CircRNA and miRNA interaction prediction system and method of graph Fourier pulse neural network

Abstract

The invention discloses a circRNA (Ribonucleic Acid) and miRNA (Micro Ribonucleic Acid) interaction prediction system and a circRNA and miRNA interaction prediction method of a graph Fourier pulse neural network. The method comprises the following steps: on the basis of high-throughput sequencing omics data of complex diseases, constructing a heterogeneous biological information network containing drugs, diseases, proteins, circRNA, miRNA and lncRNA; converting the topological features of the entities into a unified feature space by using a graph convolutional network; designing a pulse graph neural network in combination with Fourier coding and a pulse neural network, and extracting a topological structure and high-order semantic features in the network; fusing sequences, topologies and semantic features of circRNA and miRNA through a gate multilayer perceptron to obtain embedding features of circRNA and miRNA; and finally, the interaction of circRNA and miRNA is predicted by adopting a Bayesian classifier. According to the method, heterogeneous biological information is modeled from the perspective of network science, Fourier coding, spiking neurons and graph embedding learning are utilized, the action mechanism of circRNA and miRNA in complex diseases can be disclosed, and the method has good practicability and application prospects in the fields of artificial intelligence, life science, clinical medicine and the like.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary research fields of brain-like computing, artificial intelligence and life sciences, and relates to a system and method for predicting circRNA and miRNA interactions using a graph Fourier spiking neural network. Background Technology

[0002] Circular RNAs (circRNAs) play crucial roles in gene expression, splicing, and transcription. Increasing research indicates a significant link between circRNAs and disease development, their interaction with multiple target genes, and their important role in the occurrence, metastasis, proliferation, and differentiation of cancer cells. These interactions form a complex network of biomolecular interactions. Artificial intelligence technologies for identifying disease-related circRNAs can contribute to understanding the molecular pathology and pathogenesis of diseases, promoting the development of smart and personalized medicine.

[0003] With the development of electronic information and sequencing technologies, massive amounts of omics data have been accumulated in fields such as biomedicine. Predicting the association between miRNAs and diseases in biological experiments requires significant financial and time resources. Therefore, utilizing intelligent computing algorithms to predict the association between circRNAs and diseases has become an urgent problem to be solved. In 2020, He Xin et al. published an article in *Life Sciences* entitled "Research Progress on the Regulatory Role of circRNA-miRNA-mRNA Networks in Cardiovascular Diseases," elucidating that circRNAs, acting as a "sponge" for microRNAs (miRNAs), competitively inhibit the binding of miRNAs to mRNAs. Peng Li et al. disclosed a method, device, and medium for predicting miRNA-circRNA based on multi-source feature extraction in patent application No. 2024109176243; Yang Sen et al. disclosed a method and system for predicting the relationship between circRNA and miRNA using multiple views in patent application No. 2024101678326; Zheng Dawei et al. disclosed a circular RNA molecular marker related to the diagnosis of esophageal squamous cell carcinoma and its application in patent application No. 202511325463X; Li Xiangdong et al. disclosed a circular RNA related to economic traits in chickens and a detection kit in patent application No. 2025112478493.

[0004] The aforementioned methods typically use similarity information between diseases, miRNAs, and circRNAs as input to the algorithm, neglecting the interactions between molecules closely related to disease miRNAs / circRNAs, such as drugs, proteins, and lncRNAs, thus limiting the predictive performance of the computational model. This invention is based on a multi-source heterogeneous biological information network composed of drugs, proteins, diseases, circRNAs, miRNAs, and lncRNAs. It utilizes spiking neural networks and graph convolutional networks to automatically learn the multi-scale topological features of circRNAs and miRNAs, and then uses support vector machines to predict the association information between miRNAs and circRNAs. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a graph Fourier spiking neural network-based system and method for predicting circRNA and miRNA interactions. First, it constructs a multi-source heterogeneous bioinformatics network between miRNAs and circRNAs using biological data such as drugs, proteins, diseases, circRNAs, miRNAs, and lncRNAs from multi-omics data of complex diseases. Then, based on the constructed heterogeneous bioinformatics network, a graph Fourier spiking neural network is used to automatically extract the topological features and semantic information of circRNAs and miRNAs. Finally, a support vector machine is used to predict the association relationships. Specifically, to achieve the above-mentioned objectives, the specific technical solution of this invention includes: 1. A system and method for predicting circRNA and miRNA interactions using a graphical Fourier spiking neural network, characterized in that: Step 1. Obtain datasets and interaction association datasets for drugs, proteins, diseases, circRNAs, miRNAs, and lncRNAs, and construct heterogeneous biomolecular networks of circRNAs and miRNAs. The set of nodes in this heterogeneous biological information network Including drugs, proteins, diseases, circRNA, miRNA, and lncRNA; edge objects This includes the association edges between nodes representing drugs, proteins, diseases, circRNAs, miRNAs, and lncRNAs. The node type mapping function... Edge type mapping function . and These represent the number of node type sets and edge type sets on a heterogeneous biological information network, respectively.

[0006] Step 2. Calculation of node features for drugs, proteins, diseases, circRNAs, miRNAs, and lncRNAs. Calculate the sequence features of each drug separately. and polarity, solubility characteristics Protein sequence characteristics and structural features Disease semantic similarity Similarity to Gaussian kernel function Sequence characteristics of circRNA, miRNA, and lncRNA Similarity features with Gaussian kernel function .

[0007] Step 3. Construct a graph Fourier spiking neural network feature learning model: Utilize graph convolutional networks to transform the topological features of drugs, proteins, diseases, circRNAs, miRNAs, and lncRNAs into a unified feature space. The process is as follows: , It is the embedding feature of the node. These are the weights learned by the model. This involves identifying the characteristics of heterogeneous networks. Secondly, based on the heterogeneous bioinformatics network from step 1 and the node features from step 2, a deep graph Fourier pulse feature model is designed using Fourier encoding and spiking neurons to extract features from multi-source heterogeneous molecular networks. This deep model is used to automatically extract and aggregate the topological features of drugs, proteins, diseases, circRNAs, miRNAs, and lncRNAs in the heterogeneous bioinformatics network. Then, a graph convolution model is used to fuse the node-level and semantic-level topological information in the multi-source heterogeneous molecular network composed of diseases and circRNAs to obtain the features of circRNAs and miRNAs.

[0008] Step 4. Construct a circRNA and miRNA association predictor. Based on the circRNA and miRNA embedding information extracted by heterogeneous biological information networks and graph Fourier spiking neural networks, support vector machines are used to predict the association relationships of disease circRNAs.

[0009] Step 5. Design the objective function for predicting circRNA and miRNA interactions. The loss function for predicting circRNA and miRNA interactions is designed using cross-entropy loss and L1 loss functions.

[0010] Step 6. Model Training and Optimization. The model is trained using datasets of drugs, proteins, diseases, circRNAs, miRNAs, and lncRNAs, along with their interactions. The interaction prediction model training process is then optimized using gradient descent, Dropout, and regularization methods to obtain the optimal parameters, enabling reliable modeling and accurate prediction of the association between circRNAs and miRNAs. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the system structure of the present invention. First, data on drugs, diseases, proteins, circRNAs, miRNAs, and lncRNAs are acquired. Drug sequence features, disease semantic similarity, and Gaussian kernel function similarity are calculated; sequence features and Gaussian kernel function similarity features of miRNAs, lncRNAs, and circRNAs are calculated; and the polarity, solubility, sequence, and structural features of proteins are calculated. Multi-source information is integrated to obtain node features for drugs, diseases, proteins, circRNAs, miRNAs, and lncRNAs. Second, a heterogeneous biological information network containing circRNAs and miRNAs is constructed. Then, a graph Fourier spiking neural network feature learning model is constructed. This model integrates Fourier coding, spiking neural networks, and graph convolution to extract the topological structure and high-order semantic features of drugs, diseases, proteins, circRNAs, miRNAs, and lncRNAs in the network. Next, a circRNA-miRNA association predictor is constructed, and an objective function for model optimization is designed. Through model training and optimization, the prediction of circRNA-miRNA interactions is finally achieved. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0013] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, the method may include the following steps: 1. A system and method for predicting circRNA and miRNA interactions using a graphical Fourier spiking neural network, characterized in that: Step 1. Obtain datasets and interaction association datasets for drugs, diseases, proteins, circRNAs, miRNAs, and lncRNAs, and construct a heterogeneous bioinformatics network for circRNAs and miRNAs. The set of nodes in this heterogeneous biological information network Including drugs, diseases, proteins, circRNA, miRNA, and lncRNA; edge objects This includes association edges between drugs, diseases, proteins, circRNAs, miRNAs, and lncRNAs. The node type mapping function... Edge type mapping function . and These represent the number of node type sets and edge type sets on a heterogeneous biological information network, respectively.

[0014] Step 2. Calculation of node features for drugs, diseases, proteins, circRNAs, miRNAs, and lncRNAs. Calculate the sequence features of each drug. and polarity, solubility characteristics Protein sequence characteristics and structural features Disease semantic similarity Similarity to Gaussian kernel function Sequence characteristics of circRNA, miRNA, and lncRNA Similarity features with Gaussian kernel function Specifically, ① Semantic similarity of diseases The relationships between diseases are described as a directed acyclic graph (DAG), where nodes represent diseases and edges represent the relationships between diseases. In a DAG, diseases... Represented as ,in This includes diseases Including diseases The ancestor node set, It is the edge set connecting these diseases. Therefore, for a certain disease s in G, there is a disease pair with the disease... The semantic contribution value is:

[0015] As a semantic contributor, disease semantic value ,disease and semantic similarity between for: .

[0016] ② Gaussian kernel function similarity of diseases : Disease and Gaussian kernel similarity is defined as The calculation method is as follows: in and It is a disease and The interaction spectrum, as the eigenvectors of the Gaussian kernel. and Representing the first and Row vectors. It is a kernel bandwidth parameter, calculated as follows: Gaussian kernel function similarity of circRNA, miRNA, and lncRNA Gaussian kernel function similarity to disease The calculation method is the same.

[0017] ③The length is L The molecular sequences (drugs, proteins, circRNAs, miRNAs, and lncRNAs) are divided into groups containing... A string of bases k -mer, and then use word embedding algorithms to train drug, protein, circRNA, miRNA and lncRNA. k -mer sequence data yields information rich in contextual patterns. k -mer embedding features, drugs, proteins, circRNAs, miRNAs, and lncRNAs k -mer embedding features are denoted as follows: The polarity and solubility characteristics of a drug are denoted as... The structural features of proteins are denoted as... .

[0018] ④ Integrating multi-source feature information from drugs, diseases, proteins, circRNAs, miRNAs, and lncRNAs, the feature representation of drugs is denoted as... The characteristic representation of a protein is denoted as The characteristics of the disease are denoted as The characteristic representation of circRNA is denoted as The characteristic representation of miRNA is denoted as The characteristic representation of lncRNA is denoted as , This represents a connection operation along a feature channel dimension.

[0019] Step 3. Construct a graph Fourier spiking neural network feature learning model: based on the heterogeneous biological information network in Step 1 and the drug features learned in Step 2. Protein Characterization Disease node characteristics circRNA node characteristics miRNA node characteristics and lncRNA node characteristics We utilize gate mechanisms and graph-gated convolutional networks to extract local topological embedding features of nodes and transform these node features into a unified feature space. These are the weight parameters learned by the model. This is the verified correlation feature matrix in a heterogeneous molecular network. The feature learning process of a Tug-gate convolutional network is as follows: , It is a non-linear activation function. It is the sigmoid function. Then, the extracted heterogeneous biological information network features are fed into the Fourier position information encoding layer to introduce position information into the nodes. This is the position index of the node in the position code table. This represents the dimension of the hidden layer in the Fourier position feature layer. It contains the node's position information. The calculation method is as follows:

[0020] After feature mapping transformation and location information encoding, drug characteristics are obtained. Protein characteristics Disease characteristics circRNA characteristics miRNA characteristics and lncRNA characteristics Then, the topological and semantic information of nodes in the heterogeneous biological information network is extracted using a leak-integral-discharge (LIF) type spiking neural network. LIF neuron computation includes pulse coding, charging, discharging, and membrane potential reset processes. The input data of the pulse encoder is , is the input pulse data of the fully connected layer at time t. It is the pulse value before the emphase. The resolution of the pulse code is T. It is a node i The first j There are several features. The encoded pulse probability is positively correlated with the importance of node features. In heterogeneous biological information networks, the Bernoulli distribution encoder fires pulses with probability p. To encode node features, The value is 0 or 1. The calculation method is as follows: We designed a pulsed convolutional technique for extracting node features in biological information networks using leaky-integral-fire (LIF) type spiking neurons. First, the neurons are charged, then pulsed discharge and membrane potential reset. The charging phase involves using trainable weights to transmit the pulsed input as an input voltage to the LIF neuron through a fully connected layer.

[0021] express Cumulative change pulse value Used to represent nodes The membrane potential is updated as follows:

[0022] In the transmission of neural information in bioinformatics networks, there are activation, firing, and reset phases. LIF neurons first charge, then pulse firing and membrane potential reset. During the charging phase, trainable weights are used to transmit the pulse input as an input voltage to the LIF neuron. This represents the membrane potential acquired by the node at step t. The total membrane potential of nodes in a heterogeneous biological information network is updated as follows: When the accumulated voltage of a LIF neuron exceeds a threshold, it is triggered and activated, transmitting a pulse signal. The pulse firing mechanism can be represented as follows: . Here, is the firing threshold, is the membrane potential threshold of the spiking neuron, and O is the output spike. After pulse propagation, the membrane potential of the LIF neuron is reset to: Then, the neuron enters a resting state before the next iteration begins. Each fully connected layer is followed by a LIF layer. Drug features are obtained through pulsed convolution. Protein characteristics Disease characteristics circRNA characteristics miRNA characteristics and lncRNA characteristics .

[0023] Step 4. Construct a circRNA and miRNA association predictor. Based on the topological embedding features of the disease extracted in Step 3: drug features. Protein characteristics Disease characteristics circRNA characteristics miRNA characteristics and lncRNA characteristics circRNA topological embedding information Then, a low-dimensional embedding representation of circRNA and miRNA nodes is obtained by fusing them using a gated multilayer perceptron. The nonlinear computation and update process of feature dimensionality reduction is as follows: , It is a non-linear activation function. It is the sigmoid function. Finally, support vector machines are used to predict the probability of the relationship between miRNAs and circRNAs. This enables reliable modeling and accurate prediction of whether there is a relationship between circRNA and miRNA.

[0024] Step 5. Design the objective function for predicting circRNA and miRNA interactions. Train and optimize the model based on cross-entropy loss and L1 loss function. A smooth L1 loss function is introduced. Introducing a term into the entropy loss The cross-entropy loss calculation is used to evaluate the loss between the output of the miRNA-circRNA association prediction model and the actual data. The calculation process of the loss function is as follows: , It is an adaptive hyperparameter. It is a tag for association between circRNA and miRNA.

[0025] Step 6. Training and Optimization of the circRNA and miRNA Interaction Prediction Model. The model is trained based on datasets of drugs, proteins, diseases, circRNAs, miRNAs, and lncRNAs, as well as interaction datasets, to perform nonlinear transformations on the input heterogeneous biological information network data to fit the output. The model training is optimized using gradient descent, Dropout, and regularization methods to obtain the optimal parameters of the graphical Fourier spiking neural network model.

Claims

1. A system and method for predicting circRNA and miRNA interactions using a graphical Fourier spiking neural network, characterized in that: Step 1. Construct a heterogeneous biological information network of drugs, diseases, proteins, circRNAs, miRNAs, and lncRNAs. The set of nodes in this heterogeneous network Nodes include drugs, diseases, proteins, circRNAs, miRNAs, and lncRNAs, and edge objects. This includes the association edges between nodes such as drugs, diseases, proteins, circRNAs, miRNAs, and lncRNAs; Step 2. Feature calculation for drug, disease, protein, circRNA, miRNA and lncRNA nodes: Calculate the sequence features and polarity and solubility features of drugs, the sequence features and structural features of proteins, the semantic similarity and Gaussian kernel function similarity of diseases, and the sequence features and Gaussian kernel function similarity features of lncRNA, circRNA and miRNA respectively. Step 3. Construct a graph Fourier spiking neural network feature learning model: Based on the multi-layer bioinformatics network in Step 1 and the node features in Step 2, a graph convolutional network is used to transform the topological features of drugs, proteins, diseases, circRNAs, miRNAs, and lncRNAs into a unified feature space. Then, a spiking graph neural network model for the propagation and aggregation of node information in a multi-layer heterogeneous molecular network is designed using Fourier coding, spiking neural networks, and random walk algorithms to obtain the topological semantic features of drugs, diseases, proteins, circRNAs, miRNAs, and lncRNAs. Finally, a gated multilayer perceptron fusion method is used to fuse the neighboring nodes and edge information of circRNAs and miRNAs to obtain the embedding features of circRNAs and miRNAs. Step 4. Construct a predictor for the association between circRNA and miRNA. Based on the embedding features extracted from the multilayer heterogeneous biological information network in Step 1 by the pulse graph neural network model in Step 3, a Bayesian classifier is used to infer and predict the association between circRNA and miRNA. Step 5. Design the objective function for predicting circRNA and miRNA interactions. Optimize the performance of the circRNA and miRNA interaction prediction system and method; Step 6. Model Training and Optimization. The model is trained using datasets of drugs, diseases, proteins, circRNAs, miRNAs, and lncRNAs. Dropout and regularization methods are used to optimize the model training process, obtaining the optimal parameter settings for the circRNA and miRNA interaction prediction model.

2. The circRNA and miRNA interaction prediction system and method based on a graph Fourier spiking neural network according to claim 1, characterized in step 1: acquiring datasets and interaction association datasets of drugs, diseases, proteins, circRNAs, miRNAs, and lncRNAs, and constructing a multi-source heterogeneous biological information network. .

3. The circRNA and miRNA interaction prediction system and method based on a graph Fourier spiking neural network according to claim 1, step 2 is characterized in that: circRNA and miRNA node features are extracted and updated using drug, disease, protein, and lncRNA features. Sequence and structural features of drugs and proteins, semantic similarity and Gaussian kernel function similarity of diseases, sequence features and Gaussian kernel function similarity features of circRNA, miRNA, and lncRNA are calculated. Then, multi-source information is fused to obtain the features of drugs, proteins, diseases, circRNA, miRNA, and lncRNA. , , , , and .

4. The circRNA and miRNA interaction prediction system and method based on a graph Fourier spiking neural network according to claim 1, step 3 is characterized in that: the topological features of drugs, proteins, diseases, circRNAs, miRNAs and lncRNAs are transformed into a unified feature space using a graph convolutional network; a graph Fourier spiking neural network is designed to extract the topological embedding features of drugs, proteins, diseases, circRNAs, miRNAs and lncRNAs; the neighbor nodes and edge information of circRNAs and miRNAs are modeled using graph convolution; and the embedding features of circRNAs and miRNAs are obtained by fusion based on a gated multilayer perceptron.

5. The circRNA and miRNA interaction prediction system and method based on a graph Fourier spiking neural network according to claim 1, wherein step 4 is characterized in that: based on the circRNA and miRNA embedding features extracted by the deep graph Fourier spiking neural network, a support vector machine is used to predict whether there is an association between circRNA and miRNA.

6. The circRNA and miRNA interaction prediction system and method based on a graph Fourier spiking neural network according to claim 1, wherein step 5 is characterized in that: for the circRNA and miRNA prediction task, a loss function for circRNA and miRNA interaction prediction is designed using cross-entropy loss and L1 loss function.

7. The circRNA and miRNA interaction prediction system and method based on graph Fourier spiking neural network according to claim 1, step 6 is characterized in that: the model is trained by data sets of drugs, proteins, diseases, circRNAs, miRNAs and lncRNAs, and various gradient descent algorithms are used to optimize the circRNA and miRNA interaction prediction model based on Dropout and regularization methods to obtain the optimal parameter settings, so as to achieve reliable modeling and accurate prediction of whether there is a correlation between circRNAs and miRNAs.

Images