Graph convolution drug pair interaction prediction method and system based on knowledge graph

Abstract

The invention provides a graph convolution drug pair interaction prediction method based on a knowledge graph. The method comprises the following steps: extracting drug pair data samples, and generating a data set comprising a training set, a verification set and a test set; constructing a knowledge graph corresponding to the data set; establishing a GCN drug pair interaction prediction model, andlearning feature information of drugs contained in the drug pair and neighborhoods of the drugs; inputting the drug pair data samples of the training set into the GCN drug pair interaction predictionmodel, and training the GCN drug pair interaction prediction model; optimizing a loss function of a training result, and then sending the training result to the GCN drug pair interaction prediction model for training; conducting iterative computation to complete the training of the GCN drug pair interaction prediction model; inputting the drug pair data samples in the test set into the GCN drug pair interaction prediction model to obtain a test result; and analyzing the test result to obtain a prediction result. The knowledge graph-based graph convolution drug pair interaction prediction method and system provided by the invention are high in accuracy and short in training time.

Description

【Technical field】 [0001] The present invention relates to the technical field of drug pair interaction prediction, in particular to a knowledge map-based graph convolution drug pair interaction prediction method and system. 【Background technique】 [0002] Drug R&D is a high-investment and high-risk field. It takes an average of 10-15 years for a drug to go from R&D to marketing, with an average investment of about 2.6 billion US dollars, and on average, only one of every 5,000-10,000 molecules entering the R&D pipeline can Successfully developed. The use of computing methods can greatly reduce costs. Relying on data-driven machine learning methods, training models through a large number of open source experimental data and predicting downstream tasks has become a hot spot in the field of drug discovery as an auxiliary computing. [0003] In related technologies, machine learning algorithms are widely used in the prediction of drug-drug interactions. Most of the existing ar...

AI Technical Summary

Problems solved by technology

Due to the scarcity of multi-class label data available in these methods, it is difficult to obtain reasonably distributed data samples for batch training.

At the same time, the current mainstream methods only consider a single relationship between drugs, that is, drug-drug interactions, and ignore the potential associations between drugs and other entities (such as targets and genes).

Images