Virtual screening strategies: Recent advances in the identification and design of anti-cancer agents
Vikash Kumar1, Shagun Krishna1 & Mohammad Imran Siddiqi*, 1, 2,
1Molecular & Structural Biology Division CSIR-Central Drug Research Institute, Lucknow, India
2Academy of Scientific and Innovative Research, New Delhi
*Corresponding author: - E-mail:- [email protected], [email protected] Abstract:
Virtual screening (VS) is a well-established technique, which is now routinely employed in computer aided drug designing process. VS can be broadly classified into two categories, i.e., ligand-based and structure-based approach. In recent years, VS has emerged as a time saving and cost effective technique, capable of screening millions of compounds in a user friendly manner. In the area of cancer drug design, VS methods have been widely used and helped in identifying novel molecules as potential anti-cancer agents. Both ligand-based VS (LBVS) structure-based VS (SBVS) methods have been highly useful in the identification of a number of potential anti-cancer agents exhibiting activities in nanomolar range. In tune with the rapid progress in the enhancement of computational power, VS has witnessed significant change in terms of speed and hit rate and in future it is expected that VS will be a preferential alternative to high throughput screening (HTS). This review, discusses recent trends and contribution of VS in the area of anti-cancer drug discovery.
Keywords: Virtual Screening, Anti-cancer Agents, Pharmacophore, Molecular Docking 1. Introduction:
The word “Virtual” is used to signify any instances, which are not directly connected to real world and we human beings cannot perceive them through our senses. In fact the word
“virtual” is becoming unambiguous as it is appearing more frequently in real life situations.
In the present era of applied science, the term virtual screening is used in the context of Computer Aided Drug Discovery (CADD). VS is a computational technique, which is used to screen novel potential active molecules (called hits) from a chemical database.
Pharmaceutical companies and many institutions now routinely employ VS as one of drug discovery methods. The origin of VS dates back to late 1980s, when ALADDIN programme was used to screen a database at ABBOTT laboratory [1].
Currently, the VS methods have evolved to a greater extent in terms of user friendliness, utility and performance. This has led to the increased use of VS methodology and many successful examples covering different disease areas have been published during the last two decades (1994-2014). Availability of supercomputing and cloud computing facilities has made possible to screen a large chemical database (having millions of compounds) within hours without much efforts. There are many published reviews [2-5], which covers the various aspects of VS in detail. However, there is a need of a review covering the recent progress of VS in the area of anti-cancer drug discovery and in this context this review concentrates on recent VS methodologies adopted in the field of anti- cancer drug discovery.
2. Need of VS in anti-cancer drug discovery:
Compared to other diseases (parasitic), cancer is a result of malfunction of cellular machinery, where transformed (cancerous) normal cell aggressively divides and spread to
other are impose healthy focuses Tradition dividing therapy r and alop therapies them. D which p involved include biologica opportun some of consider high cos methods 3. Overv As ment basic con is to retr compoun Abott la Modern (Fig.1) a and struc
Figure.1
eas of huma a great hurd dividing cel on specific nal cancer t cells includ results in ser pecia [6]. T s in aspect t uring the la plays an imp d in cancer
monoclonal al function o nity to look f the techniqu red as a good st of cancer s.
view of virtu tioned earlie ncept of the rieve and pri nds. First su aboratory. Th
VS techniqu are classified cture-based v
. Different a
an body thro dle as these lls. This is molecular t therapy such ding cancer c rious side ef Targeted ther
that they spe ast ten years
portant role are either fu
antibodies of their targe
for new mo ues that can d alternative drug discov
ual screenin er, the VS t method has ioritize the p uccessful atte
he above dis ues are fast, d mainly in virtual scree
approaches to
ough the pro e are modifi the reason argets, frequ h as chemot cells and few ffects such a
rapy has ad ecifically sto s, a large nu
in cancer unctional pr (mAb) and ets. The liga olecules wh n screen thou
e to HTS, be very can be
ng strategies terminology s not change potential acti empt of VS scovery can , user friend
two catego ening (SBVS
o VS.
cess of meta ied normal
behind deve uently over- therapy and w normal cel
s, myelosupp dvantage ove op the proli umber of mo
initiation an roteins or st
small mole and binding ich could op usands of m ecause it is c
lowered up
s:
y was concep d till date. T ive compoun was the disc be marked dly, advanced
ries, i.e., lig S).
astasis. Targ cells and ki elopment of -expressed o radiation th lls. The cyto
presion, gas er both che feration of c olecular targ
nd progress tructural pro ecules, which
site on the t ptimally fit molecules wit cost effectiv p to certain
ptualized in The concept nds from the covery of no as the begin d and autom gand-based v
geting the ca illing them f targeted th or altered in herapy targe toxic nature trointestinal emotherapy
cancer cells gets have be
ion. Most o oteins. Targe h bind and targets offer
into it. HTS thin short tim ve and fast. N extent by e
n the late 19 behind VS i e virtual libra ovel D1 ago
nning of mo mated. The V virtual scree
ancerous cel will also ki herapy, whic cancer [6-7 et the rapidl of tradition complicatio and radiatio or either ki een identified
of the targe eted therapie
modulate th s an excellen S and VS ar me [8]. VS No doubt, th
mploying V
980’s and th is simple, th ary of divers
nist [1] at th odern VS er VS technique ening (LBVS lls
ill ch 7].
ly al on on ill d, ets es he nt re is he VS
he at se he ra.
es S)
4. VS strategies adopted in anticancer drug discovery
VS techniques employed in the drug discovery are mostly generalized and have not been specifically designed for use in discovery of anti-cancer agents. The successful applications of VS have been increasing in recent years towards the contribution for development of anticancer agents. In this section, we highlight the general methodologies and current progress in VS from the selected recent works and provide an overview of the emerged strategies in anti-cancer drug discovery. The key features of presented case studies are tabulated in table 1.
4.1. Ligand-based virtual screening (LBVS)
VS can be initiated from at least one known ligand of the target, and the method is known as LBVS. Pharmacophore-based VS (PBVS) is considered most popular method under the LBVS approach. Ehrlich first presented the theory of pharmacophore in the year 1909 [9].
He described the pharmacophore as an abstract description of a drug or biologically active molecule that entails (phoros) the necessary features accountable for the drug’s (pharmacon) biological activity. [9] During the past few decades, the perception of pharmacophore still remains perpetual but at the same time its application in drug discovery has been magnified extensively. A pharmacophore model may be generated in a ligand based approach for which at least one active query molecule and a search database is an essential prerequisite [8].
However, a set of active molecules can also be used [10]. To execute ligand-based pharmacophore screening using a set of known ligands (which are collectively called as training set), usually common chemical features that illustrate important interactions between a ligand and target are extracted. The PBVS comprises of two major steps- creation of conformational space for the ligands in training set so that the flexibility associated with conformation of ligands can be illustrated and alignment of multiple ligands to figure out the fundamental chemical features so that the pharmacophore can be generated [11]. For the development of pharmacophore model the first crucial step is the selection of accurate chemical feature. Initially the active analogue approach was used in which a pharmacophore could have any fragment or atom type [12]. However, recently available techniques utilize a generalized manner for generating pharmacophore models. Presently, various automated software packages/modules are used for the generation of pharmacophore models such as CATALYST/HipHop [13-14], Hypogen [13, 15], DISCO [16-17], GASP [16, 18] and GALAHAD [16, 19], MOE [20], PHASE [21-22], and LigandScout [23]. The software available for pharmacophore generation are based on various algorithms and the variety resides in the alignment rules for the training set molecules and also the approach of handling conformational flexibility. Various studies have been published showing comparison of these software packages [24-26] which may be referred to analyse the differences, advantages and disadvantages of these programs. PBVS has been proved to be successful in identification of a number of potential anticancer agents.
In 2007, Purushottamachar et al. [27] published the first PBVS for the identification of androgen receptor down-regulating agents (ARDAs). This was achieved with the help of a three-dimensional pharmacophore model generated with the help of HipHop software based on a training set of five natural products. The generated pharmacophore was used by them as a query to screen two databases – Maybridge database [28] containing 59,652 compounds and National Cancer Institute (NCI) database [29] containing 238,819 compounds. The hits identified by screening of these two databases were ranked according to their fit score and only those that have fit score greater than or equal to 3.05 were selected further. They had shortlisted 41compounds for further evaluation. However, based on the availability, only 17 compounds were experimentally validated. Out of these, six compounds exhibited significant
down regulation of AR protein expression at 50 and 150 µM concentration. Among these six compounds (EC50values 17.5–212 µM), five compounds were able to show considerable inhibition of the viability of human prostate cancer LNCaP cells (IC50 value 4.5-39.8 µM).
The compounds identified in this study may contribute significantly to further design of potent ARDAs since the natural product ARDAs inhibit the growth of human cancer cells at comparatively higher concentrations. Thus in this study only a pharmacophore based screening has led to the discovery of potent inhibitors.
Massarotti et al. [30] proposed a 7-point pharmacophore model to identify novel inhibitors that interact with colchi-site on the tubulin dimer with the help of MOE. The authors performed a screening of three commercially available databases out of which 1127 hits were identified. The molecules possessing similar structural similarity and those having Tanimoto coefficient greater than 0.8 as compared to reference molecule were removed. Interestingly in this study the use of seven point pharmacophore has led to the discovery of potent anti- tubulin agents, compound 36 and compound 34 with IC50 values of 194 ±5 and 200±35 nM respectively.
Similarity-based VS methods have evolved for last three decades and there are various approaches that have been attempted. The fundamental idea behind a similarity-based approach was first described in the year of 1990 by Johnson and Maggiora [31] who proposed the principle of similarity, according to which the structurally similar molecules are likely to have related properties. Consequently if this principle is true then a database molecule that is structurally analogous to a molecule of known activity against a target (reference molecule) should exhibit activity against the same target, even this molecule is expected to be more active as compared to any other molecule of database that is less similar to reference molecule. For this reason, similarity-based VS approach utilizes the comparison of reference structure with each molecule of database and then ranking of database molecules is done in order to compute similarity between them and performing actual screening on only the high ranked database molecule. Similarity-based VS can be broadly classified as Superposition- and histogram-based similarity methods and descriptor based similarity methods [32].
In superimposition methods [33-34] one molecule is mapped on the other molecule. In 2D superimposition the molecules are considered as graphs and the correlation is calculated between the atoms of corresponding molecules [34-35] whereas in the 3D superimposition approach the best superimposition in 3D form of molecules is taken into account. In case of histogram similarity methods, the 2D and 3D structures are converted into spectra that are known as histograms and then the correspondence is calculated between them [36-37]. The third category of similarity-based method is descriptor based similarity methods where the descriptors are calculated according to the structure of molecules and the molecules are considered as a single point in a multi-dimensional descriptor space. In one subdivision of this method various properties are calculated such as molecular weight, log D, number of hydrogen bond donors and number of hydrogen bond acceptors, dipole moment and several other descriptors [38-41]. Presently, one of the most widely accepted method is the 2D fingerprinting method [42]. Carhart et al. and Willett et al. in separate studies described the concept of ranking of database in which both of them have given emphasis on the usage of 2D fingerprints [43-44]. 2D fingerprints are essentially binary strings that articulate the presence or absence of a sub-structural fragment and these substructures are used as descriptors. The similarity is calculated by the common descriptors among each molecule of the database that is normalized by the number of descriptors calculated of the database molecules. There are various types of structural representations available for calculation of
molecular similarity but the 2D fingerprinting is widely accepted as the best choice for performing a similarity search [44].
There is an interesting paper published by Füllbeck et al. [45] in 2005, which describes the discovery of novel curcumin- and emodin-related compounds that have induced apoptosis in tumor cells. In this work, authors have developed a superimposition algorithm through which the comparison between curcumin and emodin as lead structures and 106 compounds of their in-house database regarding their structural properties was performed. In this study, they have utilized a 3D as well as a 2D similarity search approach to find out two groups of inhibitors.
The study also compares the significant success rate of VS as compared to high throughput screening.
A study by Wang et al. [46] described an example of connectivity based search using Scitegic Pipeline Pilot [13] and molecular shape similarity search using Schrodinger software [22]. The authors have initially performed the validation of connectivity based search by establishing a relatively small testing compound library of 2032 molecules. They conducted a connectivity similarity search for lead compound LY-1-100 against the 2032 testing compound library and it was subjected to five similarity filters in parallel using the ECFP2, ECFP4, ECFC6, FCFP4, and FCFP6 property sets and Tanimoto distances using LY-1-100 as the lead structure to identify 14 new molecules from the University of Cincinnati Drug Discovery Center Compound Library (total 342910 compounds) that are active against melanoma cells.
Recently, various machine learning methods have been developed as a LBVS tool to construct an accurate and robust screening of large chemical libraries. There are various methods available that can be implemented to perform machine learning based VS. Among these support vector machines, neural networks, decision trees and numerous regression and classification methods viz. multiple linear regression, nearest neighbours and naive bayesian classification are widely utilized. Machine learning methods apply the use of a training set of molecules that has already been reported as active or inactive against a target of interest.
These training set molecules are then evaluated to develop a decision rule with the help of which the new molecules (test set) are classified as actives or inactive for a given target [47].
Support vector machines (SVM) were first developed in the year of 1990 by Vapnik as a general data modelling approach for the pattern recognition [48]. SVM are used to create classification model. The notion behind SVM is to map data into a high dimensional space in which a constrained quadratic programming problem will be solved and a separating hyper plane with the maximal margin will be found [49]. An SVM algorithm determines a set of numbers that characterize each object and defines the objects in two classes and calculates a classification model to evaluate other objects among the two classified classes. In VS, the two classes of objects are active and inactive molecules. Various molecular descriptors are calculated with the help of which the characterization of molecules is accomplished and a classification model is derived. Once the model is trained it is used to screen the databases to identify active molecules against the given target of interest [50]. Bayesian methods have been developed on the basis of Bayes theorem and statistical approaches. The Bayesian methods that are used in VS, calculate the probability of a compound to be active against a target. Bayesian methods are utilized to rank compound databases since they calculate numerical estimates for the likelihoods of activity or inactivity of a compound according to their probability distribution [51].
In a study published by Liu et al. [52], a SVM based VS approach was explored to identify novel zinc binding groups and nonhydroxamate Histone deacetylase (HDAC) inhibitors. In
this study, the authors have utilized a training dataset to create a HDAC inhibitor VS tool that has the ability to screen large database at low false hits rate and a good yield. The study shows that the SVM based method demonstrates good performance and compared with other VS methods, SVM can be used to attain comparable yields at a very low false hit rate similar to HTS.
4.2. Structure-Based Virtual Screening (SBVS)
Availability of three dimensional structures of the targets provides another route to search hits, primarily known as structure-based virtual screeing (SBVS) or receptor based virtual screening (RBVS). Prior knowledge of ligand binding site is essential to execute SBVS.
Compared to ligand-based approaches, SBVS is considered more productive and informative.
SBVS techniques can also be integrated with LBVS techniques to make the search strategy more robust.
Among the most widely used SBVS strategies, docking-based virtual screening (DBVS) stands at the top position. Docking is a term generally used to describe the process of finding the conformation of ligand in the binding site of a protein. In the docking process, programme also calculates the score/energy associated with the protein-ligand interaction.
DBVS utilizes the ability of docking programme, to handle large number of compounds and rank them on the basis of binding energy/score in an automated manner. Available docking programmes can easily screen millions of compounds in a short time. The time required to screen the compounds mainly depends on three factors i.e., size of database, scoring function and computational power. There are many free and commercially available docking softwares such as AutoDock [53], DOCK6 [54], GLIDE [22], FlexX [55], SurfleX [56], AutoDock Vina [57] and GOLD [58]
A simple DBVS method using FlexX was applied to find the antagonists of androgen receptor (AR) [59]. In this study, 20000 compounds were docked into ligand binding domain (LBD) of AR. Binding affinity of compounds were evaluated by using multiple scoring function and also consensus scoring. After screening, 54 compounds were procured and tested for their possible antagonistic/agonistic effects. Following the experimental evaluation, one of these compounds (DIMN) showed significant antagonistic effect.
In another work, Li et al. [60] identified 13 novel EGFR kinase inhibitors using SBVS approach with the help of Schrodinger suite. Crystal structure of EGFR kinase domain complexed with lapantib was selected for the docking of molecules from SPECS database.
Glide programme [22] was used to carry out two stages of docking i.e., high throughput VS (HTVS) and standard precision (SP) respectively. After ranking with Glide score, 500 hits were retained and subjected to visual inspection. Finally 43 compounds were sorted and purchased from SPECS database [61] for experimental evaluation. At 100 µM concentration 13 compounds showed greater than 50% inhibition. Among the identified hits, further assay at 10Um concentration resulted in identification of compound13 as most potent inhibitor (Ic50=3.5 µM).
In a work of Foloppe et al. [62], 10 novel Chk1 inhibitors were identified using SBVS approach. With the help of rDock [63] programme, 700000 compounds were docked into the ATP binding site of Crystal structure of Chk1. On the basis of intermolecular score, 15000 compounds were retained and divided into set A (best scoring 2000 compounds) and set B (remaining 13000 compounds). After visual inspection of interaction pattern of hits, 480 compounds were selected from the set A. Based on 2D-MACCS fingerprint[64] available in MOE and Tanimoto similarity coefficient, diverse subset of 1000 compounds were
selected from set B. The final list of hits contained 1480 compounds. 1179 compounds were obtained and subjected to assay. After assay, 0.8% of the total assayed compounds showed
> 50% inhibition at 50 µM concentration.
In one other example, a Stat3 inhibitor showing antitumor activity was identified through SBVS approach [65]. Glide docking programme was used to dock NCI library into the pTyr peptide binding site of SH2 domain of monomeric Stat3 crystal structure (PDB: 1BG1). Best scoring hits were subjected to experimental assay and led to identification of S3I-201. The identified compound inhibited the Stat3-Stat3 complex formation.
A simple SBVS approach also helped in identification of nanomolar inhibitor of NRH:
quinone oxidoreductase 2(NQO2) from the NCI database [66]. GOLD docking software was utilized to dock all the molecules of NCI database. After visual inspection of top scoring hits, 250 hits were finally selected for experimental evaluation. NSC13000 was identified as novel NQO2 inhibitor having potential anticancer activity.
Prioritization of hits on the basis of score and interaction pattern is crucial step in the SBVS.
Visual inspection of large set of protein- ligand interaction is cumbersome process and may not consider all the important interactions. Structural interaction fingerprint (SIFT) is a simple method to represent and analyze the 3D protein-ligand interaction [67]. The method converts the protein-ligand interaction pattern into the 1D binary string and can be utilized as postdocking molecular organizer and filter tool to reorganize the docking poses [68].
Other important issue with DBVS is the treatment of role of solvent in the protein–ligand interaction. It is quite evident from experimental techniques such as X-ray crystallography and nuclear magnetic resonance (NMR) that solvent play an important role in ligand binding process. Most of the docking programmes, did not consider the water molecules during docking process, until it is explicitly included. However there are scoring functions, which include the solvent effect implicitly in the course of protein-ligand binding score/energy calculation.
A method called linear interaction energy (LIE) approximation combines the molecular mechanics (MM) calculations with experimentally available data to build a model scoring functions for protein-ligand binding free energies and incorporates highly attractive features [69]. The LIE methodology was used to design four thapsigargin (TG) derivatives based on the experimental binding data and docking of 20 TG analogues (training set) [69]. Docking of training set molecules was performed using MOE and LIE was calculated with the help of LIAISON package. It was observed that the calculated LIE showed good correlation with Ic50 value of designed TG analogues.
In addition, Adaptive Poisson-Boltzmann solver (APBS) is a tool which calculates the electrostatic binding free energy of a protein-ligand system, while considering the effect of solvation energy [70]. Similarly, solvated interaction energy (SIE) is an end point scoring method for protein-ligand system [71], which consists of force field terms supplemented by solvation terms. It has been shown that the inclusion of explicit conserved water molecules during docking, not only increases the accuracy of pose prediction but also the docking affinity [72].
In a study by Barakat et al. [73], interaction between ERCC1 and XPA was targeted. Two stage docking protocol was applied by the authors to find out the small molecule inhibitors of ERCC1/XPA interaction. In the first stage, Autodock was used to dock the CN library into the NMR structure of ERCC1. This resulted in 2000 hits having binding energy below - 5kCal/mole. In the second stage, relaxed complex scheme (RSC) methodology [74] was
applied to filter the initial hits and subsequently top 170 compounds were subjected to short molecular dynamics (MD) simulation. APBS tool was utilized to prioritize the hits and finally 14 compounds were subjected to biological evaluation. Compound 10 and 12 showed the Kd values of 27.4 µM and 66.8 µM respectively.
In DBVS method, statistical validation can be incorporated to increase the chance of getting active hits. A well-known and accepted protocol is Receiver Operating Characteristic Curve (ROC) analysis [75-76]. In ROC, a compound database is seeded with known actives and inactive compounds for a particular target. After retrieval of hits, statistical calculations are performed to calculate the sensitivity and specificity of the employed DBVS method.
One of the most challenging issues in DBVS is the treatment of receptor flexibility. In recent years, attempts have been made to address the issue of receptor’s flexibility during DBVS process. Ensemble based VS is becoming a popular method to consider the receptor flexibility in the DBVS. Soft docking is another way to handle the flexibility issue [77]. In soft docking method, few steric clashes between protein and ligand are permitted by lowering the steepness of the repulsion term in the Lennard-Jones potential function [77].
However, the soft docking method allows limited conformational changes. In the VS protocol, receptor flexibility is one of the major issues that need to be addressed. Although many docking software provide flexibility option upto certain extent, consideration of full receptor flexibility during virtual screening is still in the naïve phase.
Li et al. [78] showed that consideration of multiple crystal structure during VS improves the enrichment factor as well as helps in identification of more diverse compounds. For ensemble based VS, 46 crystal structure of Chk1 were processed using Maestro wizard of Schrodinger suite. A test set of 2,042 compounds containing 1,996 random compounds (from ZINC database) [79] and 46 crystal ligands was prepared with the help of Schrodinger suite.
Optimal ensembles of structures were selected on the basis of enrichment factor. VS using ensemble of three Chk1 crystal structures showed enrichment factor of 30.9 and subsequently the ensemble was used to screen more than 60000 compounds from ZINC database. Out of the 6 compounds purchased for Chk1 assay, 2 compounds Chk#2 and Chk#4 showed IC50 of 9.6 µM and 49.5 µM respectively.
4.3. Combined Ligand and Structure Based Virtual Screening
In order to make the VS more robust, LBVS and SBVS approaches can be integrated.
Availability of both ligand and corresponding target structure offers a way to carry out the combined VS. Few selected studies, which have incorporated the combined LBVS and SBVS strategies, are discussed below.
Ambaye et al. [80] applied VS strategies involving a shape-based similarity search, molecular docking, and 2D similarity searches to discover novel Grb7-based antitumor agents. The lead peptide antagonist 5 was employed as reference shape query. The validated shape query was used to screen the NCI database, which identified 16521 hit molecules. These were further subjected to docking using Glide which identified 17 hits. The methodology also uses the hit optimization of one compound 1 with the help of CACTUS software [29] using MACCS key structural descriptors and Tanimoto similarity coefficient. The study proposes a successful utilization of ligand shape similarity search based VS followed by molecular docking and hit optimization with 2D similarity search for identification of a small-molecule antagonists of Grb7 starting with a polypeptide search query.
In a study published by Xie et al. [81], SVM based classification model was used to differentiate between c-Met inhibitors and non-inhibitors. The authors have performed a five- fold cross validation of this model and to check whether this model is able to classify
molecules outside the training set, they have also performed an external validation. Along with this they have also combined the docking based approach. With this study the authors have identified 8 molecules and out of these, five of them are novel scaffold. The study presents a successful example of combined SVM based and DBVS that significantly increases hit rate.
In another work, Dokla et al. [82] reported a novel protocol based on a SVM model, Bayesian model and a structure based pharmacophore filters established on previously known urea based kinase inhibitors for the discovery of novel urea-based anti-neoplastic kinase inhibitors while emphasizing on both diversification of compounds and selectivity pattern.
The selectivity pattern was created using the Ftrees algorithms for similarity searching against NCI database. This study is a successful example of the application of a novel computational procedure that allows screening of urea derivatives that can act as kinase inhibitors and also the development of another computational procedure that allows verification of cancericidal activity of the hits in order to prioritize selection.
In another study, Ren et al. [83], applied multistage VS approach that is based on SVM, pharmacophore based and also DBVS approach to identify novel Pim1 inhibitors as potential anticancer agents. With the help of this approach they have screened very large chemical libraries in comparatively lesser time consisting of a hierarchical multistage VS approach based on SVM, pharmacophore based and docking based methods. With the help of this combined time saving approach, they have successfully identified 15 potential anti-cancer compounds with nanomolar or low micromolar activity among 2 million compounds. The study highlights that the multistage VS strategy could also play an important role in drug discovery in search of potential anticancer agents.
A successful protocol for similarity-based VS along with docking and hierarchical hit optimization was published by Kong et al. [84]. They have used a hybrid algorithm named SHAFTS [85] for 3D similarity search that not only identified selective active molecules against mutant B-RafV600E but also exhibited reasonable scaffold hopping capability towards few representative kinases. They have also utilized a substructure based analogue searching for optimization of hit molecules by close examining the binding mode of compound1 and they have ended up with a more potent nanomolar inhibitor compound 22q. This study offers a noteworthy example of similarity-based VS combined with hierarchical hit optimization that could be helpful not only to improve biological potency of inhibitor but also the selectivity towards oncogene mutation.
In a recent work published by Krishna et al. [86], authors have applied a multistep PBVS protocol to find out novel inhibitors of human DNA ligaseI as potential anticancer agents. In this study, with the help of previously reported inhibitors the authors have generated a simple three point pharmacophore using GASP module of Sybyl7.1 [87]. The pharmacophore was further used to screen Maybridge database and the hits identified were ranked according to the UNITY score. In the study, they have also performed the docking of top 3000 hits to the DNA binding domain of hLigI and calculated the binding energy of top thirty compounds with APBS to prioritize hits. This simple but statistically validated pharmacophore has led to the discovery of two potent hLigI inhibitors HTS01682 and NRB00556 with IC50 value 24.93±3.71 and 37.56 ±6.97 µM respectively.
In another work, Lu et al. [88] published the discovery of a nanomolar inhibitor of the human murine double minute 2 (MDM2)-p53 interaction as potential anticancer agents. In their paper, the authors described how a pharmacophore based screening protocol followed by
docking can be specifically utilized for disruption of MDM2-p53 interaction that in turns reactivates p53 function and now accepted as a new approach for anticancer drug design. The authors have proposed a simple pharmacophore model with the help of previously available non-peptide small molecule inhibitors of p53-MDM2 interaction. A database search was performed on a subset of (~150,000 compounds) of the NCI database. The authors have applied a web-based, flexible pharmacophore searching tool developed in their laboratory [89]. The screening identified 2599 hits which were further subjected to docking using GOLD with the ChemScore fitness function. The 67 hits short listed after visual inspection were subjected to a quantitative and sensitive fluorescence-polarization based (FP-based) competitive binding assay to check their ability to exhibit a fluorescently tagged p53-based peptide from the MDM2 protein. They have found 10 compounds showing a Ki value less than 10 µM in this assay, However NSC 66811 has the highest binding affinity with a Ki of 120 nM that can serve as a novel class of inhibitor of the MDM2-P53 interaction.
5. Libraries and databases specific to anticancer agents:
Most of the available chemical databases such as ZINC, NCI, Maybridge and Asinex [90]
are generalized and can be used for VS, irrespective of disease area. However, there are few databases/libraries, which are exclusively designed to aid in anticancer drug discovery.
Anticancer Agent Mechanism Database is a small database that consists of 122 compounds with known mechanism of action [91]. Conceptualization of targeted and focused libraries for the purpose of VS has shown to improve the hit rate. For example, kinase targeted library (KTL) helped in identification of new inhibitor scaffold for PDK [92].
S.no Compound name Target name
Method used Activity Ref
1. NCI-0002815 The androgen receptor
Pharmacophore- based VS
IC50= 4.5µM [27]
2. Compound 36 tubulin Pharmacophore-
based VS IC50=194±5nM [30]
3. BTB14431 COP9 Similarity-based VS
IC50= 6.4 µM (CK2) and 68.9µM(PKD)
[45]
4. LY-1-100 Melanoma
cells
Similarity-based VS
IC50=55.1±4.8nM (B16-F1 cell line)
[46]
5. Compound#1(DIMN) Androgen receptor
Docking-based VS
IC50=3µM [59]
6. Compound13 EGFR kinase Docking-based VS
IC50= 3.5 µM [60]
7. S3I-201 STAT3 Docking-based VS
IC50= 86±33 µM [65]
8. NSC13000 NQO2 Docking-based
VS IC50= 420 nM [66]
9. Compound10 and 12 ERCC1-XPA interaction
Docking-based VS
Kd= 27.4 uM, 66.8 µM [73]
10. Chk#2 and Chk#4 Chk1 Docking-based VS
Ic50 = 9.6 uM, 49.5 µM [78]
11. Compound 1 Grb7 Combined ligand and structure-
based VS
IC50= 39.9µm [80]
12. Mol1 c-Met Combined ligand and structure-
based VS
%
inhibition=84%@10µM [81]
13. Compound 12a Antineoplastic Kinase
Combined ligand and structure-
based VS
GI50= 0.9 µM [82]
14. Compound N5 Pim-1 Combined ligand and structure-
based VS
IC50= 263nM [83]
15. Compound 22q Human B-Raf protein kinase
Combined ligand and structure-
based VS
IC50= 1.15µM [84]
16. HTS01682 Human DNA
Ligase I
Combined ligand and structure-
based VS
IC50=24.93±3.71 µM [86]
17. NSC 66811 p53/MDM2 Combined ligand and structure-
based VS
Ki= 120nM [88]
Table.1: List of anti-cancer hits obtained using VS techniques.
6. Conclusion
In cancer research, a large number of successful virtual screening strategies to identify novel hits have been reported by various groups. It is true that the identified hits cannot directly enter into clinical trials, but may serve as starting point for further optimization. Although this review highlights the successful case studies, there are still some major issues left which must be addressed to increase the practical use of VS in drug discovery process. The ligand flexibility should be considered carefully in LBVS approaches. Not only this, molecular alignment is the second major issue that should be tackled carefully. Although the pharmacophore based VS methods have flourished very much but still there is scope for further improvement so that the more optimized and efficient pharmacophore methods can be developed. Most of the identified hits show activity in micromolar range; even their interactions and calculated scores suggest them as strong binders. Unfortunately, there are only few reported studies, which have shown the reasonably good correlation between the calculated affinities and experimental affinities of VS hits. In this context, emphasis should be given on the appropriate use of scoring functions. In one of the case studies, use of LIE provided the good correlation between calculated scores and experimental activities of inhibitors. SIFT method may be integrated with VS tools to prioritize the hit selection process. In cancer research, a number of the VS studies have been carried out on Kinases.
Because of selectivity issue, human kinases are difficult to target and almost all of the anticancer drugs which target kinases compete with the ATP binding. The ATP binding site is highly conserved in a family of related kinase and, therefore the question is how VS techniques can handle the selectivity issue more effectively? One of the possible solutions is to develop target-biased scoring functions. Another approach is to develop a SVM model from the known active and selective inhibitors and then applying the model on VS hits, which will classify them in selective or non-selective. Apart from these possible solutions, targeting the allosteric binding pocket will also be a very interesting option to design inhibitors. There are also other issues related to VS techniques that need to be addressed. Receptor flexibility may be incorporated in VS protocol, either through using multiple conformation of receptor
or conducting MD simulation study. In one of the case study discussed earlier it has been shown that incorporation of receptor flexibility enhances the enrichment factor.
Consideration of solvent effect will also help in calculation of accurate binding affinities of VS hits and thus making the VS techniques more productive. Integration of VS protocol with other techniques such as SVM, free energy calculation, MD simulation will definitely help to increase its success rate.
The VS route of cancer drug discovery provides an excellent opportunity to save time and money and hence to bring down the drug discovery cost. Recent years have witnessed excellent growth in computational power and its use in drug discovery.
Availability of large number of protein targets in cancer, offers an excellent opportunity to carry out VS. This review throws light on strategies used in in-silico identification of novel anti-cancer agents using various VS techniques. Overall VS techniques have contributed significantly in anti-cancer drug discovery and there are few challenges in VS techniques, which should be considered in future to make it more robust and helpful for the design and identification of novel anti-cancer agents.
Acknowledgements
This manuscript is a CSIR-CDRI communication number 8765. Authors thank CSIR
network project-GENESIS (BSC0121) for the Computational facility support in writing this manuscript. VK and SK acknowledge DBT for fellowships.
References
1. Van Drie JH, Weininger D, Martin YC. ALADDIN: an integrated tool for computer- assisted molecular design and pharmacophore recognition from geometric, steric, and substructure searching of three-dimensional molecular structures. J Comput Aided Mol Des.3 (1989)225-51.
2. Reddy AS, Pati SP, Kumar PP, Pradeep HN, Sastry GN. Virtual screening in drug discovery -- a computational perspective. Curr Protein Pept Sci. 8(2007)329-51.
3. Heikamp K, Bajorath J. The future of virtual compound screening. Chem Biol Drug Des. 81(2013):33-40.
4. Schneider G. Virtual screening: an endless staircase? Nat Rev Drug Discov.
9(2010)273-6.
5. Cheng T, Li Q, Zhou Z, Wang Y, Bryant SH. Structure-based virtual screening for drug discovery: a problem-centric review. AAPS J. 14(2012)133-41.
6. Sledge GW Jr. What is targeted therapy? J Clin Oncol. 23(2005):1614-5.
7. Kumar M, Nagpal R, Hemalatha R, Verma V, Kumar A, Singh S, Marotta F, Jain S, Yadav H. Targeted cancer therapies: the future of cancer treatment. Acta Biomed.
83(2012)220-33.
8. Bajorath J. Integration of virtual and high-throughput screening. Nat Rev Drug Discov. 1(2002):882-94.
9. Ehrlich, P. Ueber den jetzigen Stand der Chemo therapie.Ber.Dtsch. Chem.Ges.42 (1909)17–47.
10. Swann, S. L.; Brown, S.P.; Muchmore, S. W.; Patel, H.; Merta, P.; Locklear, J.; Philip J. Hajduk, P. J. A Unified, Probabilistic Framework for Structure- and Ligand-Based Virtual Screening. J. Med. Chem., 54(2010)1223- 1232.
11. Güner, O. F. (Ed.). Pharmacophore perception, development, and use in drug design (Vol. 2). Internat’l University Line. 2000.
12. Marshall, G.R.et al.(1979) The conformational parameter in drug design: the active analog approach. InComputer-Assisted Drug Design, (vol. 112) (Olson, E.C. and Christoffersen, R.E., eds) pp. 205–225, American Chemical Society.
13. Accelrys Software, Inc., San Diego, CA.
14. Barnum, D.; Greene, J.; Smellie, A.; Sprague, P.I dentification of common functional configurations among molecules J. Chem. Inf. Comput. Sci. 36(1996)563– 57.
15. Li, H.et al. (2000) HypoGen: an automated system for generating 3D predictive pharmacophore models. InPharmacophore Perception, Development, and Use in DrugDesign (Guner, O.F., ed.), pp. 171–189, International University Line.
16. Tripos International, 1699 South Hanley Rd., St. Louis, Missouri, 63144, USA
17. Martin, Yvonne C., Mark G. Bures, Elizabeth A. Danaher, Jerry DeLazzer, Isabella Lico, and Patricia A. Pavlik. "A fast new approach to pharmacophore mapping and its application to dopaminergic and benzodiazepine agonists."Journal of Computer-Aided Molecular Design 7, no. 1 (1993) 83-102.
18. Jones, G. and Willet, P. (2000) GASP: genetic algorithm superimposition program.
InPharmacophore Perception, Development, and Use in Drug Design (Gu ¨ner, O.F., ed.), pp.85–106, International University Line.
19. Richmond, Nicola J., Charlene A. Abrams, Philippa RN Wolohan, Edmond Abrahamian, Peter Willett, and Robert D. Clark. "GALAHAD: 1. Pharmacophore identification by hypermolecular alignment of ligands in 3D." Journal of computer- aided molecular design. 9 (2006)567-587.
20. Molecular Operating Environment (MOE), 2013.08; Chemical Computing Group Inc., 1010 Sherbooke St.West, Suite #910, Montreal, QC, Canada, H3A 2R7, 2013.
21. Dixon, S.L.et al. PHASE: a new engine for pharmacophore perception, 3DQSAR model development, and 3D database screening. 1. Methodology and preliminary results.J. Comput.Aid. Mol. Des.20 (2006) 647–671.
22. Schrodinger, Inc., New York, NY.
23. Wolber, G. and Kosara, R. Pharmacophores from macromolecular complexes with LigandScout. InPharmacophores and Pharmacophore Searches, (vol. 32) (Langer,T.
and Hoffmann, R.D., eds) pp. (2006) 131–150, Wiley-VCH.
24. Patel, Yogendra, Valerie J. Gillet, GianpaoloBravi, and Andrew R. Leach. "A comparison of the pharmacophore identification programs: Catalyst, DISCO and GASP." Journal of computer-aided molecular design 16, no. 8-9 (2002)653-681.
25. Poptodorov, K.et al. Pharmacophore model generation software tools. In Pharmacophores and Pharmacophore Searches (Langer, T. and Hoffmann, R.D., eds), pp. Wiley–VCH (2006) 17–47.
26. Dror, Oranit, Alexandra Shulman-Peleg, Ruth Nussinov, and Haim J. Wolfson.
"Predicting molecular interactions in silico: I. an updated guide to pharmacophore identification and its applications to drug design." Frontiers in Medicinal Chemistry.
3(2006)551-584.
27. Purushottamachar, Puranik, Aakanksha Khandelwal, Pankaj Chopra, Neha Maheshwari, Lalji K. Gediya, Tadas S. Vasaitis, Robert D. Bruno, Omoshile O.
Clement, and Vincent CO Njar. "First pharmacophore-based identification of androgen receptor down-regulating agents: discovery of potent anti-prostate cancer agents." Bioorganic & medicinal chemistry 15 (2007)3413-3421.
28. www.maybridge.com/, accessed 20th february 2014.
29. www.cactus.nci.nih.gov/download/nci/ ,accessed 20th february 2014.
30. Massarotti, Alberto, Sewan Theeramunkong, Ornella Mesenzani, Antonio Caldarelli, Armando A. Genazzani, and Gian Cesare Tron. "Identification of Novel Antitubulin Agents by Using a Virtual Screening Approach Based on a 7‐Point Pharmacophore Model of the Tubulin Colchi‐Site." Chemical biology & drug design 78(2011)913-922.
31. Johnson, M.A. and Maggiora, G.M., eds. Concepts and Applications of Molecular Similarity, John Wiley(1990).
32. Sheridan, Robert P., and Simon K. Kearsley. "Why do we need so many chemical similarity search methods?." Drug discovery today. 7(2002): 903-911.
33. Dixon, Steven L., and Kenneth M. Merz. "One-dimensional molecular representations and similarity calculations: methodology and validation. Journal of medicinal chemistry 44 (2001)3795-3809.
34. Hagadone, Thomas R. "Molecular substructure similarity searching: efficient retrieval in two-dimensional structure databases." Journal of chemical information and computer sciences. 32(1992)515-521.
35. Rarey, Matthias, and J. Scott Dixon. "Feature trees: a new molecular similarity measure based on tree matching." Journal of computer-aided molecular design 12 (1998)471- 490.
36. Schuur, Jan H., Paul Selzer, and Johann Gasteiger. "The coding of the three- dimensional structure of molecules by molecular transforms and its application to structure-spectra correlations and studies of biological activity." Journal of chemical information and computer sciences. 36(1996)334-344.
37. Ginn, Claire MR, Peter Willett, and John Bradshaw. "Combination of molecular similarity measures using data fusion." In Virtual Screening: An Alternative or Complement to High Throughput Screening? (2002)1-16. Springer Netherlands.
38. Menard, Paul R., Jonathan S. Mason, Isabelle Morize, and Susanne Bauerschmidt.
"Chemistry space metrics in diversity analysis, library design, and compound selection." Journal of chemical information and computer sciences. 38(1998)1204- 1213.
39. Pearlman, Robert S., and Karl M. Smith. "Metric validation and the receptor-relevant subspace concept." Journal of Chemical Information and Computer Sciences. 39(1999) 28-35.
40. Labute, Paul. "A widely applicable set of descriptors." Journal of Molecular Graphics and Modelling 18(2000)464-477.
41. Livingstone, David J. "The characterization of chemical structures using molecular properties. A survey." Journal of chemical information and computer sciences.
40(2000)195-209.
42. Walters, W. Patrick, Matthew T. Stahl, and Mark A. Murcko. "Virtual screening--an overview." Drug Discovery Today 3(1998)160-178.
43. Carhart, Raymond E., Dennis H. Smith, and R. Venkataraghavan. "Atom pairs as molecular features in structure-activity studies: definition and applications." Journal of Chemical Information and Computer Sciences 25(1985)64-73.
44. Willett, Peter. "Similarity-based virtual screening using 2D fingerprints." Drug discovery today 11 (2006)1046-1053.
45. Füllbeck, Melanie, Xiaohua Huang, Renate Dumdey, Cornelius Frommel, Wolfgang Dubiel, and Robert Preissner. "Novel curcumin-and emodin-related compounds identified by in silico 2D/3D conformer screening induce apoptosis in tumor cells." BMC cancer 5 (2005) 97.
46. Wang, Zhao, Yan Lu, William Seibel, Duane D. Miller, and Wei Li. "Identifying novel molecular structures for advanced melanoma by ligand-based virtual screening." Journal of chemical information and modelling. 49 (2009) 1420-1427.
47. Melville, James L., Edmund K. Burke, and Jonathan D. Hirst. "Machine learning in virtual screening." Combinatorial chemistry & high throughput screening .12(2009) 332-343.
48. Vapnik ,VN. An overview of statistical learning theory. IEEE Trans Neural Netw.1999;
10(5):988-99.
49. Yang ZR. Biological applications of support vector machines. Brief Bioinform. 2004 Dec;5(4):328-38
50. Han, L. Y., X. H. Ma, H. H. Lin, J. Jia, F. Zhu, Y. Xue, Z. R. Li, Z. W. Cao, Z. L. Ji, and Y. Z. Chen. "A support vector machines approach for virtual screening of active compounds of single and multiple mechanisms from large libraries at an improved hit- rate and enrichment factor." Journal of Molecular Graphics and Modelling. 26(2008) 1276-1286.
51. Watson, Paul. "Naive Bayes classification using 2D pharmacophore feature triplet vectors." Journal of chemical information and modelling. 48 (2008) 166-178.
52. Liu, X. H., H. Y. Song, J. X. Zhang, B. C. Han, X. N. Wei, X. H. Ma, W. K. Cui, and Y. Z. Chen. "Identifying novel type ZBGs and nonhydroxamate HDAC inhibitors through a SVM based virtual screening approach." Molecular Informatics. 29(2010) 407-420.
53. Morris GM, Huey R, Lindstrom W, Sanner MF, Belew RK, Goodsell DS, Olson AJ.AutoDock4 and AutoDockTools4: Automated docking with selective receptor flexibility. J Comput Chem.30 (2009) 2785-91.
54. Lang, P. T.; Brozell, S. R.; Mukherjee, S.; Pettersen, E. F.; Meng, E. C.; Thomas, V.;
Rizzo, R. C.; Case, D. A.; James, T. L.; Kuntz, I. D.DOCK 6: combining techniques to model RNA-small molecule complexes. RNA. 15 (2009 ) 1219–1230
55. Rarey M, Kramer B, Lengauer T, Klebe G. A fast flexible docking method using an incremental construction algorithm. J Mol Biol.261 (1996)470-89.
56. Jain AN. Surflex: fully automatic flexible molecular docking using a molecular similarity-based search engine. J Med Chem. 46(2003):499-511.
57. O. Trott, A. J. Olson, AutoDock Vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization and multithreading, Journal of Computational Chemistry. 31 (2010) 455-461
58. Verdonk, Marcel L., et al. Improved protein–ligand docking using GOLD. Structure, Function, and Bioinformatics. Proteins. 52(2003) 609-623.
59. Song CH, Yang SH, Park E, Cho SH, Gong EY, Khadka DB, Cho WJ, Lee K.Structure-based virtual screening and identification of a novel androgen receptor antagonist. J Biol Chem. 287(2012)30769-80.
60. Li S, Sun X, Zhao H, Tang Y, Lan M. Discovery of novel EGFR tyrosine kinase inhibitors by structure-based virtual screening. Bioorg Med Chem Lett. 2012 Jun15;22(12):4004-9
61. www.specs.net/snpage.php?snpageid=home, accessed 17th march 2014.
62. Foloppe N, Fisher LM, Howes R, Potter A, Robertson AG, Surgenor AE. Identification of chemically diverse Chk1 inhibitors by receptor-based virtual screening. Bioorg Med Chem.; 14(2006)4792-802.
63. Morley SD, Afshar M. Validation of an empirical RNA-ligand scoring function for fast flexible docking using Ribodock. J Comput Aided Mol Des. 18(2004):189-208.
64. Brown, R. D., & Martin, Y. C. The information content of 2D and 3D structural descriptors relevant to ligand-receptor binding. Journal of Chemical Information and Computer Sciences. 37(1997)1-9.
65. Siddiquee K, Zhang S, Guida WC, Blaskovich MA, Greedy B, Lawrence HR, Yip ML, Jove R, McLaughlin MM, Lawrence NJ, Sebti SM, Turkson J. Selective chemical probe inhibitor of Stat3, identified through structure-based virtual screening, induces antitumor activity. Proc Natl Acad Sci U S A.104 (2007)7391-6.
66. Nolan KA, Dunstan MS, Caraher MC, Scott KA, Leys D, Stratford IJ. In silico screening reveals structurally diverse, nanomolar inhibitors of NQO2 that are functionally active in cells and can modulate NF-κB signaling. Mol Cancer Ther.
11(2012)194-203.
67. Deng Z, Chuaqui C, Singh J. Structural interaction fingerprint (SIFt): a novel method for analyzing three-dimensional protein-ligand binding interactions. J Med Chem.
47(2004):337-44.
68. Zhou, R. H.; Friesner, R. A.; Ghosh, A.; Rizzo, R. C.; Jorgensen,W. L.; Levy, R. M.
New linear interaction method for binding affinity calculations using a continuum solvent model. J. Phys. Chem. B.105 (2001)10388-10397
69. Singh P, Mhaka AM, Christensen SB, Gray JJ, Denmeade SR, Isaacs JT. Applying linear interaction energy method for rational design of noncompetitive allosteric inhibitors of the sarco- and endoplasmic reticulum calcium-ATPase. J Med Chem.
48(2005):3005-14.
70. Baker NA, Sept D, Joseph S, Holst MJ, McCammon JA. Electrostatics of nanosystems:
application to microtubules and the ribosome. Proc. Natl. Acad. Sci. USA. 98(2001) 10037-10041
71. Naïm M, Bhat S, Rankin KN, Dennis S, Chowdhury SF, Siddiqi I, Drabik P, Sulea T, Bayly CI, Jakalian A, Purisima EO. Solvated interaction energy (SIE) for scoring protein-ligand binding affinities. 1. Exploring the parameter space. J Chem Inf Model.47 (2007)122-33.
72. Liu J, He X, Zhang JZ. Improving the scoring of protein-ligand binding affinity by including the effects of structural water and electronic polarization. J Chem Inf Model.;
53(2013)1306-14.
73. Barakat KH, Jordheim LP, Perez-Pineiro R, Wishart D, Dumontet C, Tuszynski JA.Virtual screening and biological evaluation of inhibitors targeting the XPA-ERCC1 interaction. PLoS One.7 (2012) e51329.
74. Lin JH, Perryman AL, Schames JR, McCammon JA. The relaxed complex method:
Accommodating receptor flexibility for drug design with an improved scoring scheme.
Biopolymers. 68(2003) 47–62.
75. Neyman, J.; Pearson, E. S. On the problem of the most efficient tests of statistical hypotheses. Philos. Trans. R. Soc, London, Ser. A.231 (1933)289−337.
76. Neyman, J.; Pearson, E. S. The testing of statistical hypotheses in relation to probabilities a priori. Proc. Cambridge Philos. Soc. 20 (1933)492−510.
77. Ferrari AM, Wei BQ, Costantino L, Shoichet BK. Soft docking and multiple receptor conformations in virtual screening. J Med Chem. 47(2004) 5076-84.
78. Li Y, Kim DJ, Ma W, Lubet RA, Bode AM, Dong Z. Discovery of novel check pointkinase 1 inhibitors by virtual screening based on multiple crystal structures. J Chem Inf Model.51 (2011) 2904-14.
79. Irwin JJ, Shoichet BK. ZINC--a free database of commercially available compounds for virtual screening. J Chem Inf Model.45 (2005) 177-82.
80. Ambaye, Nigus D., Menachem J. Gunzburg, Reece CC Lim, John T. Price, Matthew CJ Wilce, and Jacqueline A. Wilce. "The Discovery of Phenylbenzamide Derivatives as Grb7‐Based Antitumor Agents."ChemMedChem. 8 (2013)280-288.
81. Xie, Qing-Qing, Lei Zhong, You-Li Pan, Xiao-Yan Wang, Jian-Ping Zhou, Lei Di-wu, Qi Huang et al. "Combined SVM-based and docking-based virtual screening for retrieving novel inhibitors of c-Met." European journal of medicinal chemistry. 46 (2011) 3675-3680.
82. Dokla, Eman M., Amr H. Mahmoud, Mohamed SA Elsayed, Ahmed H. El-Khatib, Michael W. Linscheid, and Khaled A. Abouzid. "Applying ligands profiling using multiple extended electron distribution based field templates and feature trees similarity searching in the discovery of new generation of urea-based antineoplastic kinase inhibitors." PloS one. 7(2012) e49284.
83. Ren, Ji-Xia, Lin-Li Li, Ren-Lin Zheng, Huan-Zhang Xie, Zhi-Xing Cao, Shan Feng, You-Li Pan, Xin Chen, Yu-Quan Wei, and Sheng-Yong Yang. "Discovery of novel Pim-1 kinase inhibitors by a hierarchical multistage virtual screening approach based on SVM model, pharmacophore, and molecular docking."Journal of chemical information and modelling.51 (2011)1364-1375.
84. Kong, Xiangqian, Jie Qin, Zeng Li, Adina Vultur, Linjiang Tong, Enguang Feng, Geena Rajan et al. "Development of a novel class of B-RafV600E-selective inhibitors through virtual screening and hierarchical hit optimization." Organic & biomolecular chemistry. 10(2012)7402-7417.
85. Liu, X., Jiang, H., & Li, H. (2011). SHAFTS: a hybrid approach for 3D molecular similarity calculation. 1. Method and assessment of virtual screening. Journal of chemical information and modelling. 51(9), 2372-2385.
86. Krishna S, Singh DK, Meena S, Datta D, Siddiqi MI, Banerjee D. Pharmacophore- based screening and identification of novel human ligase I inhibitors with potential anticancer activity. J Chem Inf Model.54 (2014)781-92.
87. Morris, G. SYBYL Software, Version 6.9, Tripos Associates 2002.St. Louis, MO.
88. Lu, Yipin, Zaneta Nikolovska-Coleska, Xueliang Fang, Wei Gao, Sanjeev Shangary, Su Qiu, Dongguang Qin, and Shaomeng Wang. "Discovery of a nanomolar inhibitor of the human murine double minute 2 (MDM2)-p53 interaction through an integrated, virtual database screening strategy." Journal of medicinal chemistry. 49 (2006) 3759- 3762.
89. Fang, X.; Wang, S. A Web-based 3D-database pharmacophore searching tool for drug discovery.J. Chem. Inf. Comput. Sci. 42(2002) 192-198.
90. www.asinex.com/download-zone.html, accessed 30th april 2014.
91. www.dtp.nci.nih.gov/docs/cancer/searches/standard_mechanism.html, accessed 30th april 2014.
92. Tandon M, Wang L, Xu Q, Xie X, Wipf P, Wang QJ. A targeted library screen reveals a new inhibitor scaffold for protein kinase D. PLoS One.7 (2012):e44653.
