Multilayer network-based drug-target and effect prediction

Abstract

In this paper, we conducted to predict drug targets and effects using multilayer networks. We tried topredict the drug target based on the drug response gene expression data so that it can be applied evenwhen there is no structural information of the drug or target protein or when the structure changes,such as G-protein coupled receptor (GPCR). At this time, a molecular network was introduced to copewith the case where there is no drug response data. Existing studies are based on a relatively simplenetwork structure, and have a disadvantage in that they have not captured accurate drug responsephenomena. To overcome this drawback, a multilayer molecular network with a deeper structure wasconstructed by distinguishing the roles of transcription factors and targets. A scoring formula suitable forthe constructed multilayer molecular network was proposed, and it was verified that the targets obtainedthrough the scoring formula were more accurate than the models used in previous studies. In addition, itwas revealed through actual case studies that the score can suggest the molecular mechanism of dietarydrugs and can also be used to find the target of coronavirus. If there is a target of the drug predicted in this way, it becomes easy to predict the effect ofthe drug. However, there may be more than one drug target, and human physiology is regulated atvarious levels. In order to model multiple drug targets and a complex human body at once, we alsotried to model the human body based on a multi-layered network structure. A multilayer network wasbuilt based on different types of entities and relational information between them in three major layers:molecule, function, and phenotype, and a standardized ontology table and relational information formwere defined to collect abundant information. In-house data could also be collected through a platformthat standardizes the heterogeneity of assay data obtained through various institutions and experiments.In the multilayer network constructed based on the information thus obtained, it was possible to predictthe relationship between drugs and diseases through the shortest path length and random walk algorithm.Confirmed. Furthermore, the performance of predicting the relationship between drug combinations anddiseases was high, and through this, candidate drug combinations with the possibility of treating diseasescould be proposed.