Brief Introduction

The prediction of target proteins for the specific compound is useful to explore ligand-target interactions and their related bio-chemical mechanisms. Target prediction also plays a critical role in the drug development. Recently, with the rapid growth of bioassay data volume, a plenty of ligand-based target prediction models have been proposed. However, most of ligand-based methods predict targets based on their ligands' chemical similarity, which cannot be applied to the targets without known ligand.

Here we present a web server of target prediction named Drug Target Predictor which no longer quantifies the physicochemical properties in the binding site and then predicts the target protein by pocket similarities. Drug Target Predictor uses random forest to predict target based on a novel integrated fingerprint of the pocket-ligand complex. The web server is rapid and with high accuracy (AUC > 0.8), which enables us to discover new drug binding targets.