Brief Introduction

Target identification plays an important role in modern chemical biology and drug discovery. It not only enables understanding of the mechanism of action of bioactive molecules or drugs but also helps characterize interactions with unintended targets, explaining side effects or adverse reactions. Attention has thus been focused on compound-protein interaction (CPI) prediction models that can guide researchers to fast lanes for hit discovery.

Two class of CPI prediction techniques have been proposed for drug development: structure based virtul screening (SBVS) and ligand based virtual screening (LBVS). SBVS methods cannot be used for sets of ligands and proteins whose three-dimensional structures are unknown. LBVS methods predict CPIs by using chemical or shape similarity. LBVS often requires many known active molecules for a target of interest. In fact, many proteins lack known active molecules. Thus, other predict method is needed to discover novel compounds for these proteins have neither ligand nor 3D structure.

Here, we developed a web server, named Sequence-Based CPI Predictor (SBCPI) to predict the targets of query compound based on protein sequence. SBCPI is made up of five steps: CPI data collection, calculation of sequence descriptor and compound fingerprint, representation of interaction vectors, predictive model construction using training data sets,and predictions from test data. The web server is rapid and accurate (AUC > 0.95), which enables us to predict all possible targets of query compound in proteome-wide.