March 16, 2016, reported by Tan et al. Polypharmacology approach of drug design

involves systematic integration of the data derived from different disciplines, such as

computational modelling, synthetic chemistry, in vitro/in vivo pharmacological

testing and clinical studies (Yamanishi et al. 2008; Dar et al. 2012) and is encourag-

ing the shift to experimental and computational multi-target approaches (Hopkins

2007).

2.2

Polypharmacology Studies Using In Silico Approaches

In recent times, a number of computational approaches, such as bioinformatics,

ligand- and structure-based methods, ligand binding site similarity comparison,

network systems biology and data-mining-based methods, have been applied to

the study of polypharmacology (Tan et al. 2016). This review aims to summarize

some of the recently developed computational tools, databases and web servers that

are being used to study polypharmacology to identify possible off-targets of drugs

and for repurposing of known drugs.

2.2.1

Ligand-Based Methods

The basic principle of ligand-based target identification methods is that similar

receptors bind similar ligands. Over the past decade, there has been a rapid growth

in biological databases and biology-related web resources that makes huge amount

of chemogenomics data freely available to the research community. Databases such

as ChemBank and Chemical Entities of Biological Interest (ChEBI) contain infor-

mation of biologically important small molecules; UniProtKb and Protein Data Bank

(PDB) contain protein information, whereas protein-ligand interactions are present

in BindingDB, Therapeutic Target Database (TTD) and ChEMBL. These databases

contain an enormous amount of complex data matrices, which cannot be analysed

using traditional computational tools for studying target-ligand interactions. There-

fore, to handle the “big data” problem, ligand-based target fishing approaches are

used that are based upon machine learning models or similarity-based screening. In

the former approach, compounds are classified on the basis of activity prediction

using Binary kernel discrimination, naive Bayesian classifier, artificial neural

networks and support vector machine (SVM) (Lavecchia 2015), and a training

data set with known characteristics (active or inactive) is essential. In similarity-

based target fishing, the protein targets for screening are initially determined

followed by identification of ligands to represent those targets and finally the

similarity method for comparing ligands is selected. The ligand-based approaches

have advantages such as not being dependent upon the availability of 3D structure

information of the target and faster descriptor calculations. Their disadvantages

include false-positive results due to high similarity of inactive and active

compounds, or no hits may be obtained in the absence of ligand-target interaction

information in the databases.

2

Polypharmacology: New Paradigms in Drug Development

19