FIELD: medicine, therapy.
SUBSTANCE: invention can be used to predict and prioritize drug targets for diseases based on a disease-specific graph using machine learning methods. A graph of protein-protein interactions is created, in which nodes are protein-coding genes, edges are interactions of protein-coding genes. A disease-specific graph is created based on the graph of protein-protein interactions by adding nodes of diseases and drug targets and the following links to the graph: protein-coding gene — disease, protein-coding gene — drug target, disease — drug target, disease-disease. The marking of nodes and the construction of graph connections are carried out using disease-specific information. With the help of convolutional neural networks, the resulting graph is convolved into a vector representation. The obtained vector representation of the disease-specific graph is fed into the trained machine learning model for predicting and prioritizing drug targets and a ranked list of protein-coding genes is obtained in terms of the probability of being a drug target for diseases.
EFFECT: method provides accurate prediction of the probability of protein-coding genes to be drug targets for diseases, accelerates the process of identifying drug targets for diseases, and accelerates the development of new effective drugs by automatically ranking the list of protein-coding genes in terms of the probability of being a drug target for diseases with using machine learning methods based on a disease-specific graph.
5 cl, 5 dwg, 5 tbl
Authors
Dates
2023-06-28—Published
2022-03-14—Filed