Prediction Methods And Application Study Of GPCR-ligand Interaction Network

G-protein coupled receptors (GPCRs) are important drug targets. When a GPCR binds a ligand, the extracellular signal is transmitted to the cell which will cause cell response and regulate the life activities of cell. The elucidation of the GPCR-ligand interactions is the first step in the study of GPCR functions, the traditional docking strategies for predicting GPCRs-ligands interactions are limited by the difficulty to model correctly the3D structures. Thus, there is a strong incentive to develop new methods to discover potential ligands for GPCRs. In this article, we combined the machine learning method and statistical knowledge to predict the interactions between receptors and ligands by analyzing the network topology features. The main works are as follows:(1) For non-orphan receptors, the vector distance based method was proposed to predict the interactions between receptors and ligands. It computed the shortest paths between receptor nodes and ligands nodes, by mapping the relationship between the receptors and the ligands into a network. Receptors and ligands have same number of features after executing two feature transition methods. Then, four distance inferred functions were used to measure the relationship between the receptors and the ligands, and most of predicted results are consistent with cluster analysis and phylogenetic tree analysis.(2) For orphan receptors, the network path based method was proposed to predict the interactions between receptors and ligands. Orphan receptors are no longer isolated points of the network by adding the orphan receptor-receptor interactions. According to the feature differences to network nodes in the known receptor-ligand pairs and unknown receptor-ligand pairs, we proposed the path addition method (PAM) and path subtraction method (PSM) to construct the pairs’features. Then, we predicted the potential interactions by SVM classifier. The results show that the method can effectively predict the relationship between the receptors and the ligands, and some predicted pairs are confirmed in authoritative databases.(3) For breast cancer, the target-drug interactions, drug-drug interactions, drug-disease interactions and target-target interactions are extracted from databases: KEGG, MINT etc. we analyzed the network’s topology structure and reliability, after embedding the four above interactions into one network. Moreover,11features which describe the drug-target pair were proposed based on the network topology features, and we developed the related feature extraction tool. Finally, target-drug pairs are predicted through a statistical ranking alrorithm, some predicted results were also confirmed in authoritative databases.