embodiment of the multi-modal feature fusion RNA binding protein recognition method according to an embodiment of the present application; Figure 3 is a structural diagram of the multi-modal feature fusion RNA binding protein recognition system according to an embodiment of the present application; Figure 4 is a structural diagram of the computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] The present application will be further described below in conjunction with the drawings and embodiments.

[0025] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise indicated, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0026] It should be noted that the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit exemplary embodiments according to the present application. As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and it should also be understood that when the terms "comprise" and / or "include" are used in the specification, there is a presence of the features, steps, operations, devices, components and / or combinations thereof.

[0027] Figure 1 is a flowchart of the multi-modal feature fusion RNA binding protein recognition method according to an embodiment of the present application; refer to Figure 1 The method comprises: protein sequence information, protein physicochemical property information, protein structure information and protein RNA interaction network features are extracted to obtain protein sequence features, protein physicochemical property features, protein structure features and protein RNA interaction network features. The self-attention mechanism is respectively used for internal enhancement of protein sequence characteristics, protein physicochemical property characteristics, protein structure characteristics and protein RNA interaction network characteristics, so as to obtain protein sequence enhanced characteristics, protein physicochemical property enhanced characteristics, protein structure enhanced characteristics and protein RNA interaction network enhanced characteristics; The cross-attention mechanism is used for fusing the protein sequence enhanced characteristics, the protein physicochemical property enhanced characteristics, the protein structure enhanced characteristics and the protein RNA interaction network enhanced characteristics, so as to obtain fused characteristics; Based on the fused characteristics, a binding specificity prediction result is obtained. The coefficients of the self-attention mechanism and the coefficients of the cross-attention mechanism are subjected to regularization processing of adding information bottleneck constraints, so as to construct a loss function and optimize the self-attention mechanism and the cross-attention mechanism.

[0028] The application provides a multi-modal feature fusion RNA binding protein identification method and system, which is based on multiple biological modal characteristics of RNA (including sequence, structure, physicochemical property, modification information and the like), extracts feature representation of each mode through a deep learning network, and realizes bidirectional high-order interaction between modes by means of a self-attention mechanism and a cross-attention mechanism, so that the recognition ability of RBP specificity is significantly improved.

[0029] Figure 2 is a flow chart of another embodiment of the multi-modal feature fusion RNA binding protein identification method shown in the embodiment of the application; in the embodiment, Multi-modal original data related to RNA binding proteins are collected, including RNA sequence, structure, physicochemical property, modification site and the like, and various data are subjected to standardization, embedding coding and the like pretreatment operations. For multi-modal data characteristics such as protein sequence, protein structure, protein RNA interaction network, protein physicochemical property, a suitable neural network is respectively used for feature extraction.

[0030] Specifically, for protein sequence information, the protein large language model ESM3 is introduced to model the high-dimensional context of the original amino acid sequence, generating high-dimensional embedding vectors with rich structural and evolutionary information as the pre-trained sequence feature representation of the protein. On this basis, further use the BiLSTM model to learn the sequence features of the high-dimensional embedding vectors generated by ESM3 to extract the context dynamic representation of the protein sequence and provide high-quality feature input for the downstream prediction task; for physicochemical property information, it can be selected from hundreds of amino acid property indicators in the Amino Acid Index (A-index) database, which is a database containing a set of numerical feature indicators of amino acids, each indicator reflects the physical, chemical, and biological properties of amino acids. For structure information, if experimental analysis structure is lacking, predicted structure can be generated by protein structure prediction tools such as AlphaFold, and then structure features are extracted; for protein-RNA interaction network information, structure perception models such as graph neural network (GNN) can be used to extract potential relationship features from the interaction graph. This module aims to characterize the biological characteristics of proteins from multiple dimensions and all directions to provide high-quality input for subsequent modeling.

[0031] In multi-modal learning tasks, how to effectively fuse information from different modalities is a key problem. In order to further optimize the feature representation and improve the model performance, first apply the self-attention mechanism to the extracted feature vectors for deep feature extraction. The role of self-attention mechanism is to enhance the internal features of each modality and capture long-distance dependencies and context information within the features. Self-attention mechanism calculates the relationship (attention score) between each element and other elements in the sequence, weights these relationships and fuses information, thereby generating more rich and context-related feature representations.

[0032] For the extracted feature vectors, first use the self-attention mechanism for further feature extraction. The basic idea of self-attention mechanism is: for each element in the input sequence, calculate its relationship (attention score) with other elements, and combine information with these relationships. The formula is as follows:

[0033] where X is the input feature, here referring to the feature vector after feature processing, is a learnable weight matrix. In the attention mechanism, the inner product of the query vector Q and the key vector K calculates the correlation or attention score between each element, and the value vector V is weighted and summed according to these attention scores to generate a new feature representation.

[0034] where the attention weights are obtained by applying a Softmax operation to the dot product results, ensuring that the sum of all weights is 1. By weighting and summing the features, the model can adjust its feature representation based on the relevance between different elements, making each element's representation more accurately reflect its contextual information.

[0035] When the feature vector after self-attention mechanism processing is obtained, cross-attention mechanism is used next to fuse multi-modal features. The main goal of cross-attention mechanism is to let each modality enhance its feature representation through interaction with other modalities. Cross-attention mechanism can effectively fuse the features of two different modalities by calculating the relationship between them. Specifically, in cross-attention mechanism, the features of one modality serve as the query (Query), while the features of another modality serve as the key (Key) and value (Value). In this way, the query modality's features can use the source modality's features to enhance their own representation, achieving information fusion between the two.

[0036] The calculation formula of cross-attention mechanism is as follows:

[0037] where, represents the processed feature vector from the "target" sequence, which refers to the processed feature vector here. represents the processed feature vector from the "source" sequence, which refers to the processed feature vector here. represents the dimension of key / query (used for scaling), which is usually used for scaling operation to avoid excessive values in calculation, respectively represent the lengths of the query sequence and the key / value sequence, representing the number of elements in the target modality and the source modality.

[0038] Cross-attention mechanism first calculates the inner product of the query vector Q and the key vector K to obtain the correlation scores between elements, and then normalizes these scores into weights through the Softmax operation. Then, these weights are used to weight and sum the value vector V of the source modality, and finally a fused feature representation is obtained. This process ensures that the target modality feature is enhanced according to the information of the source modality, and vice versa, so as to effectively capture the interaction relationship between multi-modal data.

[0039] For example, protein sequence information can be fused with physicochemical property information through cross-attention mechanism to learn the context dependence and coupling relationship between them. Specifically, for queries, keys and values from different modalities, the cross-attention mechanism calculates the relationship between them and generates a fused feature representation by weighted summation. Through this two-way information flow, the model can capture the mutual influence between the features of each modality, thereby generating a high-dimensional joint representation with stronger expression capability. After processing by the self-attention and cross-attention mechanisms, the feature vector obtained contains the fusion result of the multi-modal information and has stronger feature expression capability.

[0040] Meanwhile, in order to realize guided sparsity so that most elements in the attention matrix are closer to zero, encourage the model to focus on only a few key positions, reduce redundant attention, reduce the model's attention to irrelevant or weakly related segments, and improve the model's generalization ability, the present application proposes a sparse attention method based on regularized Lq, as follows:

[0041]

[0042] wherein, is the original task loss function (such as cross-entropy, mean square error, etc.); λ is the regularization coefficient (λ>0); is the attention weight matrix obtained by the present application; q controls the sparsity, the smaller q is, the sparser it is; when q=1, it degenerates to L1 regularization; when q=2, it is L2 regularization; when q=1 / 2, it obtains L1 / 2 regularization, which has stronger sparsity but is a non-convex function. The smaller q is, the more parameters are selected, making the sparsity better.

[0043] In deep learning, information bottleneck (IB) hopes to: compress irrelevant information, reduce the mutual information between attention and input ; retain information related to prediction, maximize the mutual information between attention and output . In order to avoid indiscriminately sparsifying weak but useful connections and retain as much valid information as possible, the present application adds an information bottleneck constraint, as follows:

[0044] wherein, is a variable parameter, greater than zero. represents the conditional probability distribution of A (attention matrix or its learnable representation) with respect to input X. is the mutual information between A and input X, is the mutual information between A and the prediction target Y, in practice, a differentiable upper bound (such as the variational information bottleneck VIB method) is used to approximate the calculation and add loss.

[0045] Specific to the task, the present application respectively regularizes the self-attention coefficient and the cross-attention coefficient, and the formula is as follows:

[0046] wherein, is the original task loss function (such as cross-entropy, mean square error, etc.); λ is the regularization coefficient (λ>0), and the regularization coefficients of the two attentions can be set respectively here; A is the attention coefficient matrix, and the two attention matrices are processed separately here; q represents the regularization index, is a variable parameter, and information bottleneck constraints are added to the self-attention and cross-attention respectively.

[0047] By combining the sparse self-attention mechanism and the cross-attention mechanism, deep fusion and optimization of multi-modal features can be achieved. In the self-attention phase, the model independently processes the features of each modality, so that each element can fully consider its context information; in the cross-attention phase, the model further strengthens the correlation between modalities by transmitting information between different modalities, so that the feature representation of each modality can reflect the important information of other modalities, and then a residual layer is added to further prevent overfitting.

[0048] After completing the multi-modal feature fusion, the fused features will be input into the fully connected layer for further processing. The fully connected layer maps the high-dimensional features to a lower-dimensional output space through the combination of linear transformation and nonlinear activation function of input features. In this process, the model can make classification prediction of RNA binding protein specificity according to the extracted and fused features. The final output result is 0 and 1. In this way, the model can accurately identify and distinguish different types of RNA binding proteins, thereby providing a basis for subsequent biological analysis and experiments.

[0049] During the training process, in order to improve the performance of the model and ensure its effectiveness, the AdamW optimizer is used to optimize the network parameters. The AdamW optimizer combines the advantages of the Adam optimizer, while introducing the WeightDecay strategy, which helps to prevent overfitting and promote the generalization ability of the model. In addition, in order to further improve the convergence speed and stability of the model, the learning rate will be dynamically adjusted according to the preset learning rate strategy during the training process, so that the learning rate gradually decreases as the training progresses, thereby ensuring that the model can gradually approach the global optimal solution in the optimization process. Through these optimization strategies, the model can more efficiently adjust the network parameters during the training process, and achieve the best performance in the final testing stage.

[0050] The above combines Figure 1The multi-modal feature fusion RNA binding protein recognition method provided by the embodiment of the present application is introduced in detail, and next, the multi-modal feature fusion RNA binding protein recognition system provided by the embodiment of the present application will be introduced in combination with the drawings.

[0051] Figure 3 is a structural schematic diagram of the multi-modal feature fusion RNA binding protein recognition system shown in the embodiment of the present application, referring to Figure 3 The system provided by the present application comprises: The feature extraction module is configured to extract protein sequence information, protein physicochemical property information, protein structure information and protein RNA interaction network features, and obtain protein sequence features, protein physicochemical property features, protein structure features and protein RNA interaction network features. The self-attention mechanism module is configured to use self-attention mechanism to internally enhance the protein sequence features, protein physicochemical property features, protein structure features and protein RNA interaction network features, and obtain protein sequence enhanced features, protein physicochemical property enhanced features, protein structure enhanced features and protein RNA interaction network enhanced features. The cross-attention mechanism module is configured to use cross-attention mechanism to fuse the protein sequence enhanced features, protein physicochemical property enhanced features, protein structure enhanced features and protein RNA interaction network enhanced features, and obtain fusion features. The prediction module is configured to obtain a binding specificity prediction result based on the fusion features. Wherein, the coefficients of the self-attention mechanism and the coefficients of the cross-attention mechanism are both subjected to regularization processing of adding information bottleneck constraint to construct a loss function and optimize the self-attention mechanism and the cross-attention mechanism.

[0052] In some embodiments, the feature extraction module is specifically configured to use multiple methods to process protein information, including sequence information, structure information, function information, etc., to obtain preliminary feature vectors. In order to further enhance the representation ability of the features, make the model pay attention to important areas, and prevent overfitting, the self-attention mechanism is used to process the preliminary feature vectors in depth. The self-attention mechanism can capture the complex relationship inside the features and combine the features by weighting, so as to extract more meaningful context information and obtain more accurate feature representation.

[0053] In some embodiments, the self-attention mechanism module and the cross-attention mechanism module are specifically configured to: after feature extraction, adopt a sparse cross-attention mechanism to perform bidirectional feature fusion on information of any two modalities. The cross-attention mechanism weights and fuses information of one modality into the representation of another modality by calculating the relationship between the features of different modalities, so that the information of the two modalities can complement and enhance each other. This bidirectional fusion can fully exploit the complementarity of information of different modalities, and improve the ability to judge the specificity of RNA binding proteins. Then, a residual layer is added to further prevent overfitting. Finally, the fused feature vectors are spliced into a new feature representation, providing more abundant information for subsequent classification.

[0054] In some embodiments, the prediction module is specifically configured to: after feature fusion, adopt a fully connected layer to further process the spliced fused feature vectors. The fully connected layer maps high-dimensional features to a target space through the combination of linear transformation and nonlinear activation function, and performs classification prediction of protein specificity. In this way, the model can effectively learn the complex relationship between protein characteristics and RNA binding specificity, and finally complete the judgment of RNA binding protein specificity.

[0055] The multi-modal feature fusion RNA binding protein identification system according to the embodiments of the present application can correspond to the method described in the embodiments of the present application, and the above and other operations and / or functions of each module of the multi-modal feature fusion RNA binding protein identification system are respectively for realizing the corresponding process of each method in Figure 1 For brevity, they will not be described here.

[0056] Referring to the structural diagram of the computer device shown in Figure 4 The computer device includes a processor, a communication interface, and a computer readable storage medium. The processor, the communication interface, and the computer readable storage medium can be connected through a bus or other means. The communication interface is used to receive and send data. The computer readable storage medium can be stored in the memory of the computer device, and the computer readable storage medium is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer readable storage medium. The processor (or CPU (Central Processing Unit, central processor)) is the computing core and control core of the computer device, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to realize the corresponding steps in the multi-modal feature fusion RNA binding protein identification method embodiment.

[0057] The embodiment of the present application provides a computer readable storage medium (Memory), which is a memory device in a computer device, and is used for storing programs and data. It can be understood that the computer readable storage medium can include a built-in storage medium in the computer device, and can also include an extended storage medium supported by the computer device. The computer readable storage medium provides a storage space, and the storage space stores a processing system of the computer device. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and the instructions can be one or more computer programs (including program codes). It should be noted that the computer readable storage medium can be a high-speed RAM memory, or a non-volatile memory such as at least one disk memory; optionally, the computer readable storage medium can also be at least one computer readable storage medium located away from the processor.

[0058] In one embodiment, the computer readable storage medium stores one or more instructions; the processor loads and executes the one or more instructions stored in the computer readable storage medium, so as to realize the corresponding steps in the above-mentioned multi-modal feature fusion RNA binding protein identification method embodiment.

[0059] The embodiment of the present application provides a computer program product or a computer program, and the computer program product or the computer program includes computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the corresponding steps in the above-mentioned multi-modal feature fusion RNA binding protein identification method embodiment.

[0060] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt a computer program product in the form of one or more computer usable storage media (including but not limited to disk storage and optical storage) containing computer usable program codes.

[0061] The embodiments of methods, apparatuses (systems) and computer program products according to the present application can be described in the general context of method steps and processes, which can be implemented in one embodiment by a program of instructions on a computer-readable storage medium executed by a computer or other programmable apparatus. The apparatuses can be specially constructed for executing the embodiments of methods, apparatuses (systems) and computer program products according to the present application or can include a computer or other programmable apparatus. Figure 1 The flow or flows and / or blocks in a flowchart and / or a block diagram Figure 1 The flow or flows and / or blocks in a flowchart and / or a block diagram

[0062] The computer program instructions can also be loaded onto a computer or other programmable apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 The flow or flows and / or blocks in a flowchart and / or a block diagram Figure 1 The flow or flows and / or blocks in a flowchart and / or a block diagram

[0063] The computer program instructions can also be loaded onto a computer or other programmable apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 The flow or flows and / or blocks in a flowchart and / or a block diagram Figure 1 The flow or flows and / or blocks in a flowchart and / or a block diagram

[0064] Those skilled in the art can understand that all or part of the flow of the above-mentioned embodiment method can be completed by computer program instructions instructing related hardware, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the flow of the above-mentioned embodiment of each method. The storage medium can be a magnetic disc, an optical disc, a read-only memory (ROM) or a random access memory (RAM), etc.

[0065] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A multi-modal feature fusion RNA binding protein recognition method, characterized in that, The method comprises the following steps: extracting protein sequence information, protein physicochemical property information, protein structure information, and protein RNA interaction network characteristics to obtain protein sequence characteristics, protein physicochemical property characteristics, protein structure characteristics, and protein RNA interaction network characteristics; respectively using a self-attention mechanism to internally enhance the protein sequence characteristics, the protein physicochemical property characteristics, the protein structure characteristics, and the protein RNA interaction network characteristics to obtain protein sequence enhanced characteristics, protein physicochemical property enhanced characteristics, protein structure enhanced characteristics, and protein RNA interaction network enhanced characteristics; using a cross-attention mechanism to fuse the protein sequence enhanced characteristics, the protein physicochemical property enhanced characteristics, the protein structure enhanced characteristics, and the protein RNA interaction network enhanced characteristics to obtain fused characteristics; based on the fused characteristics, obtaining a binding specificity prediction result; wherein the coefficients of the self-attention mechanism and the coefficients of the cross-attention mechanism are subjected to regularization processing of adding an information bottleneck constraint to construct a loss function and optimize the self-attention mechanism and the cross-attention mechanism. 2.The RNA-binding protein identification method of claim 1, wherein, Before feature extraction, the method further comprises pre-processing the protein sequence information, the protein physicochemical property information, the protein structure information, and the protein RNA interaction network, and the pre-processing comprises standardization and embedding coding. 3.The RNA-binding protein identification method of claim 1, wherein, The feature extraction comprises: based on the protein sequence information, using a protein large language model ESM3 to perform high-dimensional context modeling on an original amino acid sequence to generate a high-dimensional embedding vector with rich structure and evolution information as a pre-training sequence feature representation of the protein; based on the generated high-dimensional embedding vector, using a BiLSTM model to perform sequence feature learning to extract a context dynamic representation of the protein sequence, i.e., the protein sequence feature; based on the protein physicochemical property information, using a DNN model to obtain the protein physicochemical property characteristics; based on the protein structure information, using a GAT model to obtain the protein structure characteristics; based on the protein RNA interaction network, using a GraphSAGE model to obtain the protein RNA interaction network characteristics. 4.The RNA-binding protein identification method of claim 1, wherein, The process of using the self-attention mechanism comprises internally enhancing the protein sequence characteristics, the protein physicochemical property characteristics, the protein structure characteristics, and the protein RNA interaction network characteristics to capture long-distance dependencies and context information within the characteristics to obtain the protein sequence enhanced characteristics, the protein physicochemical property enhanced characteristics, the protein structure enhanced characteristics, and the protein RNA interaction network enhanced characteristics. 5.The RNA-binding protein identification method of claim 1, wherein, The loss function is: wherein, represents a loss function, represents an original task loss function, , are regularization coefficients of self-attention mechanism and cross-attention mechanism, respectively, , are matrix of regularization coefficients of self-attention mechanism and cross-attention mechanism, respectively, and q represents a regularization index, represents a bottleneck constraint of self-attention mechanism, represents a bottleneck constraint of cross-attention mechanism, , is a variable parameter, represents an input, represents a predicted target. 6.The RNA-binding protein identification method of claim 1, wherein, The method for obtaining the binding specificity prediction result based on the fused characteristics comprises inputting the fused characteristics into a fully connected layer for linear transformation and non-linear activation function processing to obtain the binding specificity prediction result.

7. A multi-modal feature fusion RNA binding protein recognition system, characterized in that, The method comprises the following steps: The feature extraction module is configured to extract protein sequence information, protein physicochemical property information, protein structure information, and protein RNA interaction network features to obtain protein sequence features, protein physicochemical property features, protein structure features, and protein RNA interaction network features. The self-attention mechanism module is configured to use a self-attention mechanism to internally enhance the protein sequence features, the protein physicochemical property features, the protein structure features, and the protein RNA interaction network features to obtain protein sequence enhanced features, protein physicochemical property enhanced features, protein structure enhanced features, and protein RNA interaction network enhanced features. The cross-attention mechanism module is configured to use a cross-attention mechanism to fuse the protein sequence enhanced features, the protein physicochemical property enhanced features, the protein structure enhanced features, and the protein RNA interaction network enhanced features to obtain fused features. The prediction module is configured to obtain a binding specificity prediction result based on the fused features. The coefficients of the self-attention mechanism and the coefficients of the cross-attention mechanism are both subjected to regularization processing of adding information bottleneck constraints to construct a loss function and optimize the self-attention mechanism and the cross-attention mechanism.

8. A computer device, characterized in that, a processor adapted to execute a computer program; a computer readable storage medium having stored therein a computer program, the computer program, when executed by the processor, implementing the steps of the multi-modal feature fusion-based RNA binding protein identification method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is adapted to be loaded and executed by the processor to implement the steps of the multi-modal feature fusion-based RNA binding protein identification method according to any one of claims 1-6.

10. A computer program product, characterised in that, The computer program product includes a computer program, and the computer program, when executed by the processor, implements the steps of the multi-modal feature fusion-based RNA binding protein identification method according to any one of claims 1-6.

Images