MiRNA-lncRNA interaction prediction method based on multi-view projection fusion and truncated matrix decomposition

Abstract

The invention provides a miRNA-lncRNA interaction prediction system based on multi-view projection fusion and truncated matrix decomposition, and belongs to the crossing field of bioinformatics and machine learning. The method comprises the following steps: firstly, constructing a comprehensive miRNA-lncRNA similarity map by comprehensively utilizing sequence similarity, functional similarity, expression similarity and Gaussian kernel similarity of miRNA / lncRNA, and constructing a corresponding network structure based on different similarities and known interaction information; secondly, a multi-view projection fusion method is provided, similar network topology structures from multiple view angles and original interaction matrix information are fully fused, and abundant topology features in the biological network are mined while the integrity of the interaction matrix structure is maintained. Finally, a truncated matrix decomposition technology is introduced, matrix dimension reduction is achieved through the number of truncated singular values, key feature information is reserved, and therefore prediction efficiency and precision are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of miRNA-lncRNA interaction prediction, and particularly relates to a miRNA-lncRNA interaction fast prediction method based on multi-view projection fusion and truncated matrix decomposition. BACKGROUND

[0002] Increasing studies have shown that most transcriptomes (98%) are composed of non-coding RNA (ncRNA) genes. Generally, these genes do not translate into functional proteins, but regulate target gene expression and affect disease progression. In-depth studies of ncRNAs have revealed the important roles of lncRNAs and miRNAs in regulating various biological processes. lncRNAs, with a length of more than 200 nucleotides, are an important class of regulatory factors in the human genome. They are related to many biological processes and have a complex relationship with diseases such as liver and breast tumors. miRNAs are small non-coding RNA molecules, usually about 22 nucleotides in length. They control about one-third of human genes, affect various biological processes and play important roles in cell regulation and disease pathways. It has been reported that miRNAs regulate the expression and function of lncRNAs through various mechanisms, thereby playing a key role in many biological processes. Many studies have highlighted the key role of miRNA-lncRNA interactions in various biological processes, including cell metabolism, gene regulation and cancer development. Therefore, exploring these miRNA-lncRNA interactions not only enhances our understanding of the functional expression of lncRNAs and miRNAs, but also opens up new ways for biomedical research of biological processes. However, identifying large-scale mli through wet lab experiments is both time-consuming and expensive. In contrast, computational methods for predicting miRNA-lncRNA interactions are more convenient and can help biological experiments to effectively discover new interactions.

[0003] With the continuous accumulation of newly discovered miRNA-lncRNA interaction data, the size of related data sets is rapidly expanding. This growth greatly increases the computational complexity of existing models, affecting prediction efficiency. Therefore, it is crucial to develop an accurate and fast prediction algorithm. Matrix decomposition is a widely used and effective related prediction method. It can extract key features by truncating singular values, thereby reducing the computational load of the model. However, the optimal number of truncated features in matrix decomposition is still uncertain. In addition, most existing models rely on a single similarity measure, which introduces bias in computational modeling. The potential of topology based on similarity has not been fully explored or utilized SUMMARY

[0004] The purpose of the present application is to solve the difficulties existing in the above-mentioned miRNA-lncRNA interaction prediction field, provide a multi-view projection fusion and truncated matrix decomposition miRNA-lncRNA interaction prediction method, integrate multi-source data of miRNA and lncRNA, and help researchers to carry out efficient prediction.

[0005] A, based on the sequence similarity, functional similarity, expression similarity and Gaussian kernel similarity of miRNA, preprocessing. Based on the sequence similarity, functional similarity, expression similarity and Gaussian kernel similarity of lncRNA, preprocessing.

[0006] B, based on the weighted average strategy, the miRNA similarity and lncRNA similarity of different views are integrated into the comprehensive similarity of miRNA and lncRNA respectively.

[0007] C, based on the sequence similarity matrix, functional similarity matrix, expression similarity matrix and Gaussian kernel similarity matrix of miRNA and lncRNA, the corresponding similarity network is constructed. Based on the miRNA-lncRNA interaction matrix, the miRNA-lncRNA interaction network is constructed.

[0008] D, based on the multiple similarity network topologies of miRNA and lncRNA space and the initial interaction matrix structure, the multi-view projection fusion technology is used to update the miRNA-lncRNA interaction matrix.

[0009] E, based on the acceleration framework of truncated matrix decomposition.

[0010] 2. According to claim 1, based on the sequence similarity, functional similarity, expression similarity and Gaussian kernel similarity of miRNA, preprocessing. Based on the sequence similarity, functional similarity, expression similarity and Gaussian kernel similarity of lncRNA, preprocessing. The present application uses the known miRNA-lncRNA interaction data downloaded from lncRNA SNP database to calculate the Gaussian kernel similarity of miRNA and lncRNA. The miRNA and lncRNA function spectrum data are collected in miRTarBase v6.1 and Lnc-GFP method. The miRNA and lncRNA expression spectrum data are collected in microRNA.org and NONCODE database. The miRNA and lncRNA sequence information are collected in miRbase database and LNCipedia platform. Different calculation methods are used to obtain the multi-view similarity of miRNA and lncRNA.

[0011] 3. Based on the weighted averaging strategy described in claim 1, miRNA similarity and lncRNA similarity across different views are integrated into the combined similarity of miRNA and lncRNA, respectively. Using the weighted averaging strategy reduces potential bias associated with using a single similarity metric and ensures that different similarities contribute equally to the combined similarity.

[0012] 4. Constructing similarity networks based on the sequence similarity matrix, functional similarity matrix, expression similarity matrix, and Gaussian kernel similarity matrix of miRNA and lncRNA according to claim 1. Constructing a miRNA-lncRNA interaction network based on the miRNA-lncRNA interaction matrix. Constructing similarity networks and miRNA-lncRNA interaction networks for different views of miRNA and lncRNA using the sequence similarity, functional similarity, expression similarity, Gaussian kernel similarity, and miRNA-lncRNA interaction matrix of miRNA and lncRNA.

[0013] 5. The miRNA-lncRNA interaction matrix is ​​updated using a multi-view projection fusion technique based on multiple similar network topologies and an initial interaction matrix structure from miRNA and lncRNA spaces as described in claim 1. Multiple similar network topologies from different spaces and the initial interaction matrix structure are used to update the miRNA-lncRNA interaction matrix. This method captures rich information about the biological network topology while maintaining the structural integrity of the original interaction matrix.

[0014] 6. The acceleration framework based on truncated matrix decomposition according to claim 1. The present invention develops an acceleration framework based on truncated matrix decomposition. The framework achieves matrix dimensionality reduction by truncating the number of singular values, retaining key feature information, thereby improving prediction efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a miRNA-lncRNA interaction prediction process using multi-view projection fusion and truncated matrix factorization; DETAILED DESCRIPTION

[0016] In order to make the objectives, technical solutions and advantages of the present invention more clear, the interaction prediction method of the present invention is further described in detail below with reference to the accompanying drawings.

[0017] In the data preprocessing stage, multi-source data of miRNA and lncRNA from different databases were integrated to construct four miRNA similarity matrices and networks and four lncRNA similarity matrices and networks, and to construct the miRNA-lncRNA interaction network.

[0018] The different miRNA similarities and lncRNA similarities are integrated into the comprehensive similarities of miRNA and lncRNA respectively using a weighted average strategy.

[0019] A multi-view projection fusion technology is developed in the application, which utilizes multiple similar network topologies from different spaces and an initial interaction matrix structure to update the miRNA-lncRNA interaction matrix.

[0020] An accelerated framework based on truncated matrix decomposition is developed in the application. The framework realizes matrix dimension reduction by truncating the number of singular values, retains key feature information, and thus improves the prediction efficiency and accuracy.

Claims

1. A miRNA-lncRNA interaction prediction method based on multi-view projection fusion and truncated matrix decomposition, including the following parts: A. Preprocessing based on sequence similarity, functional similarity, expression similarity, and Gaussian kernel similarity of miRNAs. Preprocessing based on sequence similarity, functional similarity, expression similarity, and Gaussian kernel similarity of lncRNAs. B. Based on the weighted average strategy, the miRNA similarity and lncRNA similarity of different views are integrated into the comprehensive similarity of miRNA and lncRNA, respectively. C. Construct the corresponding similarity networks of miRNAs and lncRNAs based on their sequence similarity matrix, functional similarity matrix, expression similarity matrix, and Gaussian kernel similarity matrix. Construct the miRNA-lncRNA interaction network based on the miRNA-lncRNA interaction matrix. D. Based on multiple similar network topologies in miRNA and lncRNA space and the initial interaction matrix structure, the miRNA-lncRNA interaction matrix is ​​updated using multi-view projection fusion technology. E. Acceleration framework based on truncated matrix factorization.

2. Preprocessing based on miRNA sequence similarity, functional similarity, expression similarity, and Gaussian kernel similarity according to claim 1. Preprocessing based on lncRNA sequence similarity, functional similarity, expression similarity, and Gaussian kernel similarity. The present invention uses known miRNA-lncRNA interaction data downloaded from the lncRNASNP database to calculate the Gaussian kernel similarity of miRNA and lncRNA. MiRNA and lncRNA functional profile data are collected using miRTarBase v6.1 and the Lnc-GFP method. MiRNA and lncRNA expression profile data are collected from the microRNA.org and NONCODE databases. MiRNA and lncRNA sequence information is collected from the miRbase database and the LNCipedia platform. Different calculation methods are used to obtain multi-perspective similarity between miRNA and lncRNA.

3. The weighted averaging strategy of claim 1 integrates miRNA similarity and lncRNA similarity across different views into the combined similarity of miRNA and lncRNA, respectively. Using a weighted averaging strategy reduces potential bias associated with using a single similarity metric and ensures that different similarities contribute equally to the combined similarity.

4. Constructing corresponding similarity networks based on the sequence similarity matrix, functional similarity matrix, expression similarity matrix, and Gaussian kernel similarity matrix of miRNA and lncRNA according to claim 1. Constructing a miRNA-lncRNA interaction network based on the miRNA-lncRNA interaction matrix. Constructing corresponding similarity networks of different views of miRNA and lncRNA and miRNA-lncRNA interaction networks using the sequence similarity, functional similarity, expression similarity, Gaussian kernel similarity, and miRNA-lncRNA interaction matrix of miRNA and lncRNA.

5. The method of claim 1, wherein multiple similar network topologies based on the miRNA and lncRNA space and the initial interaction matrix structure are used to update the miRNA-lncRNA interaction matrix using a multi-view projection fusion technique. Multiple similar network topologies from different spaces and the initial interaction matrix structure are used to update the miRNA-lncRNA interaction matrix. This method captures rich information about the biological network topology while maintaining the structural integrity of the original interaction matrix.

6. The truncated matrix factorization-based acceleration framework of claim 1. The present invention develops an acceleration framework based on truncated matrix factorization. This framework reduces matrix dimensionality by truncating the number of singular values, retaining key feature information, thereby improving prediction efficiency and accuracy.

Images