A method and system for mining lncRNA-miRNA association

Abstract

This invention relates to a method and system for mining lncRNA-miRNA associations. The method includes: constructing a heterogeneous biological information network for lncRNA-miRNA interactions; performing feature initialization to form an initial feature matrix; constructing a graph spiking neural network embedding representation model; constructing an lncRNA-miRNA association predictor based on the embedding representations of connected lncRNA-miRNA node pairs; and predicting the potential interaction strength between unconnected lncRNA-miRNA node pairs using the lncRNA-miRNA association predictor. By utilizing a graph spiking neural network to mimic the dynamic information processing mechanism of biological neural systems, it adaptively learns the higher-order topological features and dynamic contextual representations of nodes, thereby more accurately and efficiently predicting potential lncRNA-miRNA interaction relationships and providing high-quality candidate targets for subsequent biological validation and clinical research.

Description

Technical Field

[0001] This invention belongs to the field of music recommendation technology, and in particular relates to a method and system for mining the association between lncRNA and miRNA. Background Technology

[0002] Long non-coding RNAs (lncRNAs) and microRNAs (miRNAs) are two important classes of non-coding RNA molecules that play crucial roles in life processes such as epigenetic regulation of gene expression, transcription, and post-transcriptional regulation. Studies have shown that lncRNAs can act as competitive endogenous RNAs (ceRNAs), binding to miRNAs through a sponge-like binding mechanism, thereby relieving the inhibitory effect of miRNAs on their target genes and forming a complex lncRNA-miRNA-mRNA regulatory network. This network is closely related to the occurrence and development of various complex diseases, including cancer, cardiovascular diseases, and neurological disorders. Therefore, systematically exploring and predicting the potential associations between lncRNAs and miRNAs is of great scientific significance for a deeper understanding of disease molecular mechanisms and the discovery of new biomarkers and drug targets.

[0003] With the development of high-throughput sequencing and bioinformatics technologies, the biomedical field has accumulated massive amounts of multi-omics data. However, verifying lncRNA-miRNA interactions using traditional wet experimental methods is time-consuming, labor-intensive, and costly. Utilizing efficient and reliable intelligent computing algorithms to predict potential associations from existing data has become an important research direction in bioinformatics.

[0004] Existing computational methods can be mainly categorized into similarity-based methods, machine learning-based methods, and network-based methods. For example, a machine learning method for predicting lncRNA-miRNA interactions has been proposed using XGBoost. This method focuses on the binding free energy, binding site score, and unpaired probability of RNA folding at potential binding sites between lncRNAs and miRNAs. A multimodal contrastive representation learning model combining Graph Convolutional Networks (GCN) and Transformer is proposed to capture local neighborhood structural features and long-distance dependencies, achieving collaborative modeling of structural and semantic information for predicting lncRNA-miRNA interactions. Furthermore, patent application number (2024119050504) discloses a method for screening thyroid cancer co-diagnostic biomarkers based on plasma exosomal lncRNAs and miRNAs, using a multi-gene regression model for prediction. Patent application number (2024110926293) discloses a method for predicting lncRNA-miRNA interactions based on sequence complementarity site information, using RF, XGBoost, SVM, Decisiontree, and LightGBM models for prediction. A multi-channel attention mechanism-based method for predicting lncRNA-miRNA association is disclosed in patent application number (2024112092351), which uses graph convolutional networks and Transformers for prediction.

[0005] Despite the progress made by the above methods, some limitations still exist: First, many methods fail to fully integrate other related biological entities such as diseases and proteins to construct a multi-source heterogeneous information network that can comprehensively reflect the biological regulatory background, resulting in insufficient information utilization; Second, when dealing with the sparsity, heterogeneity, and complex spatiotemporal dynamics of biological networks, the node information transmission mechanism (based on continuous numerical activation) of traditional graph neural networks differs fundamentally from the discrete pulse signal transmission between biological neurons, leaving room for improvement in computational efficiency and biological interpretability. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method and system for mining the association between lncRNA and miRNA.

[0007] The technical solution adopted in this invention is:

[0008] Firstly, a method for mining associations between lncRNAs and miRNAs is provided, including:

[0009] A heterogeneous biological information network for lncRNA-miRNA interactions was constructed. The topology of the heterogeneous biological information network includes a set of nodes and a set of edges. The set of nodes includes lncRNAs, miRNAs, diseases, and proteins. The set of edges includes lncRNA-miRNA associations, lncRNA-disease associations, miRNA-disease associations, protein-lncRNA interactions, and protein-miRNA interactions.

[0010] The nodes of the heterogeneous biological information network are initialized with features to form an initial feature matrix;

[0011] Based on the topology and initial feature matrix of the heterogeneous biological information network, a graph spiking neural network embedding representation model is constructed.

[0012] Based on the graph spiking neural network embedding representation model, the lncRNA embedding representation and miRNA embedding representation of the connected lncRNA-miRNA node pairs are obtained, and the lncRNA-miRNA association predictor is constructed based on the lncRNA embedding representation and miRNA embedding representation.

[0013] When it is necessary to predict unconnected lncRNA-miRNA node pairs, the potential interaction strength between unconnected lncRNA-miRNA node pairs is predicted based on the lncRNA-miRNA association predictor, and the prediction association confidence is obtained.

[0014] Furthermore, a heterogeneous biological information network for lncRNA-miRNA interactions is constructed, including:

[0015] Obtain known biomolecular interaction datasets of lncRNAs, miRNAs, diseases, and proteins from publicly available databases;

[0016] A heterogeneous biological information network for lncRNA-miRNA interactions was constructed based on known biomolecular interaction datasets. , Represents a set of nodes. Represents the set of edges and the set of nodes. This includes lncRNA, miRNA, diseases, and proteins. lncRNA is represented by 'l', miRNA by 'm', diseases by 'd', and proteins by 'p', along with edge sets. This includes lncRNA-miRNA associations, lncRNA-disease associations, miRNA-disease associations, protein-lncRNA interactions, and protein-miRNA interactions. lncRNA-miRNA associations are represented as... lncRNA-disease association is represented as miRNA-disease association is represented as Protein-lncRNA interactions are represented as Protein-miRNA interactions are represented as .

[0017] Furthermore, the nodes of the heterogeneous biological information network are initialized with features to form an initial feature matrix, including:

[0018] By integrating multiple known prior bioinformatics knowledge, multi-view initial feature representations are constructed for each node in a heterogeneous biological information network;

[0019] An initial feature matrix is ​​formed based on the multi-view initial feature representations of all nodes.

[0020] Furthermore, the graph spiking neural network embedding representation model updates and transmits the state of each node in a heterogeneous biological information network by simulating the spiking and information accumulation mechanism of biological neurons, adaptively aggregating multi-hop topological and semantic information from different types of neighboring nodes, and generating an embedding representation between lncRNA-miRNA node pairs containing spatiotemporal dynamic characteristics.

[0021] Furthermore, the methods also include:

[0022] A multi-task joint loss function combining cross-entropy loss and graph structure reconstruction loss was designed for the lncRNA-miRNA association predictor. The multi-task joint loss function is composed of a weighted sum of the binary cross-entropy loss of association prediction and the graph structure reconstruction loss.

[0023] Furthermore, the methods also include:

[0024] An algorithm combining backpropagation and temporal backpropagation is adopted, and a rectangular function is used for gradient update. The Adam optimizer is used to minimize the multi-task joint loss function to optimize the parameters of the lncRNA-miRNA association predictor.

[0025] Secondly, a lncRNA-miRNA association mining system is provided, including:

[0026] The network construction module is used to construct a heterogeneous biological information network for lncRNA-miRNA interactions. The topology of the heterogeneous biological information network includes a set of nodes and a set of edges. The set of nodes includes lncRNAs, miRNAs, diseases, and proteins. The set of edges includes lncRNA-miRNA associations, lncRNA-disease associations, miRNA-disease associations, protein-lncRNA interactions, and protein-miRNA interactions.

[0027] The feature initialization module is used to initialize the features of nodes in a heterogeneous biological information network to form an initial feature matrix.

[0028] The model building module is used to construct a graph spiking neural network embedding representation model based on the topology and initial feature matrix of the heterogeneous biological information network.

[0029] The association predictor construction module is used to obtain the lncRNA embedding representation and miRNA embedding representation of the connected lncRNA-miRNA node pairs based on the graph spiking neural network embedding representation model, and to construct the lncRNA-miRNA association predictor based on the lncRNA embedding representation and miRNA embedding representation.

[0030] The prediction module is used to predict the potential interaction strength between unconnected lncRNA-miRNA node pairs based on the lncRNA-miRNA association predictor when it is necessary to predict unconnected lncRNA-miRNA node pairs, and obtain the prediction association confidence.

[0031] Furthermore, the system also includes:

[0032] The loss function design module is used to design a multi-task joint loss function for the lncRNA-miRNA association predictor, which combines cross-entropy loss and graph structure reconstruction loss. The multi-task joint loss function is composed of a weighted sum of the binary cross-entropy loss of association prediction and the graph structure reconstruction loss.

[0033] Furthermore, the system also includes:

[0034] The association predictor optimization module is used to optimize the parameters of the lncRNA-miRNA association predictor by employing an algorithm that combines backpropagation and temporal backpropagation, using a rectangular function for gradient updates, and minimizing the multi-task joint loss function using the Adam optimizer.

[0035] The beneficial effects achieved by this invention are as follows:

[0036] A heterogeneous biological information network (BIN) for lncRNA-miRNA interactions is constructed. The topological network of this network comprises a set of nodes and an edge set. The node set includes lncRNAs, miRNAs, diseases, and proteins, while the edge set includes lncRNA-miRNA associations, lncRNA-disease associations, miRNA-disease associations, protein-lncRNA interactions, and protein-miRNA interactions. Features of the nodes in the HCN are initialized to form an initial feature matrix. Based on the HCN topological network and the initial feature matrix, a graph spiking neural network (PSN) embedding representation model is constructed. The PSN embedding representation model is used to obtain the lncRNA and miRNA embedding representations of connected lncRNA-miRNA node pairs. Based on these representations, an lncRNA-miRNA association predictor is constructed. When predicting unconnected lncRNA-miRNA node pairs, the potential interaction strength between these pairs is predicted using the PSN association predictor, resulting in a predicted association confidence score. By utilizing graph spiking neural networks to mimic the dynamic information processing mechanism of biological nervous systems, we can adaptively learn the high-order topological features and dynamic contextual representations of nodes, thereby more accurately and efficiently predicting potential lncRNA-miRNA interactions and providing high-quality candidate targets for subsequent biological validation and clinical research. Attached Figure Description

[0037] Figure 1 This is the first flowchart of the lncRNA and miRNA association mining method of the present invention;

[0038] Figure 2 This is the second flowchart of the lncRNA and miRNA association mining method of the present invention;

[0039] Figure 3 This is the first structural diagram of the lncRNA and miRNA association mining system of the present invention;

[0040] Figure 4 This is the second structural diagram of the lncRNA and miRNA association mining system of the present invention. Detailed Implementation

[0041] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0042] like Figure 1 As shown, this embodiment of the invention provides a method for mining the association between lncRNA and miRNA, including:

[0043] 101. Construct a heterogeneous biological information network for lncRNA-miRNA interactions;

[0044] In this embodiment, a dataset of known biomolecular interactions of lncRNA, miRNA, diseases, and proteins is obtained from a public database;

[0045] A heterogeneous biological information network for lncRNA-miRNA interactions was constructed based on known biomolecular interaction datasets. , Represents a set of nodes. Represents the set of edges and the set of nodes. This includes lncRNA, miRNA, diseases, and proteins. lncRNA is represented by 'l', miRNA by 'm', diseases by 'd', and proteins by 'p', along with edge sets. This includes lncRNA-miRNA associations, lncRNA-disease associations, miRNA-disease associations, protein-lncRNA interactions, and protein-miRNA interactions. lncRNA-miRNA associations are represented as... lncRNA-disease association is represented as miRNA-disease association is represented as Protein-lncRNA interactions are represented as Protein-miRNA interactions are represented as ;

[0046] Network structure is mapped through node type functions and edge type mapping function Provide a standardized description, in which A collection of node types This is a set of edge types; this heterogeneous biological information network integrates multi-dimensional associations, providing a rich topological foundation for association mining.

[0047] 102. Initialize the features of the nodes in the heterogeneous biological information network to form an initial feature matrix;

[0048] In this embodiment, multiple known prior bioinformatics knowledge is integrated to construct multi-view initial feature representations for each node in the heterogeneous biological information network; an initial feature matrix is ​​formed based on the multi-view initial feature representations of all nodes.

[0049] Feature fusion of lncRNA and miRNA nodes is based on shared disease association calculations; functional similarity between the two molecules (taking lncRNA as an example). Through its associated disease set and Measured by semantic similarity;

[0050] Specifically, diseases are first calculated based on the Medical Subject Headings (MeSH) ontology. and semantic similarity The formula for calculating functional similarity is:

[0051] ;

[0052] in, It is based on the MeSH ontology to calculate disease semantic similarity; sequence features are obtained by extracting the frequency of fixed-length nucleotide strings (k-mers). For example, for a sequence, its k-mer frequencies constitute a feature vector. Gaussian interaction spectral kernel similarity is based on molecular-disease association spectral vectors (e.g., The correlation vector LD i ) calculation, defined as:

[0053] ;

[0054] Among them, bandwidth parameter For disease nodes, the features are combined with a MeSH-based semantic similarity matrix. Similarity to Gaussian kernels based on correlation spectra The features of protein nodes are derived from the physical properties of their amino acid sequences (e.g., amino acid composition) or pre-trained sequence encoding (e.g., ProtBERT, a protein language model based on a transformer architecture specifically designed to predict key properties of peptides); all calculated feature matrices (such as FS, GS, ... After standardization (e.g., Min-Max Scaling), the feature vectors corresponding to each node are concatenated and fused to form the initial feature matrix of the network. , where |v| is the total number of nodes and FF is the total dimension of the fused features, thus providing a dense vector representation rich in biological semantics for each node.

[0055] 103. Based on the topology and initial feature matrix of the heterogeneous biological information network, a graph spiking neural network embedding representation model is constructed.

[0056] In this embodiment, the core is to construct a graph spiking neural network embedding representation model that simulates the dynamics of biological neurons; a leakage current integrated discharge model is used as the basic computational unit, and the membrane potential is... The dynamic update is determined by the following differential equation:

[0057] ;

[0058] At discrete time step t, the input current received by node v is... It consists of two parts: The formula aggregates data from neighbor nodes of different types. The pulse signal of the previous time step (Learnable weight matrix specific to the relationship) (Perform transformation and aggregation) and the initial characteristics of the nodes themselves. (Transformation is performed using a learnable matrix B);

[0059] The model simulates the spatiotemporal propagation and dynamics of pulse signals in the network by iteratively updating T time steps.

[0060] Ultimately, the pulse firing sequence for each node Encoded as a low-dimensional embedding vector , where β is the time decay factor, ultimately generating low-dimensional embedding representations of lncRNA and miRNA containing complex topological and dynamic characteristics.

[0061] The graph spiking neural network embedding representation model updates and transmits the state of each node in a heterogeneous biological information network by simulating the spiking and information accumulation mechanism of biological neurons. It adaptively aggregates multi-hop topological and semantic information from different types of neighboring nodes to generate an embedding representation between lncRNA-miRNA node pairs that contains spatiotemporal dynamic characteristics.

[0062] 104. Based on the graph spiking neural network embedding representation model, obtain the lncRNA embedding representation and miRNA embedding representation of the connected lncRNA-miRNA node pairs, and construct the lncRNA-miRNA association predictor based on the lncRNA embedding representation and miRNA embedding representation.

[0063] In this embodiment, the role of the lncRNA-miRNA association predictor is to construct a bilinear decoder to predict the association probability between lncRNA and miRNA after obtaining their embedding representations.

[0064] The main implementation process is as follows:

[0065] For each pair of lncRNA nodes l (embedding representation as) ) and miRNA node m (embedding representation as Calculate an association score; the core formula is a bilinear transformation:

[0066] ;

[0067] in, It is a learnable parameter matrix specifically designed to capture and model the complex patterns of interactions between two specific types of nodes: lncRNA and miRNA. It allows the model to learn two interaction weights embedded in different dimensions, making it more expressive than simple dot products (where M is the identity matrix) or spliced ​​feedforward networks. It is a learnable global bias term;

[0068] Subsequently, the score was... Activation function via sigmoid Mapped to the (0, 1) interval, used as the predicted association confidence score:

[0069] ;

[0070] The lncRNA-miRNA association predictor can efficiently and directly assess the potential interaction strength between any previously unconnected lncRNA-miRNA node pairs in a network.

[0071] 105. When it is necessary to predict unconnected lncRNA-miRNA node pairs, the potential interaction strength between unconnected lncRNA-miRNA node pairs is predicted based on the lncRNA-miRNA association predictor to obtain the prediction association confidence.

[0072] In this embodiment, based on the description of the prediction implementation process of the lncRNA-miRNA association predictor in step 104, the prediction association confidence between unconnected lncRNA-miRNA node pairs is finally calculated.

[0073] The beneficial effects achieved by the embodiments of the present invention are as follows:

[0074] A heterogeneous biological information network (BIN) for lncRNA-miRNA interactions is constructed. The topological network of this network comprises a set of nodes and an edge set. The node set includes lncRNAs, miRNAs, diseases, and proteins, while the edge set includes lncRNA-miRNA associations, lncRNA-disease associations, miRNA-disease associations, protein-lncRNA interactions, and protein-miRNA interactions. Features of the nodes in the HCN are initialized to form an initial feature matrix. Based on the HCN topological network and the initial feature matrix, a graph spiking neural network (PSN) embedding representation model is constructed. The PSN embedding representation model is used to obtain the lncRNA and miRNA embedding representations of connected lncRNA-miRNA node pairs. Based on these representations, an lncRNA-miRNA association predictor is constructed. When predicting unconnected lncRNA-miRNA node pairs, the potential interaction strength between these pairs is predicted using the PSN association predictor, resulting in a predicted association confidence score. By utilizing graph spiking neural networks to mimic the dynamic information processing mechanism of biological nervous systems, we can adaptively learn the high-order topological features and dynamic contextual representations of nodes, thereby more accurately and efficiently predicting potential lncRNA-miRNA interactions and providing high-quality candidate targets for subsequent biological validation and clinical research.

[0075] It should be noted that, in the above Figure 1 The illustrated examples only describe how to construct a lncRNA-miRNA association predictor to predict the association confidence between unconnected lncRNA-miRNA node pairs. However, the prediction accuracy of the constructed lncRNA-miRNA association predictor still needs to be optimized, such as... Figure 2 As shown, it specifically includes:

[0076] 201. Construct a heterogeneous biological information network for lncRNA-miRNA interactions;

[0077] 202. Initialize the features of the nodes in the heterogeneous biological information network to form an initial feature matrix;

[0078] 203. Based on the topology and initial feature matrix of the heterogeneous biological information network, a graph spiking neural network embedding representation model is constructed.

[0079] 204. Based on the graph spiking neural network embedding representation model, the lncRNA embedding representation and miRNA embedding representation of the connected lncRNA-miRNA node pairs are obtained. Based on the lncRNA embedding representation and miRNA embedding representation, the lncRNA-miRNA association predictor is constructed.

[0080] 205. A multi-task joint loss function combining cross-entropy loss and graph structure reconstruction loss was designed for the lncRNA-miRNA association predictor. The multi-task joint loss function is composed of a weighted sum of the binary cross-entropy loss of association prediction and the graph structure reconstruction loss.

[0081] In this embodiment, in order to simultaneously optimize the accuracy of association prediction and maintain the ability of the learned node embeddings to represent the original network structure, a multi-task joint loss function is designed; it consists of two parts: the first part is the binary cross-entropy loss for association prediction. The calculation method is as follows:

[0082] ;

[0083] Where Ω is the set of training sample pairs, It's a real label. It is the predicted association confidence level calculated in step 104;

[0084] The second part is the graph structure reconstruction loss. The calculation method is as follows:

[0085] ;

[0086] The aim is to encourage the learned full-node embedding matrix H to be able to approximately reconstruct the adjacency matrix corresponding to various relations r in the original network through inner product. ;

[0087] The joint loss function for multiple tasks is , where λ is a hyperparameter used to balance the two weights; the multi-task joint loss function ensures that the embeddings learned by the lncRNA-miRNA association predictor are both suitable for downstream prediction tasks and faithful to the topology of the original heterogeneous network.

[0088] 206. An algorithm combining backpropagation and temporal backpropagation is adopted, and a rectangular function is used for gradient update. The Adam optimizer is used to minimize the multi-task joint loss function to optimize the parameters of the lncRNA-miRNA association predictor.

[0089] In this embodiment, all known lncRNA-miRNA associations are used as positive samples, and unobserved lncRNA-miRNA node pairs are randomly sampled as negative samples, together forming the training set. The lncRNA-miRNA association predictor training adopts an algorithm combining backpropagation and temporal backpropagation, and uses alternative gradient strategies such as rectangular functions to solve the problem of non-differentiability of the firing function of spiking neurons, thereby achieving effective gradient updates. The joint loss function L is minimized by the Adam optimizer, and strategies such as Dropout, weight decay (L2 regularization), and early stopping are used in combination to prevent overfitting.

[0090] After the lncRNA-miRNA association predictor training converges, the association confidence is calculated for any unknown lncRNA-miRNA node pair. By sorting and ranking, we can ultimately output high-confidence potential interactions for subsequent biological experiments to verify.

[0091] 207. When it is necessary to predict unconnected lncRNA-miRNA node pairs, the potential interaction strength between unconnected lncRNA-miRNA node pairs is predicted based on the lncRNA-miRNA association predictor to obtain the prediction association confidence.

[0092] In the above Figure 2 In the illustrated embodiment, optimized training of the lncRNA-miRNA association predictor was implemented, making the calculation of the potential interaction strength of unknown lncRNA-miRNA node pairs more accurate.

[0093] like Figure 3 As shown, this embodiment of the invention provides a lncRNA and miRNA association mining system, comprising:

[0094] Network construction module 301 is used to construct a heterogeneous biological information network of lncRNA-miRNA interactions. The topology of the heterogeneous biological information network includes a set of nodes and a set of edges. The set of nodes includes lncRNA, miRNA, disease and protein. The set of edges includes lncRNA-miRNA association, lncRNA-disease association, miRNA-disease association, protein-lncRNA interaction and protein-miRNA interaction.

[0095] The feature initialization module 302 is used to initialize the features of the nodes in the heterogeneous biological information network to form an initial feature matrix.

[0096] The model building module 303 is used to construct a graph spiking neural network embedding representation model based on the topological network and initial feature matrix of the heterogeneous biological information network.

[0097] The association predictor construction module 304 is used to obtain the lncRNA embedding representation and miRNA embedding representation of the connected lncRNA-miRNA node pairs according to the graph spiking neural network embedding representation model, and to construct the lncRNA-miRNA association predictor based on the lncRNA embedding representation and miRNA embedding representation.

[0098] The prediction module 305 is used to predict the potential interaction strength between unconnected lncRNA-miRNA node pairs based on the lncRNA-miRNA association predictor when it is necessary to predict unconnected lncRNA-miRNA node pairs, and obtain the prediction association confidence.

[0099] Based on the above Figure 3 The embodiments shown are preferred embodiments of the present invention, such as... Figure 4 As shown, it also includes:

[0100] The loss function design module 401 is used to design a multi-task joint loss function that combines cross-entropy loss and graph structure reconstruction loss for the lncRNA-miRNA association predictor. The multi-task joint loss function is composed of a weighted sum of the binary cross-entropy loss of association prediction and the graph structure reconstruction loss.

[0101] The association predictor optimization module 402 is used to optimize the parameters of the lncRNA-miRNA association predictor by employing an algorithm that combines backpropagation and temporal backpropagation, using a rectangular function for gradient updates, and minimizing the multi-task joint loss function using the Adam optimizer.

[0102] The beneficial effects achieved by the embodiments of the present invention are as follows:

[0103] The lncRNA and miRNA association mining system can use graph spiking neural networks to mimic the dynamic information processing mechanism of biological neural systems, adaptively learning the high-order topological features and dynamic contextual representations of nodes, thereby more accurately and efficiently predicting potential lncRNA-miRNA interaction relationships and providing high-quality candidate targets for subsequent biological validation and clinical research.

[0104] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0105] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0106] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0107] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0108] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A method for mining associations between lncRNA and miRNA, characterized in that, include: A heterogeneous biological information network for lncRNA-miRNA interactions is constructed. The topology of the heterogeneous biological information network includes a set of nodes and a set of edges. The set of nodes includes lncRNAs, miRNAs, diseases, and proteins. The set of edges includes lncRNA-miRNA associations, lncRNA-disease associations, miRNA-disease associations, protein-lncRNA interactions, and protein-miRNA interactions. The nodes of the heterogeneous biological information network are initialized with features to form an initial feature matrix; Based on the topology of the heterogeneous biological information network and the initial feature matrix, a graph spiking neural network embedding representation model is constructed. Based on the graph spiking neural network embedding representation model, the lncRNA embedding representation and miRNA embedding representation of the connected lncRNA-miRNA node pairs are obtained, and a lncRNA-miRNA association predictor is constructed based on the lncRNA embedding representation and miRNA embedding representation. When it is necessary to predict unconnected lncRNA-miRNA node pairs, the potential interaction strength between the unconnected lncRNA-miRNA node pairs is predicted based on the lncRNA-miRNA association predictor to obtain the prediction association confidence.

2. The method for mining the association between lncRNA and miRNA according to claim 1, characterized in that, The construction of a heterogeneous biological information network for lncRNA-miRNA interactions includes: Obtain known biomolecular interaction datasets of lncRNAs, miRNAs, diseases, and proteins from publicly available databases; A heterogeneous biological information network for lncRNA-miRNA interactions was constructed based on the known biomolecular interaction dataset. The Represents a set of nodes, the Let the set of edges and the set of nodes be represented. This includes lncRNAs, miRNAs, diseases, and proteins, where lncRNAs are represented by 'l', miRNAs by 'm', diseases by 'd', and proteins by 'p', and the edge set... This includes lncRNA-miRNA associations, lncRNA-disease associations, miRNA-disease associations, protein-lncRNA interactions, and protein-miRNA interactions. The lncRNA-miRNA associations are represented as... The lncRNA-disease association is represented as follows: The miRNA-disease association is represented as follows: The protein-lncRNA interaction is represented as The protein-miRNA interaction is represented as .

3. The method for mining the association between lncRNA and miRNA according to claim 1, characterized in that, The initialization of features for the nodes of the heterogeneous biological information network to form an initial feature matrix includes: By integrating multiple known prior bioinformatics knowledge, multi-view initial feature representations are constructed for each node in the heterogeneous biological information network; An initial feature matrix is ​​formed based on the multi-view initial feature representations of all nodes.

4. The method for mining the association between lncRNA and miRNA according to claim 1, characterized in that, The graph spiking neural network embedding representation model updates and transmits the state of each node in the heterogeneous biological information network by simulating the spiking and information accumulation mechanism of biological neurons. It adaptively aggregates multi-hop topological and semantic information from different types of neighboring nodes to generate an embedding representation between lncRNA-miRNA node pairs that contains spatiotemporal dynamic characteristics.

5. The method for mining the association between lncRNA and miRNA according to claim 1, characterized in that, The method further includes: A multi-task joint loss function combining cross-entropy loss and graph structure reconstruction loss is designed for the lncRNA-miRNA association predictor. The multi-task joint loss function is composed of a weighted sum of the binary cross-entropy loss of association prediction and the graph structure reconstruction loss.

6. The method for mining the association between lncRNA and miRNA according to claim 5, characterized in that, The method further includes: An algorithm combining backpropagation and temporal backpropagation is employed, with gradient updates using a rectangular function and the Adam optimizer used to minimize the multi-task joint loss function to optimize the parameters of the lncRNA-miRNA association predictor.

7. A lncRNA and miRNA association mining system, characterized in that, include: A network construction module is used to construct a heterogeneous biological information network for lncRNA-miRNA interactions. The topology of the heterogeneous biological information network includes a set of nodes and a set of edges. The set of nodes includes lncRNAs, miRNAs, diseases, and proteins. The set of edges includes lncRNA-miRNA associations, lncRNA-disease associations, miRNA-disease associations, protein-lncRNA interactions, and protein-miRNA interactions. The feature initialization module is used to initialize the features of the nodes of the heterogeneous biological information network to form an initial feature matrix. The model building module is used to construct a graph spiking neural network embedding representation model based on the topology of the heterogeneous biological information network and the initial feature matrix. The association predictor construction module is used to obtain the lncRNA embedding representation and miRNA embedding representation of the connected lncRNA-miRNA node pairs according to the graph spiking neural network embedding representation model, and to construct the lncRNA-miRNA association predictor according to the lncRNA embedding representation and miRNA embedding representation. The prediction module is used to predict the potential interaction strength between unconnected lncRNA-miRNA node pairs based on the lncRNA-miRNA association predictor when it is necessary to predict unconnected lncRNA-miRNA node pairs, and obtain the prediction association confidence.

8. The lncRNA and miRNA association mining system according to claim 7, characterized in that, The system also includes: The loss function design module is used to design a multi-task joint loss function for the lncRNA-miRNA association predictor, which combines cross-entropy loss and graph structure reconstruction loss. The multi-task joint loss function is composed of a weighted sum of the binary cross-entropy loss of association prediction and the graph structure reconstruction loss.

9. The lncRNA and miRNA association mining system according to claim 8, characterized in that, The system also includes: The association predictor optimization module is used to optimize the parameters of the lncRNA-miRNA association predictor by employing an algorithm that combines backpropagation and temporal backpropagation, using a rectangular function for gradient updates, and minimizing the multi-task joint loss function using the Adam optimizer.

Images