2.2.2

Similarity Search Based on 2D and 3D Descriptors

In a 2D representation, the molecule is represented as a graph, without spatial

coordinates of the atoms. The atoms are represented as nodes and the bonds as

edges. A number of approaches such as SPiDER (Reker et al. 2014), self-organizing

maps (SOM) and similarity ensemble approach (SEA) have been developed for in

silico identification of ligand-target interactions. SEA is used to identify molecular

targets based on set similarities of their respective ligands (Keiser et al. 2007), and

using this approach, Lounkine et al. had predicted the activity of 656 marketed drugs

on 73 unintended side-effect targets, and nearly 50% of these predictions were later

experimentally confirmed (Lounkine et al. 2012).

A major determinant of biological activity are the 3D characteristics of a mole-

cule, as drug pairs that share high 3D similarity but low 2D similarity (i.e. a novel

scaffold) were found to exhibit pharmacologically relevant differences in terms of

specific protein target modulation (Yera et al. 2011). In chemogenomics research for

3D

similarity

searching,

most

commonly

used

measures

are

shape-

or

pharmacophore-based similarity (Willett 2009; MacCuish and MacCuish 2014).

Rapid overlay of chemical structures (ROCS) developed by OpenEye Scientific

Software is a commonly used shape-based platform that has been used for drug

repurposing

studies

(Méndez-Lucio

et

al.

2014),

in

reprofiling

existing

FDA-approved drugs (Vasudevan et al. 2012) and to identify off-targets for several

drugs (Abdul Hameed et al. 2012). Recently, there has been a surge in the develop-

ment of computational tools for 3D similarity search, which include Gaussian

ensemble screening (GES), computational polypharmacology fingerprint (CPF)

and feature point pharmacophores (FEPOPS) (Jenkins et al. 2004; Pérez-Nueno

et al. 2012, 2014).

2.2.3

Structure-Based Methods

These methods predict the binding of a ligand to the target whose 3D structure has

been obtained experimentally by X-ray crystallography or NMR. In their absence,

homology-based models may be used, but due to their low reliability, the off-target

predictions are less accurate. Using the 3D atomic coordinates of the target, molecu-

lar docking predicts binding orientation and binding affinity of molecules.

2.2.4

Inverse Docking

The technique of inverse/reverse docking, i.e., docking ligands against a variety of

targets is being used for target prediction, and subsequently, the ligands are scored

according to their binding affinity scores with the targets (Rognan 2010; Koutsoukas

et al. 2011). Tools such as idTarget, INVDOCK, TarFisDock and DRAR-CPI have

been designed for inverse docking to predict the targets and/or side effects of various

ligands (Chen and Ung 2001; Li et al. 2006; Luo et al. 2011;Wang et al. 2012).

20

T. R. Sahrawat and R. C. Sobti