A computational model for GPCR-ligand interaction prediction.

Shiva Karimi,Maryam Ahmadi,Farjam Goudarzi,Reza Ferdousi

doi:10.1515/jib-2019-0084

Abstract

G protein-coupled receptors (GPCRs) play an essential role in critical human activities, and they are considered targets for a wide range of drugs. Accordingly, based on these crucial roles, GPCRs are mainly considered and focused on pharmaceutical research. Hence, there are a lot of investigations on GPCRs. Experimental laboratory research is very costly in terms of time and expenses, and accordingly, there is a marked tendency to use computational methods as an alternative method. In this study, a prediction model based on machine learning (ML) approaches was developed to predict GPCRs and ligand interactions. Decision tree (DT), random forest (RF), multilayer perceptron (MLP), support vector machine (SVM), and Naive Bayes (NB) were the algorithms that were investigated in this study. After several optimization steps, receiver operating characteristic (ROC) for DT, RF, MLP, SVM, and NB algorithm were 95.2, 98.1, 96.3, 95.5, and 97.3, respectively. Accordingly final model was made base on the RF algorithm. The current computational study compared with others focused on specific and important types of proteins (GPCR) interaction and employed/examined different types of sequence-based features to obtain more accurate results. Drug science researchers could widely use the developed prediction model in this study. The developed predictor was applied over 16,132 GPCR-ligand pairs and about 6778 potential interactions predicted.

Highlights

IntroductionPeptides and proteins with different structures and various biological impress play an essential role in the human body’s main activities (e.g., hormones, neurotransmitters, modulators) and diagnoses/therapy of the disease [1,2,3]
Peptides and proteins with different structures and various biological impress play an essential role in the human body’s main activities and diagnoses/therapy of the disease [1,2,3]
Decision tree (DT), random forest (RF), multilayer perceptron (MLP), support vector machine (SVM), and Naive Bayes (NB) were the algorithms that were investigated in this study

Summary

Introduction

Peptides and proteins with different structures and various biological impress play an essential role in the human body’s main activities (e.g., hormones, neurotransmitters, modulators) and diagnoses/therapy of the disease [1,2,3]. The main part of target proteins and nearly more than half of drug targets are enzymes and G protein-coupled receptors (GPCRs). GPCRs are an interesting group of druggable proteins that are vastly considered drug targets (almost 26.8% of drug targets are GPCRs) [2, 4, 5]. GPCRs are classified in five distinct groups (i.e., glutamate [G], rhodopsin-like [R], adhesion [A], frizzled/taste [F] and secretin [S]) [6]. A significant part of GPCR-targeted drug development is related to an interaction between GPCRs and other molecules (e.g., natural ligands, synthetic agonists, and other proteins) [4]

Methods

Results

Discussion

Conclusion