A prediction method and system for the interaction between circular RNA and RBP

Abstract

The present invention relates to the field of bioinformatics, and in particular to a method, system, and computer-readable storage medium for predicting the interaction between circular RNA and RBP. The method comprises: constructing a PreRBP model, comprising a feature extraction module and a prediction module; the feature extraction module comprises a TransRNAm module and three EMLinter modules, each of which takes a miRNA and a circular RNA as input and utilizes an attention mechanism and a gating mechanism to extract interaction features between the circular RNA and the miRNA, thereby enabling the discovery of interactions between RNAs in subsequent predictions. The present invention utilizes the commonalities and influences between other RNA post-transcriptional regulatory mechanisms and RBP binding to improve the accuracy of predicting the interaction between circular RNA and RBP.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics, and in particular to a method, system, and computer-readable storage medium for predicting the interaction between circular RNA and RBP. Background Art

[0002] Circular RNAs are ubiquitous in eukaryotic cells and have garnered widespread attention due to their conserved nature. They are closely linked to the pathogenesis of cancer. RNA-binding proteins (RBPs) are proteins that bind to RNA. Research has shown that RBPs play a key role in nearly all stages of the circular RNA (circRNA) lifecycle. Furthermore, RBPs can influence the expression of messenger RNA (mRNA) by binding to circRNAs.

[0003] In recent years, research on the interactions between circular RNAs and RBPs has continued to advance, encompassing computational approaches, traditional machine learning methods, and deep learning-based approaches. Computational methods, such as RNAcontext, consider both the original RNA sequence and RNA structure information, assuming that the structural expression profiles at different nucleotide positions are independent and distinct, and computationally derive predictions. Traditional machine learning methods, such as GraphProt, propose a method for identifying RBP-RNA interactions based on the Support Vector Machine (SVM) algorithm.

[0004] Deep learning-based methods include: DeepBind, a classic method that uses deep learning technology to predict the preferences of DNA and RNA binding proteins; RBPDL, which predicts the RBP sites of linear RNA and circular RNA through a deep multi-scale residual network combined with AdaBoost; CRIP, which uses a convolutional neural network combined with a classic architecture of a bidirectional long short-term memory network to predict RBP sites, and improves the model prediction accuracy through a codon-based encoding method; DeepA-RBPBS, which uses a convolutional neural network combined with BiGRU and an inserted attention module for prediction; DeCban, which optimizes sequence embedding and model architecture, introduces a dual embedding method and a cross-branch attention network, and alleviates the problem of difficulty in learning long sequences.

[0005] However, existing prediction methods for the interaction between circular RNA and RBP only consider the information of the circular RNA or protein itself, but do not consider the complex mechanisms involved in RNA post-transcriptional regulation, which reduces the prediction accuracy. Summary of the Invention

[0006] Therefore, the technical problem to be solved by the present invention is to overcome the fact that the existing technology only considers the information of circular RNA or protein itself, resulting in a reduced accuracy in predicting the interaction between circular RNA and RBP.

[0007] To solve the above technical problems, the present invention provides a method for predicting the interaction between circular RNA and RBP, comprising:

[0008] Input circular RNA into the TransRNAm module to obtain the circular RNA feature vector;

[0009] The three different miRNAs and circular RNAs were input into three EMLinter modules to obtain the first fusion feature vector, the second fusion feature vector and the third fusion feature vector respectively;

[0010] Each EMLinter module includes an attention branch and a gating branch;

[0011] In the attention branch, circular RNA and miRNA are respectively passed through their corresponding embedding layers and convolution blocks to obtain the first intermediate features of circular RNA and the first intermediate features of miRNA, and then passed through the approximate attention mechanism to obtain the second intermediate features of circular RNA and the second intermediate features of miRNA. The first intermediate features of circular RNA and the second intermediate features of miRNA are spliced ​​together and input into their corresponding maximum pooling layers to obtain the first target features of circular RNA and the first target features of miRNA. After splicing, they pass through the fully connected layer to obtain the output features of the attention branch.

[0012] In the gating branch, circular RNA and miRNA pass through their corresponding embedding layer, BiGRU gating mechanism, convolution block and maximum pooling layer respectively to obtain the second target feature of circular RNA and the second target feature of miRNA. After splicing, they pass through the fully connected layer to obtain the output feature of the gating branch;

[0013] The output features of the attention branch and the output features of the gate branch are concatenated to obtain a fused feature vector;

[0014] The circular RNA feature vector, the first fusion feature vector, the second fusion feature vector, and the third fusion feature vector are spliced ​​and input into the prediction module for binary classification to obtain the prediction results of the interaction between circular RNA and RBP.

[0015] Preferably, three different miRNA selection methods include:

[0016] The miRNAs in the database were divided into three categories using the KMeans clustering method, and one miRNA was selected from each category.

[0017] Preferably, the three miRNAs selected are hsa-miR-193b-3p, hsa-miR-20b-5p and hsa-miR-376b-3p.

[0018] Preferably, in the attention branch and the gating branch, the convolution block includes three sequentially connected convolution layers, and each convolution layer is connected to a normalization layer.

[0019] Preferably, the prediction module comprises three linear layers and a softmax activation function connected in sequence.

[0020] Preferably, before miRNA and circular RNA are input into the EMLinter module, circular RNA is padded to 22743 bp and miRNA is padded to 25 bp.

[0021] Preferably, the loss function for training the TransRNAm module, EMLinter module and prediction module is:

[0022] ;

[0023] in, represents the loss function, represents the total number of samples, represents the i-th sample, represents the true label of the i-th sample, Represents the prediction model composed of TransRNAm module, EMLinter module and prediction module.

[0024] Preferably, the TransRNAm module includes a Word2Vec model, a Transformer encoding block and a convolutional neural network connected in sequence; the output features of the Transformer encoding block and the output features of the convolutional neural network are spliced ​​through jump connections to obtain a circular RNA feature vector.

[0025] The present invention also provides a prediction system for the interaction between circular RNA and RBP, comprising:

[0026] The circular RNA feature extraction unit is used to input the circular RNA into the TransRNAm module to obtain the circular RNA feature vector;

[0027] The fusion feature extraction unit is used to input three different miRNAs and circular RNA into three EMLinter modules respectively to obtain the first fusion feature vector, the second fusion feature vector and the third fusion feature vector respectively; each EMLinter module includes an attention branch and a gating branch; in the attention branch, the circular RNA and the miRNA are respectively passed through their corresponding embedding layers and convolution blocks to obtain the first intermediate features of the circular RNA and the first intermediate features of the miRNA, and then respectively passed through the approximate attention mechanism to obtain the second intermediate features of the circular RNA and the second intermediate features of the miRNA; the first intermediate features of the circular RNA and the second intermediate features of the miRNA are respectively passed through the approximate attention mechanism to obtain the second intermediate features of the circular RNA and the second intermediate features of the miRNA; After feature splicing, the first intermediate feature and the second intermediate feature of the miRNA are spliced ​​and input into their corresponding maximum pooling layers respectively to obtain the first target feature of the circular RNA and the first target feature of the miRNA. After splicing, they pass through the fully connected layer to obtain the output feature of the attention branch. In the gating branch, the circular RNA and the miRNA pass through their corresponding embedding layer, BiGRU gating mechanism, convolution block and maximum pooling layer respectively to obtain the second target feature of the circular RNA and the second target feature of the miRNA. After splicing, they pass through the fully connected layer to obtain the output feature of the gating branch. The output features of the attention branch and the output features of the gating branch are spliced ​​to obtain a fused feature vector.

[0028] The prediction unit is used to splice the circular RNA feature vector, the first fusion feature vector, the second fusion feature vector and the third fusion feature vector and input them into the prediction module for binary classification to obtain the prediction result of the interaction between circular RNA and RBP.

[0029] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-mentioned method for predicting the interaction between circular RNA and RBP.

[0030] The above technical solution of the present invention has the following beneficial effects compared with the prior art:

[0031] The present invention discloses a method for predicting the interaction between circular RNA and RBP. The method uses the TransRNAm module to extract the characteristics of circular RNA and capture key regions of the upstream and downstream sequences of the RBP binding site. Furthermore, three different miRNAs are introduced as input, and three EMLinter modules are constructed. Each EMLinter module takes a circular RNA and a miRNA as input, and uses an attention mechanism and a gating mechanism to extract the interaction characteristics between the circular RNA and the miRNA. This allows the interaction between RNAs to be discovered in subsequent predictions. The method also utilizes the commonalities and influences between other RNA post-transcriptional regulatory mechanisms and RBP binding to improve the accuracy of predicting the interaction between circular RNA and RBP. Furthermore, the present invention is applicable not only to predicting the interaction between circular RNA and RBP, but also to predicting the interaction between linear RNA and RBP, showing strong versatility. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:

[0033] Figure 1 is a structural diagram of the PreRBP model of the present invention;

[0034] Figure 2 This is the structural diagram of the TransRNAm module;

[0035] Figure 3 This is the structure diagram of the EMLinter module. Figure 3 (a) in the figure is the structural diagram of the attention branch. Figure 3 (b) is the structural diagram of the gated branch. Figure 3 (c) in the figure is the structural diagram of the convolution block;

[0036] Figure 4 It is a comparison chart of the prediction results of the method of the present invention and other existing methods. DETAILED DESCRIPTION

[0037] The present invention will be further described below with reference to the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it. However, the embodiments are not intended to limit the present invention.

[0038] MicroRNA (miRNA) is an important participant in RNA post-transcriptional regulation and plays a fine regulatory role on gene expression at the post-transcriptional level. Therefore, miRNA can be used to improve the prediction accuracy of the interaction between circular RNA and RBP.

[0039] Example 1 of the present invention provides a method for predicting the interaction between circular RNA and RBP, which specifically comprises the following steps:

[0040] Step 1: Data preparation.

[0041] The present invention requires the acquisition of two parts of data. The first part is to obtain data on different circular RNAs targeting different RBPs, using 0 and 1 to mark whether the circular RNA has no (0) or has (1) interaction with the RBP, and the sequence information of the circular RNA. The second part is the sequence information of some miRNAs that interact with the corresponding RBPs, and the characteristics of these miRNAs are used to help analyze whether the circular RNA interacts with the RBP.

[0042] Step 2: Construct the PreRBP model.

[0043] Reference Figure 1 As shown in FIG, the PreRBP model includes a feature extraction module and a prediction module; the feature extraction module includes a TransRNAm module and three EMLinter modules.

[0044] The TransRNAm module is used to extract the characteristic representation of circular RNA, capture key nucleotide sites, and capture the key regions of the upstream and downstream sequences of the RBP binding site.

[0045] The circular RNA was input into the TransRNAm module to obtain a circular RNA feature vector of size 1×768.

[0046] The input circular RNA sequence is converted to a length of 601 nucleotides by trimming or padding.

[0047] Reference Figure 2 As shown in FIG, the TransRNAm module includes a Word2Vec model, a Transformer encoding block, and a convolutional neural network connected in sequence; the output features of the Transformer encoding block and the output features of the convolutional neural network are spliced ​​through jump connections to obtain a circular RNA feature vector.

[0048] The Word2Vec model is used to generate a coding matrix from RNA nucleotide sequences. A 601-nucleotide circular RNA is converted into a 599 x 300 matrix using the Word2Vec model. Each group of three nucleotides is grouped together using a sliding window, shifting backward one nucleotide at a time. For a 601-nucleotide sequence, there are 599 groups.

[0049] The Transformer encoding block and convolutional neural network are used to extract the features of circular RNA, thereby obtaining an abstract representation of circular RNA.

[0050] The Transformer encoding block takes the encoding matrix output by the Word2Vec model as input, and includes a multi-head attention layer, a first normalization layer, a feedforward neural network and a second normalization layer connected in sequence; the input features and output features of the multi-head attention layer are added through a skip connection as the input of the first normalization layer, and the input features and output features of the feedforward neural network are added through a skip connection as the input of the second normalization layer.

[0051] The convolutional neural network includes a first convolutional layer, a second convolutional layer, a first maximum pooling layer, a third convolutional layer and a second maximum pooling layer which are connected in sequence.

[0052] The EMLinter module is a multi-view module responsible for predicting interactions between miRNAs and circRNAs. The three EMLinter modules independently extract interaction features between circRNAs and three different miRNAs without sharing parameters. Its primary function is to discover interactions between RNAs and incorporate information about RNA interference regulation.

[0053] Three different miRNAs and circular RNAs were input into three EMLinter modules, and the first fusion feature vector, the second fusion feature vector, and the third fusion feature vector were obtained, all with a size of 1×256.

[0054] Due to the large number of miRNAs, three different miRNA selection methods are provided in this embodiment:

[0055] Considering the way and dimension of data read into the model, the iFeatureOmega tool was used to convert the sequence of each miRNA in the database into a 176-dimensional feature vector, and then divided into three categories using the KMeans clustering method, and one miRNA was selected in each category.

[0056] The three miRNAs selected in this example are hsa-miR-193b-3p, hsa-miR-20b-5p, and hsa-miR-376b-3p. These three miRNAs are the most representative miRNAs selected based on the clustering algorithm and can best represent the characteristics of similar miRNAs.

[0057] Reference Figure 3 As shown in Figure 2, each EMLinter module takes a miRNA and a circular RNA as input and includes an attention branch and a gating branch. Before inputting miRNA and circular RNA into the EMLinter module, circular RNA is padded to 22743 bp and miRNA is padded to 25 bp.

[0058] Reference Figure 3As shown in (a), in the attention branch, circular RNA and miRNA are respectively passed through their corresponding embedding layers and convolution blocks to obtain the first intermediate features of circular RNA and the first intermediate features of miRNA, and then respectively passed through the approximate attention mechanism to obtain the second intermediate features of circular RNA and the second intermediate features of miRNA; the first intermediate features and the second intermediate features of circular RNA and the first intermediate features and the second intermediate features of miRNA are spliced ​​together and input into their corresponding maximum pooling layers respectively to obtain the first target features of circular RNA and the first target features of miRNA, and then the output features of the attention branch are obtained after splicing through the fully connected layer.

[0059] Reference Figure 3 As shown in (b), in the gating branch, circular RNA and miRNA pass through their corresponding embedding layer, BiGRU gating mechanism, convolution block and maximum pooling layer respectively to obtain the second target feature of circular RNA and the second target feature of miRNA. After splicing, they pass through the fully connected layer to obtain the output feature of the gating branch.

[0060] Reference Figure 3 As shown in (c), in the attention branch and the gating branch, the convolution block includes three sequentially connected convolution layers, and each convolution layer is connected to a normalization layer.

[0061] The convolution block is used to extract RNA features, and the attention mechanism and the gating mechanism are used to extract deeper features.

[0062] The output features of the attention branch and the output features of the gating branch are concatenated to obtain a fused feature vector.

[0063] The circular RNA feature vector, the first fusion feature vector, the second fusion feature vector, and the third fusion feature vector were concatenated into a feature vector of size 1×1536, which was input into the prediction module for binary classification to obtain the prediction results of the interaction between circular RNA and RBP.

[0064] The prediction module includes three linear layers connected in sequence and a softmax function. The dimensions of the output feature vectors of the three linear layers are 512, 256, and 2, respectively. Since the prediction of circRNA-RBP interactions is a binary classification problem, a softmax activation function is used to obtain the final prediction results.

[0065] If the prediction result of the interaction between circular RNA and RBP is 1, it means that circular RNA and RBP interact with each other; if the prediction result of the interaction between circular RNA and RBP is 0, it means that circular RNA and RBP do not interact with each other.

[0066] Step 3: Train the PreRBP model.

[0067] Specifically, the pre-trained weights of the TransRNAm module and the EMLinter module are used as the initial weights of the PreRBP model, and the parameters of each layer are updated by fine-tuning.

[0068] The cross entropy function was used to calculate the loss function during the training process, and the Adam optimizer was used for training. The learning rate was set to 0.001, and a learning rate decay coefficient of 0.98 was used to improve the stability of the training.

[0069] The formula of the loss function is:

[0070] ;

[0071] in, Represents the loss function, and the goal of the training process is to minimize the loss function; represents the total number of samples, represents the i-th sample, represents the true label of the i-th sample, Represents the PreRBP model.

[0072] Step 4: Obtain the circular RNA to be detected and input it into the trained PreRBP model together with three different miRNAs to obtain the prediction results of the interaction between circular RNA and RBP.

[0073] On the same dataset, this example compares the prediction results of the PreRBP model with those of other existing methods, such as Figure 4 The comparison index used in this embodiment is AUC, which is a floating point number between 0 and 1. When it is greater than 0.5, it means that the prediction ability is greater than random guessing, and the closer it is to 1, the stronger the prediction ability.

[0074] Depend on Figure 4 It can be seen that the predictive ability of the PreRBP model is significantly stronger than other existing methods, and it performs better in predicting the interaction between circular RNA and RBP.

[0075] This example also used the constructed PreRBP model to conduct experiments targeting linear RNAs. Although linear RNAs and circular RNAs differ in structure and function, their binding mechanisms with RBPs are likely similar. Therefore, the same methods used to predict circular RNA-binding proteins can be used to predict linear RNA-binding proteins.

[0076] Across different datasets, the input data processing, model construction, and training process for linear RNAs are similar to those for circRNAs. This example compared prediction results on 31 publicly available datasets with the state-of-the-art linear RNA model, iCircRBP-DHN. Results for 15 datasets significantly outperformed iCircRBP-DHN, while the remaining datasets were comparable. This demonstrates the generalizability of the proposed method, which is applicable not only to circRNAs but also to linear RNAs.

[0077] In summary, the present invention discloses a method for predicting the interaction between circular RNA and RBP. The method utilizes the TransRNAm module to extract the characteristics of circular RNA and capture key regions of the upstream and downstream sequences of the RBP binding site. Three different miRNAs are additionally introduced as input, and three EMLinter modules are constructed. Each EMLinter module takes a circular RNA and a miRNA as input, and utilizes an attention mechanism and a gating mechanism to extract the interaction characteristics between the circular RNA and the miRNA. This allows the interaction between RNAs to be discovered in subsequent predictions. The method utilizes the commonalities and influences between other RNA post-transcriptional regulatory mechanisms and RBP binding to improve the accuracy of predicting the interaction between circular RNA and RBP without the need for time-consuming, labor-intensive, and costly biological experiments. Furthermore, the present invention is applicable not only to predicting the interaction between circular RNA and RBP, but also to predicting the interaction between linear RNA and RBP, and has strong versatility.

[0078] Based on the method for predicting the interaction between circular RNA and RBP described in Example 1, Example 2 of the present invention provides a prediction system for the interaction between circular RNA and RBP, comprising:

[0079] The circular RNA feature extraction unit is used to input the circular RNA into the TransRNAm module to obtain the circular RNA feature vector;

[0080] The fusion feature extraction unit is used to input three different miRNAs and circular RNA into three EMLinter modules respectively to obtain the first fusion feature vector, the second fusion feature vector and the third fusion feature vector respectively; each EMLinter module includes an attention branch and a gating branch; in the attention branch, the circular RNA and the miRNA are respectively passed through their corresponding embedding layers and convolution blocks to obtain the first intermediate features of the circular RNA and the first intermediate features of the miRNA, and then respectively passed through the approximate attention mechanism to obtain the second intermediate features of the circular RNA and the second intermediate features of the miRNA; the first intermediate features of the circular RNA and the second intermediate features of the miRNA are respectively passed through the approximate attention mechanism to obtain the second intermediate features of the circular RNA and the second intermediate features of the miRNA; After feature splicing, the first intermediate feature and the second intermediate feature of the miRNA are spliced ​​and input into their corresponding maximum pooling layers respectively to obtain the first target feature of the circular RNA and the first target feature of the miRNA. After splicing, they pass through the fully connected layer to obtain the output feature of the attention branch. In the gating branch, the circular RNA and the miRNA pass through their corresponding embedding layer, BiGRU gating mechanism, convolution block and maximum pooling layer respectively to obtain the second target feature of the circular RNA and the second target feature of the miRNA. After splicing, they pass through the fully connected layer to obtain the output feature of the gating branch. The output features of the attention branch and the output features of the gating branch are spliced ​​to obtain a fused feature vector.

[0081] The prediction unit is used to splice the circular RNA feature vector, the first fusion feature vector, the second fusion feature vector and the third fusion feature vector and input them into the prediction module for binary classification to obtain the prediction result of the interaction between circular RNA and RBP.

[0082] Embodiment 2 of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for predicting the interaction between circular RNA and RBP are implemented.

[0083] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0084] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0085] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0086] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0087] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. A method for predicting the interaction between circular RNA and RBP, characterized in that: include: Three different miRNAs were selected by dividing the miRNAs in the database into three categories using the KMeans clustering method, and selecting one miRNA in each category and padding it to 25 bp; The circular RNA is padded to 22743 bp and input into the TransRNAm module to obtain a circular RNA feature vector; the TransRNAm module includes a Word2Vec model, a Transformer encoding block, and a convolutional neural network connected in sequence; the output features of the Transformer encoding block and the output features of the convolutional neural network are spliced ​​through jump connections to obtain the circular RNA feature vector; The three different miRNAs and circular RNAs were input into three EMLinter modules to obtain the first fusion feature vector, the second fusion feature vector and the third fusion feature vector respectively; Each EMLinter module includes an attention branch and a gating branch; In the attention branch, circular RNA and miRNA are respectively passed through their corresponding embedding layers and convolution blocks to obtain the first intermediate features of circular RNA and the first intermediate features of miRNA, and then passed through the approximate attention mechanism to obtain the second intermediate features of circular RNA and the second intermediate features of miRNA. The first intermediate features of circular RNA and the second intermediate features of miRNA are spliced ​​together and input into their corresponding maximum pooling layers to obtain the first target features of circular RNA and the first target features of miRNA. After splicing, they pass through the fully connected layer to obtain the output features of the attention branch. In the gating branch, circular RNA and miRNA pass through their corresponding embedding layer, BiGRU gating mechanism, convolution block and maximum pooling layer respectively to obtain the second target feature of circular RNA and the second target feature of miRNA. After splicing, they pass through the fully connected layer to obtain the output feature of the gating branch; The output features of the attention branch and the output features of the gate branch are concatenated to obtain a fused feature vector; The circular RNA feature vector, the first fusion feature vector, the second fusion feature vector, and the third fusion feature vector are spliced ​​and input into a prediction module for binary classification to obtain a prediction result of the interaction between circular RNA and RBP; the prediction module includes three linear layers and a softmax activation function connected in sequence.

2. The method for predicting the interaction between circular RNA and RBP according to claim 1, wherein: The three miRNAs selected were hsa-miR-193b-3p, hsa-miR-20b-5p and hsa-miR-376b-3p.

3. The method for predicting the interaction between circular RNA and RBP according to claim 1, wherein: In the attention branch and the gating branch, the convolution block includes three sequentially connected convolution layers, and each convolution layer is connected to a normalization layer.

4. The method for predicting the interaction between circular RNA and RBP according to claim 1, wherein: The loss function for training the TransRNAm module, EMLinter module, and prediction module is: ； in, represents the loss function, represents the total number of samples, represents the i-th sample, represents the true label of the i-th sample, Represents the prediction model composed of TransRNAm module, EMLinter module and prediction module.

5. A prediction system for the interaction between circular RNA and RBP, characterized in that: include: A circular RNA feature extraction unit is used to pad the circular RNA to 22,743 bp and input it into a TransRNAm module to obtain a circular RNA feature vector. The TransRNAm module includes a Word2Vec model, a Transformer encoding block, and a convolutional neural network connected in sequence. The output features of the Transformer encoding block and the output features of the convolutional neural network are spliced ​​via jump connections to obtain a circular RNA feature vector. Three different miRNAs are selected by dividing the miRNAs in the database into three categories using the KMeans clustering method, selecting one miRNA from each category and padding it to 25 bp. A fusion feature extraction unit is used to input three different miRNAs and circular RNA into three EMLinter modules respectively to obtain a first fusion feature vector, a second fusion feature vector, and a third fusion feature vector, respectively. Each EMLinter module includes an attention branch and a gating branch. In the attention branch, the circular RNA and miRNA are respectively passed through their corresponding embedding layers and convolution blocks to obtain the first intermediate features of the circular RNA and the first intermediate features of the miRNA, and then respectively pass through the approximate attention mechanism to obtain the second intermediate features of the circular RNA and the second intermediate features of the miRNA. The first intermediate features of the circular RNA and the second intermediate features of the miRNA are spliced ​​together and then input into their corresponding maximum pooling layers respectively to obtain the first target features of the circular RNA and the first target features of the miRNA. After splicing, the first intermediate features of the circular RNA and the second intermediate features of the miRNA are spliced ​​together and then input into their corresponding maximum pooling layers respectively to obtain the first target features of the circular RNA and the first target features of the miRNA. After splicing, the output features of the attention branch are obtained by passing through the fully connected layer. In the gating branch, circular RNA and miRNA pass through their corresponding embedding layer, BiGRU gating mechanism, convolution block and maximum pooling layer respectively to obtain the second target feature of circular RNA and the second target feature of miRNA. After splicing, they pass through the fully connected layer to obtain the output feature of the gating branch; The output features of the attention branch and the output features of the gate branch are concatenated to obtain a fused feature vector; The prediction unit is used to splice the circular RNA feature vector, the first fusion feature vector, the second fusion feature vector and the third fusion feature vector and input them into the prediction module for binary classification to obtain the prediction result of the interaction between circular RNA and RBP; the prediction module includes three linear layers and a softmax activation function connected in sequence.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for predicting the interaction between circular RNA and RBP according to any one of claims 1 to 4.

Images