CN-121237229-B - Method and system for predicting association of annular RNA and drug

CN121237229BCN 121237229 BCN121237229 BCN 121237229BCN-121237229-B

Abstract

The invention discloses a method and a system for predicting association between annular RNA and a drug, wherein the method comprises the steps of S1, respectively constructing a multi-source similarity network of the annular RNA and a multi-source similarity network of the drug based on a known association data set of the annular RNA and the drug, S2, inputting the similarity networks into a shared constraint driven collaborative feature learning module to output a collaborative feature matrix, S3, inputting the collaborative feature matrix into a minimum entropy graph structure learning module to obtain a graph structure matrix, S4, inputting the graph structure matrix and the collaborative feature matrix into a graph convolution network for association prediction between the annular RNA and the drug. The collaborative feature learning method provided by the proposal fully utilizes different information among the multi-source data, so that the multi-source information is fully interacted to obtain more comprehensive features, in addition, the graph structure learning module learns more discriminant features by using a minimum entropy learning method, enhances graph topological features, further optimizes feature representation, and effectively improves prediction accuracy and generalization capability of the model.

Inventors

ZOU QUAN
ZHANG XUE
WANG CHUNYU
NIU MENGTING

Assignees

电子科技大学长三角研究院(衢州)

Dates

Publication Date: 20260505
Application Date: 20251201

Claims (8)

1. A method for predicting association between annular RNA and medicine based on collaborative feature learning and minimum entropy diagram structure learning is characterized by comprising the following steps: S1, respectively constructing a multisource similarity network of the annular RNA and a multisource similarity network of the medicine based on a known association data set of the annular RNA and the medicine; S2, inputting the multi-source similarity network of the annular RNA and the multi-source similarity network of the medicine into a collaborative feature learning module driven by sharing constraint, and outputting a collaborative feature matrix; s3, inputting the collaborative feature matrix to a minimum entropy diagram structure learning module to obtain a diagram structure matrix; S4, inputting the graph structure matrix and the synergetic feature matrix into a graph convolution network for the association prediction of the annular RNA and the medicine; In step S2, the collaborative feature learning module includes a first self-encoder, a sharing constraint module, and a second self-encoder, and step S2 includes: Inputting the multisource similarity network to a first self-encoder, generating node feature matrixes corresponding to the similarity networks, and inputting the node feature matrixes to a sharing constraint module; the sharing constraint module calculates the similarity between any two nodes by adopting the pearson correlation coefficient, screens out node pairs with obvious similarity, and constructs a constraint matrix containing the information of the node pairs with obvious similarity; Performing intersection operation on constraint matrixes corresponding to all similarity networks of the annular RNA to obtain a shared constraint set of the annular RNA; performing intersection operation on constraint matrixes corresponding to all similarity networks of the medicines to obtain a sharing constraint set of the medicines; Respectively inputting the annular RNA sharing constraint set and the drug sharing constraint set and the low-dimensional node characteristics output by the corresponding first self-encoder into a subsequent second self-encoder together, and finally obtaining the enhanced similarity characteristics of the annular RNA and the drug respectively; splicing the annular RNA enhanced similarity characteristic and the drug enhanced similarity characteristic to obtain the synergetic feature matrix; the minimum entropy diagram structure learning module comprises an optimization view and an inference view; the optimized view is constructed based on the initial adjacent matrix A and the cooperative characteristic matrix X; The inference view is based on an adjacency matrix Constructing and constructing a cooperative feature matrix X; The adjacent matrix The initial adjacent matrix A is processed by nonlinear activation function through a atlas Xi Qi and is obtained through normalization and symmetry processing; The initial adjacency matrix A is constructed based on a circular RNA-drug association matrix.
2. The method for predicting circular RNA and drug association based on collaborative feature learning and minimum entropy diagram structure learning of claim 1, wherein in step S1, the multi-source similarity network of circular RNAs comprises, sequence similarity, entropy similarity, gaussian interaction kernel similarity of circular RNAs; The multi-source similarity network of the medicines comprises the structural similarity, the information entropy similarity and the Gaussian interaction nuclear similarity of the medicines.
3. The method for predicting circular RNA and drug association based on collaborative feature learning and minimum entropy diagram structure learning of claim 2, wherein three similarity networks for circular RNAs are constructed by: Host gene information of each circular RNA is obtained from a biological database, and the similarity of the circular RNA sequences is calculated based on the host gene information; Calculating the information entropy of a single annular RNA based on the known association data set of the annular RNA and the medicine, and calculating the similarity of the information entropy of the annular RNA according to the common association medicine set of the annular RNA and other annular RNAs; And constructing a circular RNA-drug association matrix based on the known association data set, and calculating to obtain the corresponding circular RNA Gaussian interaction kernel similarity based on column vectors corresponding to circular RNAs in the circular RNA-drug association matrix.
4. The method for predicting circular RNA and drug association based on collaborative feature learning and minimum entropy diagram structure learning of claim 3, wherein three similarity networks for drugs are constructed by: Searching SMILES structure data of each drug from a biological information database, obtaining a topological fingerprint corresponding to each drug by using a related tool, and calculating based on the topological fingerprint to obtain the structural similarity of the drug; Calculating the information entropy of a single drug based on the known association data set, and calculating to obtain the similarity of the information entropy of the drug according to the common association annular RNA set of the drug and other drugs; Based on the row vectors corresponding to the drugs in the circular RNA-drug association matrix, the corresponding drug Gaussian interaction kernel similarity is calculated.
5. The method for predicting circular RNA and drug associations based on collaborative feature learning and minimum entropy diagram structure learning of claim 1, wherein the minimum entropy diagram structure learning module introduces edge discard and feature masking strategies and employs different feature masking probabilities for the optimized view and the inference view and the same edge discard probability; After the optimized view and the inferred view are enhanced by edge discarding and feature masking measurement, the node representations of the enhanced two views are extracted using a graph encoder and mapped into final feature vectors via a multi-layer perceptron.
6. The method for cyclic RNA and drug association prediction based on collaborative feature learning and minimum entropy diagram structure learning of claim 5, wherein in step S3, the minimum entropy diagram structure learning module learns by minimizing the output entropy value, the minimum entropy loss function The following are provided: , the final feature vectors of the optimized view and the reasoning view are respectively; Respectively is 、 The dimension of the maximum value; 、 Normalized features of final feature vectors of the optimized view and the reasoning view respectively; Representing a cross entropy loss function.
7. The method for predicting circular RNA and drug association based on collaborative feature learning and minimum entropy diagram structure learning of claim 1, wherein step S4 comprises: the graph structure matrix obtained in the step S3 and the collaborative feature matrix output in the step S2 are input into a graph convolution network together; The training is performed with mean square error as the loss function, targeting the difference between the true and predicted values of the circular RNA-drug association.
8. A loop RNA and drug association prediction system based on collaborative feature learning and minimum entropy diagram structure learning, for performing the method of any one of claims 1-7.

Description

Method and system for predicting association of annular RNA and drug Technical Field The invention belongs to the field of computer bioinformatics, and particularly relates to a method and a system for predicting association between annular RNA and medicine based on collaborative feature learning and minimum entropy diagram structure learning. Background In the field of bioinformatics, the prediction of association of circular RNAs with drugs is a key technology for drug localization and development of disease treatment targets. The traditional related prediction method is mostly dependent on a single biological characteristic construction model, integrates data by adopting a simple characteristic splicing or fixed weight fusion mode, and is difficult to fully mine hidden complementary information in multi-source data. Aiming at the associated prediction of RNA and diseases, the inventor provides a double-view scheme [ publication number CN119400252B ] for constructing a similarity network and a meta-path network, and the scheme introduces a sharing unit, a multichannel attention mechanism and a contrast learning strategy, so that cross-interaction of multi-view information and dynamic distribution of view weights are realized. In one aspect, however, the protocol is used for the prediction of association of a circular RNA with a disease, and does not involve the prediction of association of a circular RNA with a drug. On the other hand, the scheme is designed to focus on multi-view information interaction and view weight optimization, and when a scene with sparse data and high noise of a graph structure is handled, the sharing unit is difficult to capture effective cross-view information due to insufficient integrity of a meta-path network, so that the prediction performance is greatly reduced, and the scheme cannot be applied to a prediction scene with known association of less annular RNA and medicine. Disclosure of Invention The invention aims to provide a method and a system for predicting association between annular RNA and a drug based on collaborative feature learning and minimum entropy diagram structure learning, aiming at the problems existing in the prior art. In order to achieve the above purpose, the present invention adopts the following technical scheme: a method for predicting association between a circular RNA and a drug based on collaborative feature learning and minimum entropy diagram structure learning, the method comprising: S1, respectively constructing a multisource similarity network of the annular RNA and a multisource similarity network of the medicine based on a known association data set of the annular RNA and the medicine; S2, inputting the multi-source similarity network of the annular RNA and the multi-source similarity network of the medicine into a collaborative feature learning module driven by sharing constraint, and outputting a collaborative feature matrix; s3, inputting the collaborative feature matrix to a minimum entropy diagram structure learning module to obtain a diagram structure matrix; S4, inputting the graph structure matrix and the synergetic feature matrix into a graph rolling network for the association prediction of the annular RNA and the medicine. In the above method for predicting association between circular RNAs and drugs based on collaborative feature learning and minimum entropy diagram structure learning, in step S1, the multi-source similarity network of the circular RNAs includes sequence similarity, information entropy similarity, and gaussian interaction kernel similarity of the circular RNAs; The multi-source similarity network of the medicines comprises the structural similarity, the information entropy similarity and the Gaussian interaction nuclear similarity of the medicines. In the above-mentioned loop RNA and drug association prediction method based on collaborative feature learning and minimum entropy diagram structure learning, three similarity networks of loop RNA are respectively constructed by the following modes: Host gene information of each circular RNA is obtained from a biological database, and the similarity of the circular RNA sequences is calculated based on the host gene information; Calculating the information entropy of a single annular RNA based on the known association data set of the annular RNA and the medicine, and calculating the similarity of the information entropy of the annular RNA according to the common association medicine set of the annular RNA and other annular RNAs; And constructing a circular RNA-drug association matrix based on the known association data set, and calculating to obtain the corresponding circular RNA Gaussian interaction kernel similarity based on column vectors corresponding to circular RNAs in the circular RNA-drug association matrix. In the above-mentioned loop RNA and drug association prediction method based on collaborative feature learning and minimum entropy diagram structure learn