Drug ATC Code Prediction Method Based on Graph Transformation Network

By constructing a drug-target protein-disease composite feature matrix and using graph transformation networks to learn heterogeneous network association information, the problem of insufficient utilization of drug association information in existing technologies is solved, and more efficient drug ATCCode prediction is achieved.

CN114420310BActive Publication Date: 2024-11-12HENAN UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210063363.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-18
Publication Date
2024-11-12
Estimated Expiration
2042-01-18

AI Technical Summary

Technical Problem

Existing drug ATCCode prediction methods fail to fully utilize information related to the association between drugs and target proteins, diseases, etc., resulting in low prediction efficiency.

Method used

A drug ATC code prediction method based on graph transformation network (DACPGTN) is adopted. By constructing a drug-target protein-disease composite feature matrix, the potential correlation information in the heterogeneous network is learned by graph transformation network, and the drug ATC code is predicted by combining an end-to-end prediction module.

Benefits of technology

It improves the accuracy and efficiency of drug ATCCode prediction, especially in multi-label classification problems, outperforming existing methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114420310B_ABST
    Figure CN114420310B_ABST
Patent Text Reader

Abstract

The present invention discloses a drug ATC Code prediction method based on a graph transformation network, namely, DACPGTN. Firstly, target proteins and diseases related to drugs are obtained, and seven drug similarities are obtained through drug interaction information based on different evaluation criteria, and target proteins and disease similarity information associated with drugs are searched or calculated, and similarity information is used as a feature to jointly construct a corresponding composite feature matrix. Secondly, considering introducing known association information existing between three entities of drugs, target proteins and diseases, a heterogeneous graph representing multiple different edge relationships is constructed, and a Graph Transformer Layer in a graph transformation network is used to learn association information of multiple heterogeneous adjacency matrices, so as to learn the potential multiple association graph structure between drugs, target proteins and diseases, and finally the association information graph structure obtained by the Graph Transformer Layer and the drug, target protein and disease composite feature matrix are input into an end-to-end prediction module for learning, so as to make a final drug ATC Code prediction. The present invention is simple and effective, and is compared with other methods and tested on a data set to show that the invention has good performance in drug ATC Code prediction.
Need to check novelty before this filing date? Find Prior Art