Medicine relocation method based on deep learning multi-source heterogeneous network

Abstract

The invention belongs to the field of computer bioinformatics, and discloses a medicine relocation method based on a deep learning multi-source heterogeneous network. The method comprises the following steps of: obtaining a probability co-occurrence matrix data set by using a random walk method; calculating a data set using a shifted Positive Point Mutual Information (PPMI) matrix method; and training a graph convolution encoder model by using the calculated multi-source data set, obtaining low-dimensional embedded representation of medicine information as input data of a variational autoencoder to perform parameter training, and combining the trained model with a known medicine disease incidence matrix to perform final medicine relocation prediction. The invention avoids the limitations of a traditional feature extraction method, such as: the method highly depends on medical care personnel to extract high-quality characteristics with distinctiveness, has certain difficulty and low accuracy, and realizes high-precision medicine relocation prediction by means of a graph convolution encoder model and a variational autoencoder network model.

Description

Technical field: [0001] The invention belongs to the technical field of computer biological information network embedding and machine learning, and relates to a drug relocation method for multi-source heterogeneous networks, in particular to a drug relocation method for multi-source heterogeneous networks based on deep learning. Background technique: [0002] New drug development is an extremely time-consuming, laborious and high-risk process. Fully exploring new uses of existing drugs and repositioning drugs has been widely valued by the biomedical industry. How to discover drug-disease with potential therapeutic relationship from a large number of unproven relationship pairs is the research focus of drug repositioning. With the help of machine learning models, starting from the existing relationship between drugs and diseases, analyzing and integrating drug information can increase the degree of enrichment of potential drug-disease relationship pairs to reduce the false p...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the problem of unbalanced drug information data samples, the present invention provides a drug relocation method based on deep learning multi-source heterogeneous network. In order to avoid the limitations of traditional feature extraction methods, such as: highly dependent on the experience It is highly subjective, takes a lot of time and energy to complete, and there are often certain difficulties in extracting distinguishable high-quality features, and the accuracy rate is low. The present invention relies on the graph convolutional encoder model and variational self-encoding neural network, automatically learn the low-dimensional network features of multi-source heterogeneous drugs, and complete the drug repositioning work of drug-disease association prediction

Images