Content uploaded by Mutasem Omar Taha
Author content
All content in this area was uploaded by Mutasem Omar Taha on Jan 30, 2018
Content may be subject to copyright.
Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=iedc20
Download by: [University of Lethbridge] Date: 12 November 2015, At: 12:59
Expert Opinion on Drug Discovery
ISSN: 1746-0441 (Print) 1746-045X (Online) Journal homepage: http://www.tandfonline.com/loi/iedc20
QSAR studies in the discovery of novel type-II
diabetes therapies.
Areej Abuhammad & Mutasem O. Taha
To cite this article: Areej Abuhammad & Mutasem O. Taha (2015): QSAR studies in the
discovery of novel type-II diabetes therapies., Expert Opinion on Drug Discovery, DOI:
10.1517/17460441.2016.1118046
To link to this article: http://dx.doi.org/10.1517/17460441.2016.1118046
Accepted author version posted online: 11
Nov 2015.
Submit your article to this journal
View related articles
View Crossmark data
1
Publisher: Taylor & Francis
Journal: Expert Opinion on Drug Discovery
DOI: 10.1517/17460441.2016.1118046
QSAR studies in the discovery of novel type-II diabetes therapies.
Areej Abuhammad
Mutasem O. Taha.*
*Corresponding Author. Email: mutasem@ju.edu.jo
Department of Pharmaceutical Sciences, Faculty of Pharmacy, The University of Jordan,
Amman 11942 Jordan.
Abstract
Introduction: Type-II diabetes mellitus (T2DM) is a complex chronic disease that represents a
major therapeutic challenge. Despite extensive efforts in T2DM drug development, therapies
remain unsatisfactory. Currently, there are many novel and important anti-diabetic drug targets
under investigation by many research groups worldwide. One of the main challenges to
develop effective orally active hypoglycemic agents is off-target effects. Computational tools
have impacted drug discovery at many levels. One of the earliest methods is quantitative
structure–activity relationship (QSAR) studies. QSAR strategies help medicinal chemists
understand the relationship between hypoglycemic activity and molecular properties. Hence,
QSAR may hold promise in guiding the synthesis of specifically designed novel ligands that
demonstrate high potency and target selectivity.
Areas covered: This review aims to provide an overview of the QSAR strategies used to
model anti-diabetic agents. In particular, this review focuses on drug targets that raised recent
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
2
scientific interest and/or led to successful anti-diabetic agents in the market. Special emphasis
has been made on studies that led to the identification of novel anti-diabetic scaffolds.
Expert opinion: Computer-aided molecular design and discovery techniques like QSAR have
a great potential in designing leads against complex diseases such as T2DM. Combined with
other in silico techniques, QSAR can provide more useful and rational insights to facilitate the
discovery of novel compounds. However, since T2DM is a complex disease that includes
several faulty biological targets, multi-target QSAR studies are recommended in the future to
achieve efficient antidiabetic therapies.
Article Highlights:
• Type II Diabetes involves several malfunctioned targets
• QSAR studies have been used and still have potential in developing new anti-diabetic
therapies
• CoMFA, CoMSIA and pharmacophore modeling represent most QSAR research
towards new Type II Diabetes therapies
• Mixing QSAR studies with other molecular modeling methodologies holds great
potential for the discovery of new anti-diabetic therapies
• Future QSAR research on anti-diabetic therapies should focus on multiple QSARs to
develop therapies that hit multiple diabetic targets simultaneously.
Keywords
QSAR, Type 2 Diabetes, GSK-3β, PTP-1B, DPP-IV, SGLT2, PPAR.
1. Introduction
1.1. Type II Diabetes Mellitus
Diabetes is a lifelong condition that poses a huge global health challenge. According to the
World Health Organization 1, the prevalence of diabetes in 2014 was estimated to be 9%
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
3
among adults older than 18 years 1, 2. It is expected that the disease will become the 7th leading
cause of death in 2030 affecting over 400 million people 1. Diabetes is a term used to describe
a group of metabolic disorders characterized by hyperglycemia. The resultant high blood
glucose level generally leads to several serious complications3-6. Type 2 diabetes mellitus
(T2DM) accounts for almost 90% of worldwide diabetes cases 1. T2DM is a progressive
disease characterized by insulin resistance in peripheral tissues and/or impaired insulin
secretion by the pancreas. Although the disease is highly connected to obesity and lack of
exercise, it can be successfully controlled by oral medications. This type of diabetes usually
affects adults after the age of 40, however, recently the prevalence of the disease has increased
among children.
Extensive research has been conducted on the search for drug targets to treat hyperglycemia
associated with T2DM 7. Figure 1 summarizes the major drug targets that have been explored
to develop hypoglycemic agents. One of the most attractive approaches for diabetes drug
development is QSAR modelling. QSAR studies allow the design of novel lead molecules and
the estimation of bioactivity prior to synthesis and biological testing 8, 9. Thus successful QSAR
modelling can help cut time and cost needed for drug discovery. Despite the availability of
drugs with different mechanism of action to treat diabetes, there is still need for new lead
molecules to fill in the pipeline with insulin independent therapies.
1.2. Recently approved anti-diabetic agents
The main goal of drug therapy in T2DM is to maintain a near-normal blood glucose level and
thus prevent or delay the onset of diabetes related complications such as retinopathy,
nephropathy and neuropathy 10, 11. Traditionally, blood glucose control is achieved by lifestyle
adjustment, insulin, sulfonylureas, and metformin as the first-line antidiabetic agents 12.
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
4
However, when considering long-term use, issues such as efficacy, tolerability and
convenience have to be considered 13. Many conventional hypoglycaemic agents exhibit
reduced efficacy over time, leading to inadequate glycaemic control 13. The escalating
epidemic of T2DM worldwide has encouraged researchers and pharmaceutical companies to
investigate new drug targets. In particular, therapies that can overcome the limitations
associated with first-line antidiabetic drugs.
Recently, several novel classes of blood glucose–lowering medications were developed. These
new agents are usually used as adjuvants in T2DM management to enhance glycaemic control
by the old treatment options. A summary of small-molecule antidiabetic agents that received
FDA approval within the last two decades is shown in Table 1. Table 2 shows small-molecules
antidiabetic agents in clinical trials.
2. Background on Quantitative Structure Activity Relationship
Basically, quantitative structure—activity relationship (QSAR) modelling aims at unveiling
relationship(s) that connects certain observed bioactivity/property with structural features of
certain list of bioactive molecules. Figure 2 summarizes the major steps carried out in QSAR
modelling. QSAR requires that the structures of modelled molecules to be representable in
quantitative terms and then to search/find the best mathematical function connecting the
quantitative values representing each structure with the corresponding experimental
activity/property. The quantitative descriptions (representations) of molecules are termed as
descriptors. Presently, there is a wide variety of computable molecular descriptors obtainable
from different resources.
QSAR methods can be classified into different types based on how their descriptors are
calculated. The simplest are zero-dimensional (0D)-QSAR models. These are developed based
on descriptors derived from information extracted from molecular formula, e.g., molecular
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
5
weight, number of atoms, atom types, sum of atomic properties. On the other hand, one-
dimensional (1D)-QSAR models correlate activity/property data with overall molecular
properties, e.g., solubility, logP, functional groups, etc…. Likewise, 2D-QSAR equations link
activity data with structural patterns, e.g., Wiener index, connectivity indices,
electrotopological state indices, flexibility indices, etc… Similarly, 3D-QSAR deals with the
position, orientation and conformation of training molecules in three-dimensional space. 3D-
QSAR tie activity/property with interaction fields (steric and electrostatic field) surrounding
the training molecules. Comparative molecular field analysis (CoMFA) and comparative
molecular similarity indices analysis (CoMSIA) are most commonly encountered examples on
this approach. Conversely, 4D-QSAR models are similar to 3D-QSAR except that they
represent each training molecule in different possible conformations, stereoisomers,
orientations, tautomers or protonation states. On the other hand, 5D-QSAR allow for induced-
fit representations in 4D-QSAR models, while 6D-QSAR models incorporate solvation models
in 5D-QSAR 14.
One of the most recent developments in the QSAR field is a new methodology known as
docking-based Comparative Intermolecular Contacts Analysis (db-CICA). This novel approach
is based on the number and quality of contacts between docked ligands and amino acid
residues within the binding pocket. It assesses a particular docking configuration based on its
ability to align a set of ligands within a corresponding binding pocket in such a way that potent
ligands come into contact with binding site spots distinct from those approached by low-
affinity ligands and vice versa. Optimal dbCICA models can be translated into valid
pharmacophore models that can be used as 3D search queries to mine structural databases for
new bioactive compounds. dbCICA was implemented to search for new inhibitors of candida
N-myristoyl transferase, 15 glycogen phosphorylase (GP), 15 Hsp90α, 16 check point kinase 17
and glucokinase activators 18.
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
6
QSAR models can be also classified based on the type of chemometric methods used to derive
them, i.e., linear or non-linear. Linear methods include linear regression, multiple linear
regression, principal component analysis, and partial least-squares. However, several nonlinear
methods were recently introduced for building predictive QSAR models, e.g., support vector
machine, artificial neural networks, k-nearest neighbors (kNN) and Bayesian neural nets14.
Still, both linear and non-linear methods require some sort of search methodology to scan long
lists of molecular descriptors to find the best combination of descriptors that can explain
bioactivity variation within a particular training set of compounds. Search methodologies range
from simple forward, backward regression analysis, genetic function algorithm or Monte Carlo
search to rather complicated methods such as simulated annealing 19.
Validation of QSAR models is a fundamental step to ensure their subsequent successful
application. It is essential to check the ability of QSAR model(s) to predict
bioactivities/properties of external compounds, i.e., not used in model building. However,
QSAR validation measures include both internal and external testing. Different statistical
metrics are implemented to ensure the internal consistency of QSAR models, these include
determination coefficient (r2), standard error of estimate, adjusted r2 (
2
a
r
) and variance ratio.
Other validation parameters include cross-validated (leave-one-out) r2 (q2), standard error of
prediction, new
2
m
r
parameters like
2)
(trainingm
r
,
2)(trainingm
r∆
; predicted r2 (
2
pred
r
),
2)
(testm
r
,
2)(testm
r∆
; other global validation parameters include
2)(overallm
r
,
2)(overallm
r∆
. Additionally,
Y-randomization testing (average
2
r
r
,
2
p
cr
) is usually done to validate the stability of QSAR
models 14, 20-23.
It remains to be mentioned that QSAR modellers are at the mercy of data providers who may
publish erroneous data. Thus, dataset curation is crucial for any QSAR modelling including
QSAR studies on anti-diabetic agents. The most important steps in curating modelled datasets
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
7
include removal of inorganics, organometallics, counterions, and mixtures; structural cleaning,
ring aromatization, normalization of specific chemotypes, curation of tautomeric forms;
deletion of duplicates; and manual checking of the structures and biological activities 24.
3. Recent Advances in QSAR modeling
QSAR modelling has witnessed several recent advances on different aspects ranging from
selecting valid datasets to the eventual selection of valid QSAR model.
A recently addressed issue in QSAR modelling is the diversity of training compounds.
Needless to say that in order for QSAR models to be of significant predictive value, their
training compounds should be as diverse as possible. A recent comprehensive study has shown
that atom topology-based descriptors, i.e., fingerprint-based descriptors and pharmacophore-
based descriptors, are most successful for assessing compounds diversity for QSAR modelling
25.
Another recently addressed aspect in QSAR is the modelability of compound datasets.
Interestingly, not all molecular datasets are modelable using QSAR analysis. Apparently,
datasets exhibiting activity cliffs can be rather resistant to QSAR analysis. Activity cliff means
a pair of molecules having similar structures but exhibit wide difference in activity values 26, 27.
Activity cliff concept has been also applied in structure-based design to elucidate key
interacting atoms in the binding site and to facilitate pharmacophore elucidation 28. Recent
advances in activity cliff concept were reported by Bajorath et al. 29, 30.
It remains to be mentioned that QSAR modellers need to avoid some common errors to
improve the QSAR modelling. Recent studies have discussed many commonly encountered
errors in QSAR/QSPR literature and provided suggestions on how to avoid such mistakes 31, 32,
33. These errors can be highlighted in the following: inappropriate endpoint data; failure to
account for data heterogeneity; collinear descriptors; obscure descriptors; inadequate or
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
8
undefined applicability domain; unacknowledged omission of data points; use of inadequate
data; too narrow range of end point values; over-fitting of data; use of excessive numbers of
descriptors in a QSAR/QSPR; inadequate statistics; misinterpretation of statistics; inadequate
training or test set selection; failure to validate a QSAR/QSPR correctly; lack of mechanistic
interpretation.
4. QSAR studies of anti-diabetic compounds
Although QSAR studies for anti-diabetic agents were reported as early as late 1990s 34-36, the
only review found in the literature dealing with molecular informatics, including QSAR,
applied to general anti-diabetic agents was published in 2007 37. Nevertheless, few reviews
describing QSAR approaches on specific diabetes-related targets have been reported 38-41. This
left a gap in the literature prompting this review.
In this report we focused on drug targets that raised recent scientific interest and/or led to
successful anti-diabetic agents in the market. Accordingly, we have reviewed QSAR reports
(Table 3) reported for five major targets involved in T2DM, namely, glycogen synthase kinase
3β (GSK-3β), protein tyrosine phosphatase 1B (PTP-1B), dipeptidyl peptidase IV (DPP-IV),
sodium-dependent glucose cotransporter 2 (SGLT2), and peroxisome proliferator-activated
receptor (PPAR). We gave special emphasis for QSAR studies that reported the discovery of
new bioactive hits against their corresponding targets.
4.1. Glycogen synthase kinase 3β
GSK-3 is a multifunctional ubiquitously expressed cytosolic serine/threonine kinase that exists
in two isoforms GSK-3α and GSK-3β. These isoforms have nearly 98% identity but are
functionally distinct. GSK-3β has been established as a regulator of glycogen synthase2
activity and involved in several signalling events in the cell-cycle. Therefore, it has attracted
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
9
the interest of many research groups worldwide as a promising target for the development of
anti-diabetic therapies 42, 43,44. GSK-3β binding pocket shares homology with other kinases at
the ATP binding region resulting in a big medicinal chemistry challenge towards the
establishment of potent and selective GSK-3β inhibitors 45-47.
Most 3D-QSAR studies carried out on GSK-3β inhibitors focused on comparative molecular
field analysis (CoMFA) and comparative molecular similarity indices analysis (CoMSIA).
These studies usually rely on the conformation/pose of docked potent ligands as templates for
molecular alignment prior to 3D-QSAR modelling. Patel and Bharatam 48 modelled 59
pyrazolo-pyridazine GSK-3 inhibitors. Their models explained the structural difference
between the selective and non-selective GSK-3 inhibitors. Fang et al. 49 generated CoMFA and
CoMSIA models for 30 known benzofuran-indole-maleimides GSK-3β inhibitors. They
coupled their 3D-QSAR models with FlexX-based docking to screen 10429 drug like-
compounds against GSK-3β. They confirmed the validity of their models by virtually capturing
23 confirmed submicromolar GSK-3β inhibitors. Similarly, Akhtar and Bharatam 50 modelled
57 anilino-aryl-maleimide GSK-3β inhibitors. Their CoMSIA and CoMFA models identified
key features explaining anti-GSK-3β bioactivities and were used to design new virtual
molecules of better predicted binding affinities. Likewise, Prasanna et al. 51 utilized CoMFA,
CoMSIA and docking studies on a number of 3-anilino-4-phenylmaleimides GSK-3 inhibitors
to compare the binding sites of GSK-3α and GSK-3β isoforms. They concluded identical
binding interactions except that Asp-133 in the β isoform is replaced by Glu-196 in the α
isoform. Dessalew et al. 52 used CoMFA and CoMSIA modelling on 49 pyrazolopyrimidines to
guide the design of new virtual GSK-3 inhibitors.
Crisan et al. 53 implemented Dragon descriptors to generate projection to latent structures
(PLS) models maleimide GSK-3 inhibitors. This was combined with shape/volume structural
analysis of binding site interactions. Their computational combination yielded QSAR models
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
10
with significant correlations. They concluded the importance of topological, charge, geometry,
shape/volume and molecular flexibility for maleimides' anti-GSK-3 bioactivities.
Lather et al. 54 compared classical QSAR modelling with 3D-QSAR analysis using 36
indirubin-based GSK-3β inhibitors. The authors carried out their classical QSAR modelling
using descriptors generated by CODESSA and Molconn-Z, while they relied on molecular
alignment to generate 3D-QSAR-based pharmacophore model using PHASE software. The
resulting pharmacophore hypothesis agreed with corresponding crystallographic structure of
ligand-GSK-3β complex.
Fu et al. 55 used 728 diverse GSK-3β inhibitors for QSAR modelling. However, to account for
data heterogeneity resulting from bioassay variability, the authors constructed a hierarchical
QSAR model which adopts a two-level structure. The base of the model included regression
models built on data collected using different particular bioassay methods. In the upper level of
the model, all compounds from different research groups are collected in a single data set, and
a classification model is built to separate compounds into different structural subclasses. They
implemented three machine learning algorithms: support vector machines, binary particle
swarm algorithm and random forests to construct classification model at the higher level and
multiple regression models at the lower level. They then used their hierarchical QSAR model
as a screening platform. Hits were further docked using ensemble docking to refine the selected
hits. The overall computational workflow identified two low micromolar anti-GSK-3β hits
(Figure 3).
Taha et al. 56 explored the pharmacophoric space of GSK-3β using diverse set of 132
inhibitors. They subsequently employed genetic algorithm and multiple linear regression to
select best possible combination of pharmacophores and physicochemical descriptors to yield
best QSAR model. The optimal QSAR included two pharmacophores suggesting the existence
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
11
of two distinct binding modes available to ligands within GSK-3β binding pocket. The
following equation shows their optimal QSAR model (Equation 1).
log(1/IC50) = − 5.52 + 0.21[Hypo4/5 − 4.40] + 0.45[Hypo6/3 − 5.67] − 1.31[S_aasC − 3.47]
+ 0.44κ2 − 3.81[12.64 − S_aaN] − 1.31[5.65 − κ2] − 2.14[Jurs-FPSA-2 − 1.13]
+ 3.69[12.90 − S_aaN] + 0.26[1.96 − S_sCH3] Equation 1
where Hypo4/5 and Hypo6/3 represent the two optimal pharmacophores, κ2 is the second order
κ shape index, Jurs-FPSA-2 is the fraction of negatively charged solvent-accessible surface
area, S_sCH3, S_aaN, and S_aasC are the electrotopological sum descriptors for methyl,
heteroaromatic nitrogens, and aromatic carbons, respectively. Several descriptors emerged in
their QSAR in spline format (truncated power splines, denoted by bolded brackets ([]). The
QSAR equation and associated pharmacophores identified three established drugs to have
nanomolar GSK-3β inhibitory IC50 values (Figure 3). Docking studies supported the binding
modes suggested by the pharmacophore/QSAR modelling.
4.2. Protein tyrosine phosphatase 1B (PTP-1B)
Insulin binding to insulin receptor initiates a cascade of cellular events that triggers the intrinsic
tyrosine kinase activity and results in autophosphorylation of critical tyrosine residues of
insulin receptor, which subsequently phosphorylates its various substrates to mediate metabolic
and mitogenic effects of insulin. Protein tyrosine phosphatase (PTP) 1B is an ubiquitous
cytosolic enzyme that directly interacts with activated IR to dephosphorylate phosphotyrosine
residues, resulting in down regulation of insulin action 57. Accordingly, PTP-1B emerged as a
potential target for treating T2DM 58. PTP-1B deficient mice showed increased insulin
sensitivity and obesity resistance, demonstrating that PTP-1B plays a major role in modulating
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
12
both insulin sensitivity and fuel metabolism. Based on this PTP-1B has drawn considerable
attention as a target for treating type 2 diabetes and obesity. However, the fact that PTP-1B is
highly homologous with T-cell protein-tyrosine phosphatase (TCPTP) makes it rather
challenging to develop safe and effective PTP-1B inhibitors 59. Several QSAR and QSAR-
related studies have been recently reported to probe ligand binding to PTP 1B.
Sachan et al. 60 modeled 29 formylchromones PTP-1B inhibitors using classical 0D-2D QSAR
analysis via multiple linear regression coupled with genetic function approximation (GFA).
The models suggested that PTP-lB inhibitory bioactivities are strongly dependent on electronic,
thermodynamic and shape related parameters.
Malla et al. 61, 62 performed atom-based 3D-QSAR analyses on formylchromane- and
thiophene-based PTP 1B inhibitors, in two separate studies, implementing partial least square
(PLS) analysis within PHASE module of Schrodinger suite. They identified two five-point
pharmacophore hypotheses: in the first case the pharmacophore included one hydrogen
acceptor, two negative ionic, and two aromatic rings, while the second model included three
hydrogen bond acceptors and two aromatic rings.
Thareja et al. 63, 64 implemented self-organizing molecular field analysis (SOMFA), a relatively
new 3D-QSAR methodology based CoMFA and self-organizing maps, on sulphonamide and
thiazolidinedione anti-PTP 1B derivatives (two separate studies). The resulting SOMFA
models suggest significant electrostatic and shape potential contributions to ligand-PTP-1B
binding.
Sobhia and Bharatam 65 performed CoMSIA studies on 1,2-naphthoquinone inhibitors of
PTP1B. The most active molecule was minimized using simulated annealing dynamics and
considered as the bioactive template conformation for molecular alignment purposes.
Database-inertial alignment was followed for aligning the molecules. Nair and Sobhia 66
combined CoMFA 3D-QSAR modeling, performed on pyridazine PTP1B inhibitors, with de
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
13
novo ligand design via the software LeapFrog to generate hypothetical binding cavity using
CoMFA maps. Novel virtual ligands were optimized using this concept.
Cheng et al. 67 developed two hologram quantitative structure-activity relationships (HQSAR)
models using two series of PTP1B inhibitors. They screened two virtual combinatorial libraries
using the optimal HQSAR models and identified several potential PTP-1B inhibitors.
Ma et al. 68 developed 3D-QSAR pharmacophore models for PTP1B and TCPTP inhibitors
using the CATALYST-HYPOGEN module of DiscoveryStudio software suite. Virtual
screening and in vitro evaluation revealed new selective inhibitors of PTP1B over TCPTP
(Figure 3).
Taha et al. 69 Implemented a novel combination of unsupervised pharmacophore modeling
(using HipHop-REFINE software) and classical QSAR analysis, via genetic algorithm and
multiple linear regression, to model a group of diverse PTP-1B inhibitors. There optimal
QSAR model was as described by Equation 2.
Log(1/IC50) = – 2.037 + 0.071 Hypol + 0.195 AlogP98 + 0. 0253 SsOH +0.327 SsssCH
– 0.0067 SsCl2 + 0.0047 CHI-V-12 Equation 2
Hypol represents the fit values of the training compounds against the best pharmacophore
model, SsOH, SsssCH and SsCl are the sums of topological E-state values of hydroxyl,
trivalent carbon and chloro substitutions, respectively. CHI-V-1 is the first-order valence
connectivity index. AlogP98 is the calculated logarithm of partition coefficient from the
implementation of the atom-type-based method using the latest published set of parameters.
They used their QSAR model and associated pharmacophoric hypothesis to identify five new
h-PTP 1B inhibitors retrieved from the National Cancer Institute (NCI) database (Figure 3).
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
14
4.3. Dipeptidyl peptidase IV (DPP-IV)
The glucagon-like peptide 1 (GLP-1) is an incretin hormone released from the stomach during
meals, and serves as an enhancer of glucose-stimulated insulin release from pancreatic β-cells.
GLP-1 is rapidly degraded in the plasma by the serine protease dipeptidyl peptidase IV ( DPP-
IV) 70. Dipeptidyl peptidase IV is a type II membrane glycoprotein that is expressed in a
variety of cell types, such as T cells, B cells, natural killer cells and monocytes. Animal studies
and initial clinical trials have shown that DPP-IV inhibitors are capable of significantly
lowering fasting and postprandial glucose concentration, with minimal risk of hypoglycaemia
71. Therefore, inhibition of DPP-IV has emerged as an attractive target in the treatment of
T2DM prompting several QSAR-based studies to probe these inhibitors.
Paliwal et al. 72 compared QSAR results produced from 47 pyrrolidine DPP-IV inhibitors via
multiple linear regression versus those produced by partial least squares. The author's analyses
suggested comparable statistical results from both methods. The models propose certain role
played by shape flexibility index, Ipso atom E-state index and electrostatic parameters (e.g.,
dipole moment) in the anti- DPP-IV bioactivities.
Patil et al. 73 attempted hologram QSAR (HQSAR) modelling on 34 β-amino amide DPP-IV
inhibitors. The resulting HQSAR maps were helpful in identifying the structural features of
ligands that enhance or reduce anti- DPP-IV bioactivity.
3D-QSAR studies on DPP-IV inhibitors focused mainly on CoMFA and CoMSIA modelling.
Jiang 74 constructed CoMFA models based on a set of arylmethylamine DPP-IV inhibitors
using certain potent inhibitor as template molecule. The conformation of the template molecule
was determined by molecular dynamic and mechanic (MD/MM) simulations. Murugesan et al.
75 performed CoMFA and CoMSIA modelling on a list of pyrrolidine analogues to demark the
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
15
structural requirements of DPP-IV active site. Their CoMFA model has shown that steric and
electrostatic fields have major contributions in binding. However, their CoMSIA model has
shown four molecular fields to impact affinity, namely, steric, electrostatic, hydrogen bond-
donor, and hydrogen bond-acceptor. Jiang 74 performed CoMFA, CoMSIA and docking studies
on DPP-IV inhibitors. Their studies suggested the importance of steric, electrostatic fields,
hydrophobic and hydrogen-bond donor/acceptor fields for DPP-IV inhibitory activity. Saqib
and Siddiqi 76 carried out CoMFA and CoMSIA modelling on triazolopiperazine amide DPP-
IV inhibitors. The study suggests that incorporating bulky and electropositive groups above the
triazolopiperizine ring while adding electronegative and hydrogen bond acceptor groups below
the triazolopiperizine ring enhance the activity.
Jain and Ghate 77 generated a five-point pharmacophore model comprised of two hydrogen
bond acceptors, a hydrophobic, a positive, and an aromatic ring features using azabicyclo DPP-
IV inhibitors. Subsequently, they used pharmacophore-based alignment for CoMFA and
CoMSIA modelling. Subsequent pharmacophore-based virtual screening retrieved seven
virtual DPP-4 inhibitory hits. Zeng et al. 78 implemented receptor (docking)-based alignment to
generate CoMFA and CoMSIA models on a series of fluoropyrrolidine amides DPP-IV
inhibitors. The modelled compounds were docked into the binding site of DPP-IV using
GOLD flexible docking. Pissurlenkar et al. 79 implemented receptor (docking)-based alignment
to generate CoMFA and CoMSIA models using 99 molecules based on phenylalanine,
thiazolidine, and fluorinated pyrrolidine scaffolds. The models were used to design some new
virtual inhibitors.
Agrawal et al. 80 modelled aminomethyl-phenyl-1-isoquinolone DPP-IV inhibitors into 3D-
QSAR-based pharmacophore using PHASE module of Schrodinger software. The
pharmacophore model included one hydrogen bond acceptor, one positive and two aromatic
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
16
ring features. The model suggests that hydrogen bond acceptor aromatic ring and hydrophobic
features are required for DPP-IV inhibitory activity.
Al-masri et al. 81 Explored the pharmacophoric space of DPP-IV 358 known DPP-IV
inhibitors. They then used genetic algorithm and multiple linear regression analysis to select
combination of pharmacophoric models and physicochemical descriptors to build QSAR
model. Two orthogonal pharmacophores emerged in the QSAR equation (Equation 3)
suggesting the existence of at least two distinct binding modes accessible to ligands within the
DPP-IV binding pocket. Their optimal QSAR model is as follows:
where Hypo32/8 and Hypo4/10 represent the two pharmacophores. Jurs-RNCG is the charge of
most negative atom divided by the total negative charge of the molecule. Apol is the sum of
atomic polarizabilities. κ3α and κ3 are the third order and third order alpha-modified Kier’s
shape indices, respectively.
χ
1
is the first order Kier and Hall connectivity index. SssCH2 and
SssssC are electrotopological state sum indices for methylene and quaternary carbon atoms,
respectively. AtypeC1, AtypeC2, AtypeO58, AtypeO60, AtypeF81 and AtypeH47 are atom-
type-based descriptors encoding for the hydrophobic contributions of individual atoms
(Carbon, Oxygen, Fluorine and Hydrogen atoms, respectively). The study identified the
fluoroquinolone gemifloxacin as potent DPP-IV inhibitor (Figure 3).
4.4. Sodium-dependent glucose cotransporter 2 (SGLT2)
Log (1/IC
50
) = ─ 2.290 + 0.024 (Hypo32/8)2 + 0.126 [Hypo4/10 ─ 3.569] ─ 17.099 [JursRNCG ─ 0.121]
+ 6 x10-4 [1.022x104─ Apol] + 0.181 (κ3α)2 ─ 0.159 (κ3)2 + 0.009 (
χ
1
)2 ─ 0.195 [12.86 ─
χ
1
] ─ 0.187
[SssCH2 ─ 1.037] + 0.082 [SssssC + 11.00] ─ 0.104 (AtypeO60)2
─ 0.293 (AtypeF81)2 ─ 0.544 [2 ─ AtypeO58] + 0.708 [6.00 ─ AtypeH47]
─ 1.508 [AtypeC2 – 7.00] ─ 0.167 (AtypeC1) Equation 3
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
17
SGLT2 a membrane protein that acts as high-capacity, low-affinity transporter responsible for
the reabsorption of 90% of renal glucose in human kidneys. It has recently emerged as a novel
target for treatment of diabetes. Canagliflozin was the first SGLT-2 inhibitor approved by the
FDA. Dapagliflozin and empagliflozin were approved later in 2014 82, 83. Interest in SGLT2
prompted different groups to model this target using QSAR techniques.
Prasoona et al. 84 used support vector machines (SVM) with optimized Gaussian kernel
function to identify non-linear QSAR across 40 SGLT2 inhibitory molecules. The study
identified atomic van der Waals volumes, atomic masses, sum of geometrical distances
between oxygen/sulfur atoms and the degree of unsaturation as crucial structural requirements
for SGLT2 inhibitory potencies.
Several 3D-QSAR studies on SGLT2 inhibitors were recently published; however, they were
all focused on CoMFA and CoMSIA modelling. Xu et al. 85 carried out selectivity CoMFA and
CoMSIA studies to highlight the structure requirements for highly selective SGLT2 inhibitors.
They also performed homology modelling, molecular dynamics simulation and binding free
energy calculations of SGLT2 and SGLT1 of potent and selective ligands. Their studies
collectively showed that the stretch of the methylthio group on Met241 had an essential effect
on the different binding modes between SGLT1 and SGLT2. Hydrogen bond analysis and
binding free energy calculations suggested that SGLT2-ligand complex is more stable than the
corresponding SGLT1 complex. Similarly, Zhi et al. 86 constructed CoMFA and CoMSIA
models on 46 SGLT-2 inhibitors. The contour maps from these models provided insights into
the requirements for more active SGLT-2 inhibitors. Vyas et al. 87 applied CoMFA and
CoMSIA studies on 180 C-aryl glucoside SGLT2 inhibitors. They implemented 3 different
alignment strategies. The best CoMFA and CoMSIA models were obtained through Distill
rigid body alignment. Nakka and Guruprasad 88 generated a 3D homology model for human
SGLT2. They then docked certain potent C-aryl glucoside SGLT2 inhibitor into their
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
18
homology model and used the docked pose to guide molecular field analysis/genetic partial
least-squares (MFA-G/PLS) on 43 C-aryl glucoside SGLT2 inhibitors.
Suryanarayanan et al. 89 used PHASE (from Schroedinger Inc.) to perform pharmacophore and
atom-based 3D-QSAR modelling on a series of N-β-D-xylosylindole SGLT2 inhibitors. They
reported a six-featured pharmacophore model exhibiting two hydrogen bond acceptors, two
hydrogen bond donors, one hydrophobic, and one aromatic ring.
Tang et al. 90 generated SGLT2 pharmacophoric model using Discovery Studio V2.1 (from
Biovia Inc.). Their model included one hydrogen bond donor, five excluded volumes, one ring
aromatic and three hydrophobic features. The group found that SGLT2 pharmacophore exhibits
distinct features from those of SGLT1 pharmacophore. The authors used their SGLT2
pharmacophore as 3D query to screen NCI and Maybridge structural databases to identify new
potential inhibitors.
4.5. Peroxisome proliferator-activated receptor gamma (PPAR-γ)
Peroxisome proliferator-activated receptors (PPAR) are considered attractive molecular targets
for developing new anti-diabetes drugs 91. These receptors are fatty acid-activated transcription
factors that belong to the nuclear hormone receptor family. They play pivotal role in glucose
and lipid homeostasis. There are three PPAR subtypes designated as PPAR-α , PPAR-γ, and
PPAR-δ 92, 93. However, PPAR-γ, which is expressed in adipose tissue, lower intestine, and
cells involved in immunity, is the most extensively investigated PPAR. Moreover, PPARδ,
which regulates several metabolic processes, has also been investigated for development of
new drugs for treating diabetes mellitus 94, 95. Several QSAR models were attempted for PPAR
full agonists with the ultimate aim of developing insulin-sensitizing PPAR modulators of
minimal classical activation of PPAR and hence of reduced side effects.
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
19
Khanna et al. 96 introduced additivity of molecular fields to correlate molecular fields of dual
activators and their pIC50 values. The authors tested their concept on PPARα and PPARγ dual
activators. Three CoMFA models namely α-model, γ-model and dual-model have been
developed. They validated this concept by comparing contour maps, CoMFA results with
FlexX docking results and analyzing some newly designed molecules.
Maltarollo et al. 97 carried out classical QSAR modelling on a set of PPARδ agonists. They
found that energy of the lowest unoccupied molecular orbital, dipole moment, atomic charge at
certain carbon atom, stereochemical volume, lipophilic partition coefficient, atomic
polarizability and aromaticity influence bioactivity. Conclusions from docking studies seemed
to agreed with their QSAR model. Giaginis et al. 98 constructed classical QSAR models
employing tyrosine-based PPAR-γ agonists using principle component analysis and partial
least squares. Their models suggested that molecular weight, rotatable bonds and lipophilicity
exert positive influence on PPAR-γ agonistic bioactivity, while excess negative and positive
charges created disfavoured bioactivity.
Sundriyal and Bharatam 99, 100 developed the sum CoMFA concept to design Dual PPARα/γ or
pan PPARa/g/d agonists by using the sum of activities of compounds against individual
subtypes as a dependent parameter. The generated models were found to be statistically
significant.
Liao et al. 101 employed CoMFA and CoMSIA on a set of thiazolidinediones- and tyrosine-
based PPARγ agonists. To obtain consistent alignment, they used the co-crystallized structures
of rosiglitazone and farglitazar within PPARγ (from the Protein Data Bank) as alignment
templates. The authors concluded some SAR rules from their study. Similarly, Garcia et al. 102
performed classical QSAR, hologram QSAR and CoMFA on oxazole-based PPAR-δ agonists.
Gee et al. 103 combined QSAR modelling and docking to virtually screen the ZINC library.
Their QSAR model was based on 1,517 compounds (including 177 full PPARδ agonists).
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
20
Upon using the optimal QSAR model as search query it identified 42,378 potential PPARδ
agonists from the ZINC library. Subsequent docking study identified 30 potentially active
ligands. Four of which were tested in vitro, and one of them was found to have Ki <5 μM
(Figure 3).
Guasch et al. 104 implemented docking-based alignment to build atom-based 3D-QSAR models
for tyrosine-based PPARγ agonists using the software PHASE from Schrödinger LLC.
Al-Najjar et al. 105 explored the pharmacophoric space of 88 PPAR-γ activators. They then
used classical QSAR modelling to identify the best combination of pharmacophore models and
2D physicochemical descriptors capable of explaining PPAR-γ bioactivation. Three orthogonal
pharmacophores emerged in the QSAR equation (Equation 4). The following equation shows
their best QSAR model:
Log(1/IC50) = - 2.4 + 0.30(S1R5H8) + 5.63x10-2 (S3R2H2)+ 9.76x10-2(S1R3H1)+5.1FPSA -
9.3LUMO+0.05(Rotable Bonds)-0.28HBD Equation 4
Where S1R3H1, S1R5H8 and S3R2H2 represent the optimal QSAR-selected pharmacophoric
models, HBD is the number of hydrogen bond donors in a particular molecule, RotableBonds
is the number of rotable bonds, FPSA is the molecular fractional polar surface area, and
LUMO is the energy of the lowest unoccupied molecular orbital calculated employing the
density functional theory method. Subsequent screening of the national cancer institute (NCI)
list of compounds captured new hits with the most potent showing EC50 values of 15 and
224 nM (Figure 3).
5. Expert opinion
Despite the extensive research focusing on the discovery of novel antidiabetic agents, there is
still unmet medical need for T2DM management and treatment. The implementation of
computational methods in early-stage drug discovery has contributed significantly to drug
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
21
development in a cost-effiecient way. In particular, ligand-based design methods in
combination with other drug discovery tools has impacted the discovery of many drugs
available in the market. QSAR has become a routine tool in ligand-based drug discovery
projects for the analysis of large sets of candidate molecules (Figure 4).
Clearly, the role of QSAR in selecting the best candidate molecules from large compounds
libraries and in identifying molecular features that control activity represent key contributions
in the drug discovery field. QSAR modelling contributes to discovery of initial hits and hit-to-
lead optimization.
Still, it is not clear from the literature whether the implementation of QSAR–based methods
have solely contributed to the discovery of any experimental or approved new antidiabetic
agents. In fact, most published QSAR studies, including QSAR studies on anti-T2DM agents,
were rarely validated by discovery of new bioactive hits. Instead, authors tend to heavily rely
on theoretical and statistical means to validate their QSAR models. Incidentally, this trend
flooded the literature with QSAR publications devoid of their fundamental goal, i.e., discovery
and optimization of new bioactive hits. We believe this situation must be corrected by
requesting researchers to experimentally validate their QSAR models by testing them as
screening and/or opitmization tools.
It is also noticeable that most of QSAR studies have focused on a single drug-target. Therefore,
a true paradigm shift in anti-diabetic drug discovery will require the implementation of multi-
target models at an early-stage discovery efforts including QSAR modelling.
Current QSAR modelling efforts lack any focus on absorption, distribution, metabolism and
elimination (ADME) issues in the final QSAR models. However, since pharmacokinetic issues
are the major source of attrition in drug discovery pipelines, it is highly recommended to
accommodate pharmacokinetic factors in QSAR modelling of anti-diabetic agents. We suggest
that this can be done by using ADME-curated training lists for QSAR modelling. That is,
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
22
selecting training lists that satisfy desirable pharmacokinetic descriptor-ranges for QSAR
modelling (e.g., Lipiski's violations, Veber's rules, etc…).
Anti-diabetic targets have been extensively explored by different QSAR methods. These
methods range from direct 2D QSAR and 3D QSAR models to more elaborate studies
combining QSAR with structure-based (e.g., docking, molecular dynamics simulation, etc…)
and pharmacophore models. These hybride studies are expected to be more fruitful in enriching
our understanding of the molecular bases of anti-diabetic agents. Still, any tangible effects of
such studies remain to be seen in the future.
Sophisticated descriptors have been developed to characterize the 3D-geometry and chemistry
of small molecules. Researchers tend to rely more on 3D descriptors, however some studies
combined both 2D and 3D descriptors. Still, QSAR studies on anti-diabetic agents are devoid
of statistical comparisons that compare knowledge gained from 2D versus 3D descriptors when
applied separately. Accordingly, we recommend that future QSAR modelling studies of
antidiabetic agents should compare knowledge gained from 2D QSAR versus 3D QSAR
modelling.
In conclusion, QSAR modelling holds excellent potential as means of reducing attrition in anti-
diabetes drug discovery workflows by focusing on novel hit molecules that exhibit optimal
structural features necessary for binding to targeted anti-diabetic receptors. Nevertheless,
additional future developments can still be performed to refine QSAR modelling efforts in this
field and optimize their outcomes.
Financial and Competing Interests Disclosure
The authors have no relevant affiliations or financial involvement with any organization or
entity with a financial interest in or financial conflict with the subject matter or materials
discussed in the manuscript. This includes employment, consultancies, honoraria, stock
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
23
ownership or options, expert testimony, grants or patents received or pending, or royalties.
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
24
References
1. WHO. WHO Factsheets. http://wwwwhoint/mediacentre/factsheets/fs312/en/ 2015.
2. Guariguata L, Whiting D, Hambleton I, et al. Global estimates of diabetes prevalence
for 2013 and projections for 2035. Diabetes research and clinical practice 2014;103(2):137-49.
3. de Boer IH. Kidney disease and related findings in the diabetes control and
complications trial/epidemiology of diabetes interventions and complications study. Diabetes
care 2014;37(1):24-30.
4. Aiello LP. Diabetic retinopathy and other ocular findings in the diabetes control and
complications trial/epidemiology of diabetes interventions and complications study. Diabetes
care 2014;37(1):17-23.
5. Huang ES, Laiteerapong N, Liu JY, et al. Rates of complications and mortality in older
patients with diabetes mellitus: the diabetes and aging study. JAMA internal medicine
2014;174(2):251-58.
6. Dart AB, Martens PJ, Rigatto C, et al. Earlier onset of complications in youth with type
2 diabetes. Diabetes care 2014;37(2):436-43.
7. Ahrén B. Creative use of novel glucose-lowering drugs for type 2 diabetes: where will
we head in the next 50 years? Diabetologia 2015:1-5.
8. Dutta S. APPLICATION OF QSAR IN DRUG DESIGN AND DRUG DISCOVERY.
drugs 2015;9:10.
9. Vaidya A, Jain S, Jain S, et al. Quantitative Structure-Activity Relationships: A Novel
Approach of Drug Design and Discovery. Journal of Pharmaceutical Sciences and
Pharmacology 2014;1(3):219-32.
10. Boussageon R, Bejan-Angoulvant T, Saadatian-Elahi M, et al. Effect of intensive
glucose lowering treatment on all cause mortality, cardiovascular death, and microvascular
events in type 2 diabetes: meta-analysis of randomised controlled trials. Bmj 2011;343.
11. Plutzky J, Viberti G, Haffner S. Atherosclerosis in type 2 diabetes mellitus and insulin
resistance: mechanistic links and therapeutic targets. Journal of Diabetes and its Complications
2002;16(6):401-15.
12. Krentz AJ, Bailey CJ. Oral antidiabetic agents: current role in type 2 diabetes mellitus.
drugs 2005;65(3):385-411.
13. Stein SA, Lamos EM, Davis SN. A review of the efficacy and safety of oral antidiabetic
drugs. Expert opinion on drug safety 2013;12(2):153-75.
● This review discusses important validation methods for QSAR models
14. Roy K, Mitra I. On various metrics used for validation of predictive QSAR models with
applications in virtual screening and focused library design. Combinatorial chemistry & high
throughput screening 2011;14(6):450-74.
15. Taha MO, Habash M, Al-Hadidi Z, et al. Docking-based comparative intermolecular
contacts analysis as new 3-D QSAR concept for validating docking studies and in silico
screening: NMT and GP inhibitors as case studies. Journal of chemical information and
modeling 2011;51(3):647-69.
16. Al-Sha’er MA, Taha MO. Application of docking-based comparative intermolecular
contacts analysis to validate Hsp90α docking studies and subsequent in silico screening for
inhibitors. Journal of molecular modeling 2012;18(11):4843-63.
17. Jaradat N, Khanfar M, Habash M, et al. Combining docking-based comparative
intermolecular contacts analysis and k-nearest neighbor correlation for the discovery of new
check point kinase 1 inhibitors. Journal of computer-aided molecular design 2015
2015/06/01;29(6):561-81.
18. Taha M, Habash M, Khanfar M. The use of docking-based comparative intermolecular
contacts analysis to identify optimal docking conditions within glucokinase and to discover of
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
25
new GK activators. Journal of computer-aided molecular design 2014 2014/05/01;28(5):509-
47.
●● This review provides an overview of different feature selection techniques applied in
QSAR modeling
19. Goodarzi M, Dejaegher B, Heyden YV. Feature Selection Methods in QSAR Studies.
Journal of AOAC International 2012;95(3):636-51.
20. Pratim Roy P, Paul S, Mitra I, et al. On Two Novel Parameters for Validation of
Predictive QSAR Models. Molecules 2009;14(5):1660.
21. Roy K, Chakraborty P, Mitra I, et al. Some case studies on application of “rm2” metrics
for judging quality of quantitative structure–activity relationship predictions: Emphasis on
scaling of response data. Journal of computational chemistry 2013;34(12):1071-82.
●● The following three articles clarify good practices in QSAR modelling
22. Golbraikh A, Tropsha A. Beware of q2! Journal of Molecular Graphics and Modelling
2002;20(4):269-76.
23. Tropsha A. Best Practices for QSAR Model Development, Validation, and
Exploitation. Molecular Informatics 2010;29(6-7):476-88.
24. Fourches D, Muratov E, Tropsha A. Trust, but verify: on the importance of chemical
structure curation in cheminformatics and QSAR modeling research. Journal of chemical
information and modeling 2010;50(7):1189-204.
● The following four articles deal with QSAR modelability of compound datasets and
activity cliffs
25. Koutsoukas A, Paricharak S, Galloway WRJD, et al. How Diverse Are Diversity
Assessment Methods? A Comparative Analysis and Benchmarking of Molecular Descriptor
Space. Journal of chemical information and modeling 2014 2014/01/27;54(1):230-42.
26. Golbraikh A, Muratov E, Fourches D, et al. Data Set Modelability by QSAR. Journal of
chemical information and modeling 2014 2014/01/27;54(1):1-4.
27. Guha R, Van Drie JH. Structure−Activity Landscape Index: Identifying and
Quantifying Activity Cliffs. Journal of chemical information and modeling 2008
2008/03/01;48(3):646-58.
28. Seebeck B, Wagener M, Rarey M. From Activity Cliffs to Target-Specific Scoring
Models and Pharmacophore Hypotheses. ChemMedChem 2011;6(9):1630-39.
29. Stumpfe D, Hu Y, Dimova D, et al. Recent Progress in Understanding Activity Cliffs
and Their Utility in Medicinal Chemistry: Miniperspective. Journal of medicinal chemistry
2013;57(1):18-28.
30. Hu Y, Stumpfe D, Bajorath J. Advancing the activity cliff concept. F1000Research
2013;2.
●● The following three articles discuss pitfalls and errors in QSAR modelling
31. Dearden J, Cronin M, Kaiser K. How not to develop a quantitative structure–activity or
structure–property relationship (QSAR/QSPR). SAR and QSAR in environmental research
2009;20(3-4):241-66.
32. Cramer RD. The inevitable QSAR renaissance. Journal of computer-aided molecular
design 2012;26(1):35-38.
33. Cherkasov A, Muratov EN, Fourches D, et al. QSAR modeling: where have you been?
Where are you going to? Journal of medicinal chemistry 2014;57(12):4977-5010.
34. Kulkarni SS, Gediya LK, Kulkarni VM. Three-dimensional quantitative structure
activity relationships (3-D-QSAR) of antihyperglycemic agents. Bioorganic & medicinal
chemistry 1999;7(7):1475-85.
35. Kurogi Y. Three-dimensional quantitative structure-activity relationships (3D-QSAR)
of antidiabetic thiazolidinediones. Drug design and discovery 1999;16(2):109-18.
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
26
36. So S-S, Karplus M. A comparative study of ligand-receptor complex binding affinity
prediction methods based on glycogen phosphorylase inhibitors. Journal of computer-aided
molecular design 1999;13(3):243-58.
37. Bharatam P, Patel D, Adane L, et al. Modeling and informatics in designing anti-
diabetic agents. Current pharmaceutical design 2007;13(34):3518-30.
38. Luan F, Xu X, Liu HT, et al. QSAR studies of PTP1B inhibitors: recent advances and
perspectives. Current medicinal chemistry 2012;19(25):4208-17.
39. Alexiou P, Pegklidou K, Chatzopoulou M, et al. Aldose reductase enzyme and its
implication to major health problems of the 21st century. Current medicinal chemistry
2009;16(6):734-52.
40. Khanna S, Bahal R, Bharatam PV. In silico Studies on PPARγ Agonistic Heterocyclic
Systems. QSAR and Molecular Modeling Studies in Heterocyclic Drugs I: Springer 2006:149-
80.
41. M Hayes J, D Leonidas D. Computation as a tool for glycogen phosphorylase inhibitor
design. Mini reviews in medicinal chemistry 2010;10(12):1156-74.
42. Gokhale KM. Role of Glycogen Synthase Kinase (GSK-3) in Type-2 Diabetes and
GSK-3 Inhibitors as Potential Anti-Diabetics. International Journal of Pharmaceutical and
Phytopharmacological Research 2014;3(3).
43. Yarchoan M, Arnold SE. Repurposing diabetes drugs for brain insulin resistance in
Alzheimer disease. Diabetes 2014;63(7):2253-61.
44. Arfeen M, Bharatam PV. Design of glycogen synthase kinase-3 inhibitors: an overview
on recent advancements. Current pharmaceutical design 2013;19(26):4755-75.
45. Takahashi-Yanaga F. Activator or inhibitor? GSK-3 as a new drug target. Biochemical
pharmacology 2013;86(2):191-99.
46. Avrahami L, Licht-Murava A, Eisenstein M, et al. GSK-3 inhibition: achieving
moderate efficacy with high selectivity. Biochimica et Biophysica Acta (BBA)-Proteins and
Proteomics 2013;1834(7):1410-14.
47. Davis MI, Hunt JP, Herrgard S, et al. Comprehensive analysis of kinase inhibitor
selectivity. Nature biotechnology 2011;29(11):1046-51.
48. Patel DS, Bharatam PV. Selectivity criterion for pyrazolo[3,4-b]pyrid[az]ine
derivatives as GSK-3 inhibitors: CoMFA and molecular docking studies. European journal of
medicinal chemistry 2008 May;43(5):949-57.
49. Fang J, Huang D, Zhao W, et al. A new protocol for predicting novel GSK-3beta ATP
competitive inhibitors. Journal of chemical information and modeling 2011 Jun 27;51(6):1431-
8.
50. Akhtar M, Bharatam PV. 3D-QSAR and molecular docking studies on 3-anilino-4-
arylmaleimide derivatives as glycogen synthase kinase-3beta inhibitors. Chemical biology &
drug design 2012 Apr;79(4):560-71.
51. Prasanna S, Daga PR, Xie A, et al. Glycogen synthase kinase-3 inhibition by 3-anilino-
4-phenylmaleimides: insights from 3D-QSAR and docking. Journal of computer-aided
molecular design 2009 Feb;23(2):113-27.
52. Dessalew N, Patel DS, Bharatam PV. 3D-QSAR and molecular docking studies on
pyrazolopyrimidine derivatives as glycogen synthase kinase-3beta inhibitors. Journal of
molecular graphics & modelling 2007 Mar;25(6):885-95.
53. Crisan L, Pacureanu L, Avram S, et al. PLS and shape-based similarity analysis of
maleimides--GSK-3 inhibitors. Journal of enzyme inhibition and medicinal chemistry 2014
Aug;29(4):599-610.
54. Lather V, Kristam R, Saini JS, et al. QSAR Models for Prediction of Glycogen
Synthase Kinase-3β Inhibitory Activity of Indirubin Derivatives. QSAR & Combinatorial
Science 2008;27(6):718-28.
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
27
55. Fu G, Liu S, Nan XF, et al. Quantitative Structure-Activity Relationship Analysis and a
Combined Ligand-Based/Structure-Based Virtual Screening Study for Glycogen Synthase
Kinase-3. Molecular Informatics 2014 Sep;33(9):627-40.
56. Taha MO, Bustanji Y, Al-Ghussein MAS, et al. Pharmacophore Modeling, Quantitative
Structure–Activity Relationship Analysis, and in Silico Screening Reveal Potent Glycogen
Synthase Kinase-3β Inhibitory Activities for Cimetidine, Hydroxychloroquine, and
Gemifloxacin. Journal of medicinal chemistry 2008 2008/04/01;51(7):2062-77.
57. He R-j, Yu Z-h, Zhang R-y, et al. Protein tyrosine phosphatases as potential therapeutic
targets. Acta Pharmacologica Sinica 2014.
58. Gurzov EN, Stanley WJ, Brodnicki TC, et al. Protein tyrosine phosphatases: molecular
switches in metabolism and diabetes. Trends in Endocrinology & Metabolism 2015;26(1):30-
39.
59. Tamrakar AK, Maurya CK, Rai AK. PTP1B inhibitors for type 2 diabetes treatment: a
patent review (2011-2014). Expert opinion on therapeutic patents 2014;24(10):1101-15.
60. Sachan N, Kadam SS, Kulkarni VM. Human protein tyrosine phosphatase 1B
inhibitors: QSAR by genetic function approximation. Journal of enzyme inhibition and
medicinal chemistry 2007 Jun;22(3):267-76, 371-3.
61. Malla P, Kumar R, Kumar M. Validation of formylchromane derivatives as protein
tyrosine phosphatase 1B inhibitors by pharmacophore modeling, atom-based 3D-QSAR and
docking studies. Chemical biology & drug design 2013 Jul;82(1):71-80.
62. Malla P, Kumar R, Mattewal SK, et al. A paradigm for development of novel PTP 1B
inhibitors: Pharmacophore modelling, atom-based 3D-QSAR and docking studies. Med Chem
Res 2014 Feb;23(2):927-38.
63. Thareja S, Kokil GR, Aggarwal S, et al. Sulphonamides as inhibitors of protein tyrosine
phosphatase 1B: a three-dimensional quantitative structure-activity relationship study using
self-organizing molecular field analysis approach. Chemical & pharmaceutical bulletin 2010
Apr;58(4):526-32.
64. Thareja S, Aggarwal S, Bhardwaj T, et al. Self-organizing molecular field analysis of 2,
4-thiazolidinediones: A 3D-QSAR model for the development of human PTP1B inhibitors.
European journal of medicinal chemistry 2010;45(6):2537-46.
65. Sobhia ME, Bharatam PV. Comparative molecular similarity indices analysis
(CoMSIA) studies of 1,2-naphthoquinone derivatives as PTP1B inhibitors. Bioorganic &
medicinal chemistry 2005;13(6):2331-38.
66. Nair PC, Sobhia ME. CoMFA based de novo design of pyridazine analogs as PTP1B
inhibitors. Journal of molecular graphics & modelling 2007 Jul;26(1):117-23.
67. Cheng Y, Zhou M, Tung CH, et al. Studies on two types of PTP1B inhibitors for the
treatment of type 2 diabetes: Hologram QSAR for OBA and BBB analogues. Bioorganic &
medicinal chemistry letters 2010 Jun 1;20(11):3329-37.
68. Ma Y, Jin YY, Wang YL, et al. The discovery of a novel and selective inhibitor of
PTP1B over TCPTP: 3D QSAR pharmacophore modeling, virtual screening, synthesis, and
biological evaluation. Chemical biology & drug design 2014 Jun;83(6):697-709.
69. Taha MO, Bustanji Y, Al-Bakri AG, et al. Discovery of new potent human protein
tyrosine phosphatase inhibitors via pharmacophore and QSAR analysis followed by in silico
screening. Journal of Molecular Graphics and Modelling 2007;25(6):870-84.
70. Potenza M, Rayfield EJ. Targeting the Incretin System in Type 2 Diabetes Mellitus.
Mount Sinai Journal of Medicine: A Journal of Translational and Personalized Medicine
2009;76(3):244-56.
71. Karagiannis T, Paschos P, Paletas K, et al. Dipeptidyl peptidase-4 inhibitors for
treatment of type 2 diabetes mellitus in the clinical setting: systematic review and meta-
analysis. Bmj 2012;344.
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
28
72. Paliwal S, Seth D, Yadav D, et al. Development of a robust QSAR model to predict the
affinity of pyrrolidine analogs for dipeptidyl peptidase IV (DPP- IV). Journal of enzyme
inhibition and medicinal chemistry 2011 Feb;26(1):129-40.
73. Patil S, Sharma R, Abhishek A. Comparative Study to Predict Dipeptidyl Peptidase IV
Inhibitory Activity of β-Amino Amide Scaffold. Indian journal of pharmaceutical sciences
2015 Mar-Apr;77(2):142-50.
74. Jiang YK. Molecular docking and 3D-QSAR studies on beta-phenylalanine derivatives
as dipeptidyl peptidase IV inhibitors. Journal of molecular modeling 2010 Jul;16(7):1239-49.
75. Murugesan V, Sethi N, Prabhakar YS, et al. CoMFA and CoMSIA of diverse
pyrrolidine analogues as dipeptidyl peptidase IV inhibitors: active site requirements. Mol
Divers 2011;15(2):457-66.
76. Saqib U, Siddiqi MI. 3D-QSAR studies on triazolopiperazine amide inhibitors of
dipeptidyl peptidase-IV as anti-diabetic agents. SAR and QSAR in environmental research
2009 2009/07/01;20(5-6):519-35.
77. Jain SV, Ghate M. Atom-based pharmacophore modeling, CoMFA/CoMSIA-based 3D-
QSAR studies and lead optimization of DPP-4 inhibitors for the treatment of type 2 diabetes.
Med Chem Res 2014;23(7):3436-50.
78. Zeng J, Liu G, Tang Y, et al. 3D-QSAR studies on fluoropyrrolidine amides as
dipeptidyl peptidase IV inhibitors by CoMFA and CoMSIA. Journal of molecular modeling
2007 Sep;13(9):993-1000.
79. Pissurlenkar RR, Shaikh MS, Coutinho EC. 3D-QSAR studies of Dipeptidyl peptidase
IV inhibitors using a docking based alignment. Journal of molecular modeling 2007
Oct;13(10):1047-71.
80. Agrawal R, Jain P, Narayan Dikshit S, et al. Atom-based 3D-QSAR and Docking
Studies of Various 3-aminomethyl-1, 2-dihydro-4-phenyl-1-isoquinolones Derivatives as an
Effective DPP-IV Inhibitors and its Validation. Immunology, Endocrine & Metabolic Agents
in Medicinal Chemistry (Formerly Current Medicinal Chemistry-Immunology, Endocrine and
Metabolic Agents) 2013;13(3):221-31.
81. Al-masri IM, Mohammad MK, Taha MO. Discovery of DPP IV Inhibitors by
Pharmacophore Modeling and QSAR Analysis followed by in silico Screening.
ChemMedChem 2008;3(11):1763-79.
82. Gallo LA, Wright EM, Vallon V. Probing SGLT2 as a therapeutic target for diabetes:
Basic physiology and consequences. Diabetes and Vascular Disease Research 2015;12(2):78-
89.
83. Ghosh RK, Ghosh SM, Chawla S, et al. SGLT2 inhibitors: a new emerging therapeutic
class in the treatment of type 2 diabetes mellitus. The Journal of Clinical Pharmacology
2012;52(4):457-63.
84. Prasoona RK, Jyoti A, Mukesh Y, et al. Optimization of Gaussian Kernel Function in
Support Vector Machine aided QSAR studies of C-aryl glucoside SGLT2 inhibitors.
Interdisciplinary sciences, computational life sciences 2013 Mar;5(1):45-52.
85. Xu J, Yuan H, Ran T, et al. A selectivity study of sodium-dependent glucose
cotransporter 2/sodium-dependent glucose cotransporter 1 inhibitors by molecular modeling.
Journal of molecular recognition : JMR 2015 Mar 5.
86. Zhi H, Zheng J, Chang Y, et al. QSAR studies on triazole derivatives as sglt inhibitors
via CoMFA and CoMSIA. Journal of Molecular Structure 2015;1098:199-205.
87. Vyas VK, Bhatt HG, Patel PK, et al. CoMFA and CoMSIA studies on C-aryl glucoside
SGLT2 inhibitors as potential anti-diabetic agents. SAR and QSAR in environmental research
2013;24(7):519-51.
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
29
88. Nakka S, Guruprasad L. Structural Insights into the Active Site of Human Sodium
Dependent Glucose Co-Transporter 2: Homology Modelling, Molecular Docking, and 3D-
QSAR Studies. Australian Journal of Chemistry 2012;65(9):1314-24.
89. Suryanarayanan V, Sudha A, Rajamanikandan S, et al. Atom-based 3D QSAR studies
on novel N-β-d-xylosylindole derivatives as SGLT2 inhibitors. Med Chem Res 2013
2013/02/01;22(2):615-24.
90. Tang C, Zhu X, Huang D, et al. A specific pharmacophore model of sodium-dependent
glucose co-transporter 2 (SGLT2) inhibitors. Journal of molecular modeling 2012
Jun;18(6):2795-804.
91. Ansquer JC, Foucher C. The PPAR System in Diabetes. Lipoproteins in Diabetes
Mellitus: Springer 2014:357-72.
92. Liu Z-M, Hu M, Chan P, et al. Early investigational drugs targeting PPAR-α for the
treatment of metabolic disease. Expert opinion on investigational drugs 2015;24(5):611-21.
93. Ahmadian M, Suh JM, Hah N, et al. PPAR [gamma] signaling and metabolism: the
good, the bad and the future. Nature medicine 2013;99(5):557-66.
94. Maiese K, Chong ZZ, Shang YC, et al. Novel directions for diabetes mellitus drug
discovery. Expert opinion on drug discovery 2013;8(1):35-48.
95. Wright MB, Bortolini M, Tadayyon M, et al. Minireview: challenges and opportunities
in development of PPAR agonists. Molecular endocrinology 2014;28(11):1756-68.
96. Khanna S, Sobhia ME, Bharatam PV. Additivity of molecular fields: CoMFA study on
dual activators of PPARalpha and PPARgamma. Journal of medicinal chemistry 2005 Apr
21;48(8):3015-25.
97. Maltarollo VG, Homem-de-Mello P, Honorio KM. Role of physicochemical properties
in the activation of peroxisome proliferator-activated receptor delta. Journal of molecular
modeling 2011 Oct;17(10):2549-58.
98. Giaginis C, Theocharis S, Tsantili-Kakoulidou A. Quantitative Structure-Activity
Relationships for PPAR-γ Binding and Gene Transactivation of Tyrosine-Based
Agonists Using Multivariate Statistics. Chemical biology & drug design 2008;72(4):257-64.
99. Sundriyal S, Bharatam PV. ‘Sum of activities’ as dependent parameter: A new
CoMFA-based approach for the design of pan PPAR agonists. European journal of medicinal
chemistry 2009;44(1):42-53.
100. Shah P, Mittal A, Bharatam PV. CoMFA analysis of dual/multiple PPAR activators.
European journal of medicinal chemistry 2008;43(12):2784-91.
101. Liao C, Xie A, Zhou J, et al. 3D QSAR studies on peroxisome proliferator-activated
receptor γ agonists using CoMFA and CoMSIA. Journal of molecular modeling
2004;10(3):165-77.
102. Garcia TS, Silva DC, Gertrudes JC, et al. Molecular features related to the binding
mode of PPARdelta agonists from QSAR and docking analyses. SAR and QSAR in
environmental research 2013;24(2):157-73.
103. Gee VM, Wong FS, Ramachandran L, et al. Identification of novel peroxisome
proliferator-activated receptor-gamma (PPARgamma) agonists using molecular modeling
method. Journal of computer-aided molecular design 2014 Nov;28(11):1143-51.
104. Guasch L, Sala E, Valls C, et al. Development of docking-based 3D-QSAR models for
PPARgamma full agonists. Journal of Molecular Graphics and Modelling 2012;36:1-9.
105. Al-Najjar BO, Wahab HA, Tengku Muhammad TS, et al. Discovery of new nanomolar
peroxisome proliferator-activated receptor gamma activators via elaborate ligand-based
modeling. European journal of medicinal chemistry 2011 Jun;46(6):2513-29.
106. Mittermayer F, Caveney E, Oliveira CD, et al. Addressing Unmet Medical Needs in
Type 2 Diabetes: A Narrative Review of Drugs under Development. Current Diabetes Reviews
2015;11(1):17-31.
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
30
107. Taygerly JP, McGee LR, Rubenstein SM, et al. Discovery of INT131: A selective
PPARγ modulator that enhances insulin sensitivity. Bioorganic & medicinal chemistry
2013;21(4):979-92.
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
31
Figure 1: the key organs involved in T2DM and potential drug targets for hypoglycemic agents. The liver: glucagon receptor (antagonists),
glycogen phosphorylase (inhibitors), glucose-6-phosphatase and fructose-1,6-bisphosphatase (inhibitors). Pancreatic islet b-cells: glucagon-like
peptide 1 (GLP-1) mimetic, GLP-1 Receptor (agonists) and dipeptidylpeptidase IV (DPP-IV) (ihibitors). Liver and muscle (and fat): insulin
receptor (agonists), protein tyrosine phosphatase (PTP)-1B (inhibitors), Glycogen synthase kinase 3β (GSK-3β). Kidney: sodium-dependent
glucose cotransporter 2 (SGLT2). Fat tissue: peroxisome proliferator-activated receptor gamma (PPARγ).Intestine: Intestinal lipase and α-
glucosidease. Potential anti-obesity agents that produce reduced appetite and/or increased energy expenditure (Fat tissue, brain and GIT).
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
32
Figure 2: general Steps involved in QSAR modelling.
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
33
N
H
O
S
N
N
H
N
N
Br
N
H
O
S
N
N
H
N
I
O
Compound 2
IC
50
= 7.05 µMCompound 3
IC
50
= 5.38 µM
Fu et al
53
NOH
H
N
Cl
N
Hydroxychloroquine
IC
50
= 33 nM
ONN
H
2
N
N
FO
N
OH
O
Gemifloxacin
IC
50
= 88 nM
N
N
HS
NH
N
N
H
N
Cimetidine
IC
50
= 13 nM
Taha et al
54
HN NH
O
O
OHN NH
O
O
O
Cl
Cl
Compound 4
IC
50
= 4.2 µMCompound 6
IC
50
= 4.6 µM
Ma et al
66
ON
O
O
N
H
COOH
COOH
SO
ON
H
N
H
OCOOH
COOH
O
O
HN
COOH
NH
O
2HN
O
O
N
O
N
H
COOH
COOH
H
N
N
O
O
O
O
HN
COOH
NH
O
2HN
HN
N
O
O
Compound 155
IC
50
= 3.30 µMCompound 156
IC
50
= 3.17 µMCompound 158
IC
50
= 0.47 µM
Compound 159
IC
50
= 1.86 µM
Compound 157
IC
50
= 0.76 µM
Taha et al
67
O
N
N
NH
2
N
F
O
N
HO
O
Gemifloxacin
IC
50
= 1.12 µM
Al-masri et al
79
N
O
N
N
O
O
ZINC40251491
Ki = 4.87 µM
Gee et al
98
SO
O
O
NS
HOOC
O
ON
NS
N
COOH
O
O
Compound 93
EC
50
= 15.3 nM Compound 96
EC
50
= 224 nM
Al-Najjar et al
100
Figure 3: the new scaffold discovered by the application of different QSAR strategies.
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
34
Figure 4: the complementary role of the different drug discovery methodologies.
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
35
Table 1: recently FDA-approved anti-diabetic small molecules.
Trade name
Anti-diabetic agents
Chemical structure
Company
Year of FDA
approval
Mechanism of action
Monotherapies
Farxiga
Dapagliflozin
Cl O
O
HO
HO OH OH
Bristol-Myers Squibb
2014
SGLT2 inhibitor.
Jardiance
Empagliflozin
O
O
O
HO
HO OH OH
Cl
Boehringer Ingelheim
2014
SGLT2 inhibitor.
Invokana
Canagliflozin
SF
O
HO
HO OH OH
Janssen
Pharmaceuticals
2013
SGLT2 inhibitor
Nesina
Alogliptin
N
N
ONO
NNH2
Takeda
2013
DPP-IV inhibitor
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
36
Tradjenta
Linagliptin
H
NH
2
N
N
N
O
N
N
NON
Boehringer Ingelheim
2011
DPP-4 inhibitor
Onglyza
Saxagliptin
H
2
N
O
N
N
OH
Bristol-Myers Squibb
2009
DPP4 inhibitors
Januvia
Sitagliptin phosphate
NH2O
NN
NN
FFF
F
FF
Merck
2006
DPP4 inhibitors
ACTOS
Pioglitazone hydrochloride
NO
SO
NH
O
Takeda
1999
PPAR-γ agonist
Avandia
Rosiglitazone maleate
N
O
S
OHN ON
SmithKline Beecham
1999
PPAR-γ agonist
Combination Therapies
Synjardy
Empagliflozin + metformin
Boehringer Ingelheim
2015
Xigduo XR
Dapagliflozin + metformin
AstraZeneca
2014
Duetact
Pioglitazone + glimepiride
Takeda
2006
Jentadueto
Linagliptin + metformin
Eli Lilly
2012
Juvisync
Sitagliptin + simvastatin
Merck
2011
ACTOplus me
Pioglitazone + metformin
Takeda
2005
Avandamet
Rosiglitazone + metformin
GlaxoSmithKline
2002
Metaglip
Glipizide + metformin
Bristol-Myers Squibb
2002
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
37
Table 2: anti-diabetic small molecules in clinical trials.106
Anti-diabetic agents
Chemical structure
Company
Mechanism of action
Clinical trial
Omarigliptin
S
O
O
N
N
N
O
F
F H
2
N
Merck - ML-3102
DPP4 inhibitor
Phase 3
Trelagliptin
N
O
NO
N F
N
H2N
Takeda- SYR-472
DPP4 inhibitor
Phase 3
INT131107
N
O
Cl
Cl
H
NS
O
O
Cl Cl
InteKrin
Therapeutics PPAR-γ modulator Phase 2
PN2034 NA
Wellstat
Therapeutics
PPAR-γ modulator Phase 2
Lobeglitazone/CKD-501
N
O
O
HN
OS
NN
O
O
Chong Kun Dang PPAR-γ agonist with
partial PPARá affinity Phase 3
GT505
NA
Genfit
PPAR-α/δ co-agonist
Phase 2
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
38
DB959
N
O
O
CO2Na
O
DARA
BioSciences PPAR-γ /δ co-agonist Phase 1
HPP 593
NO
SO
NH
O
NovoNordisk PPAR-δ agonist Phase 1
SAR 351034
NA
Sanofi-Aventis
PPAR-δ agonist
Phase 1
Chiglitazar/CS038
N
O
COONa
NH
O
F
Chipscreen
Biosciences PPAR-α/γ dual agonist Phase 3
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
39
Table 3: summary of QSAR and 3D-QSAR studies performed for T2DM.
Target Reference Data set QSAR Model validation External validation
Chemical class Training/test R2 Q2 R2
GSK3β
Fang et al. 49
Benzofuranyl -indolyl
maleimides
30/8
CoMFA
0.984
0.602
0.905
CoMSIA
0.983.
0.665
0.761
Crisan et al. 53
Maleimides
105/13
2D PLS
0.938
0.866
0.875
Akhtar and
Bharatam 50
Anilino Aryl
maleimides
57/17
CoMFA
0.652
0.958
0.82
CoMSIA
0.614
0.948
0.87
Fu et al. 55
Diverse
587/141
Support vector
machines
0.967
NA
0.752
Prasanna et al.
51
Anilino Aryl
maleimides
56/18
CoMFA
0.942
0.844
0.779
CoMSIA
0.932
0.833
0.803
Lather et al. 54
Indirubin
36/8
2D QSAR
0.93
NA
0.60
3D-Pharmacophore
0.97
NA
0.91
Patel and
Bharatam
48
Pyrazolopyridazine
59/14
CoMFA
0.6
0.97
0.55
Taha et al. 56
Diverse
132/29
3D-Pharmacophore
0.663
0.592
0.695
Dessalew et al.
52
Pyrazolopyrimidine
49/12
CoMFA
0.98
0.53
0.47
CoMSIA
0.92
0.48
0.48
PTP 1B
Malla et al. 61
Formylchromane
26/10
3D-Pharmacophore
0.96
0.774
0.729
Malla et al. 62
Thiophene
48/22
3D-Pharmacophore
0.968
0.737
0.790
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
40
Thareja et al. 63
Sulphonamides
22/8
SOMFA
0.797
0.751
0.616
Thareja et al. 64
2,4-Thiazolidinediones
20/7
SOMFA
0.708
0.658
NA
Cheng et al. 67
Benzoic acid
33/6
HQSAR
0.940
0.592
0.786
Benzofuran and
benzothiophene
biphenyls
50/10
0.863
0.667
0.866
Ma et al. 68
Diverse
22/34
3D-Pharmacophore
0.776
NA
NA
Sachan et al. 60
formylchromones
29/7
GFA 2D QSAR
0.766
0.689
0.785
Sobhia and
Bharatam
65
1,2-naphthoquinone
25/9
CoMSIA
0.988
0.454
NA
Nair and Sobhia
66
Pyridazine
28/9
CoMFA
0.990
0.619
0.630
Taha et al. 69
Diverse
154/NA
3D Pharmacophore
0.757
0.68
NA
SGLT2
Xu et al. 85
C-aryl glycoside
32/8
CoMFA
0.967
0.548
0.974
CoMSIA
0.943
0.542
0.938
Zhi et al. 86
C-aryl glucoside
39/7
CoMFA
0.985
0.792
0.872
CoMSIA
0.895
0.633
0.839
Vyas et al. 87
C-aryl glucoside
144/36
CoMFA
0.905
0.602
0.622
CoMSIA
0.902
0.618
0.582
Nakka and
Guruprasad
88
C-aryl glucoside
43/10
MFA
0.857
0.783
0.829
Suryanarayanan
et al.
89
Xylosylindole
32/7
3D pharmacophore
0.9527
0.9045
NA
Tang et al. 90
Diverse
25/85
3D pharmacophore
0.955
NA
NA
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
41
Prasoona et al.
84
of C-aryl glucoside
36/10
SVM
0.8480
0.6622
NA
PPARγ
Guasch et al. 104
Tyrosine-based
derivatives
25/24
3D-pharmacophore
0.9223
0.6385
NA
Khanna et al. 96
5-aryl
thiazolidinedione and
oxazolidinedione
27/5 CoMFA 0.985 0.782 NA
Sundriyal and
Bharatam
99
Indylacetic acid
derivatives
28/9 CoMFA 0.981 0.757 NA
Giaginis et al. 98
Tyrosine derivatives
83/10
2D-QSAR
0.74
0.67
NA
Liao et al. 101
Tyrosine derivatives
77/18
CoMFA
0.974
0.642
NA
CoMSIA
0.979
0.686
NA
Gee et al. 103
Thiazolidinediones
210
2D QSAR/Neural
network
pessimistic AUC =0.993
sensitivity = 95.7%
specificity = 96.9 %
NA
Maltarollo et al.
97
Diverse
35/10
2DQSAR
0.90
0.80
NA
Garcia et al. 102
Phenoxy acetic acid
derivatives
86/20
2DQSAR
0.87
0.83
NA
HQSAR
0.90
0.73
NA
CoMFA
0.94
0.88
NA
Al-Najjar et al.
105
Diverse
72/16
3D Pharmacophore
0.80
0.73
0.67
DPP-IV
Patil et al. 73
β-amino amide
26/8
HQSAR
0.971,
0.971
0.91
Jain and Ghate
77
azabicyclo-derived
33/5
CoMFA
0.989
0.630
0.584
CoMSIA
0.977
0.610
0.549
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
42
Murugesan et
al. 75
pyrrolidine analogues
140/50
CoMFA
0.882
0.651
0.706
CoMSIA
0.803
0.661
0.706
Paliwal et al. 72
pyrrolidine analogues
39/8
MLR
0.87
0.84
NA
PLS
0.68
0.82
Jiang 74
β-phenylalanine
derivatives
73/18
CoMFA
0.954
0.759
0.737
CoMSIA
0.959
0:750
0.636
Saqib and
Siddiqi 76
triazolopiperazine amide
36/9
CoMFA
0.868
0.589
0.816
CoMSIA
0.868
0.586
0.863
Al-masri et al.
81
Diverse
287/71
3D-Phamracophore
0.74
0.69
0.51
Pissurlenkar et
al. 79
phenylalanine,
thiazolidine, and
fluorinated pyrrolidine
analogs.
74/28
CoMFA
0.95
0.58
0.69
CoMSIA
0.96
0.71
0.66
25/9
CoMFA
0.91
0.62
0.73
CoMSIA
0.87
0.62
0.58
Zeng et al. 78
fluoropyrrolidine amides
27/6
CoMFA
0.982
0.555
0:884
CoMSIA
0.953
0.613
0.914
Downloaded by [University of Lethbridge] at 12:59 12 November 2015
