DSpace at VNU: A Rational Workflow for Sequential Virtual Screening of Chemical Libraries on Searching for New Tyrosinase Inhibitors

DSpace at VNU: A Rational Workflow for Sequential Virtual Screening of Chemical Libraries on Searching for New Tyrosinas...

Current Topics in Medicinal Chemistry, 2014, 14, 1473-1485 1473

A Rational Workflow for Sequential Virtual Screening of Chemical Libraries on Searching for New Tyrosinase Inhibitors

Huong Le-Thi-Thu1,*, Gerardo M Casañola-Martín2,3,4, Yovani Marrero-Ponce5,

Antonio Rescigno6, Concepción Abad2 and Mahmud Tareq Hassan Khan7

1 School of Medicine and Pharmacy, Vietnam National University, Hanoi (VNU) 144 Xuan Thuy, Cau Giay, Hanoi,

Viet-nam; 2 Departament de Bioquímica i Biologia Molecular, Universitat de València, E-46100 Burjassot, Spain; 3 Unidad

de Investigación de Diseño de Fármacos y Conectividad Molecular, Departamento de Química Física, Facultad de

Farmacia, Universitat de València, Spain; 4 Centro de Información y Gestión Tecnológica, Ministerio de Ciencia

Tecno-logía y Medio Ambiente (CITMA), 65100, Ciego de Avila, Cuba; 5 Enviromental and Computational Chemistry Group,

Facultad de Química Farmacéutica, Universidad de Cartagena, Cartagena de Indias, Bolivar, Colombia; 6 Sezione di

Chimica Biologica, Dip Scienze e Tecnologie Biomediche, Università di Cagliari, Cittadella Universitaria, 09042

Mon-serrato (CA), Italy; 7 Present address: Holmboevegen 3B, 9010 Tromso, Norway

Abstract: The tyrosinase is a bifunctional, copper-containing enzyme widely distributed in the phylogenetic tree This

en-zyme is involved in the production of melanin and some other pigments in humans, animals and plants, including skin

pigmentations in mammals, and browning process in plants and vegetables Therefore, enzyme inhibitors has been under

the attention of the scientist community, due to its broad applications in food, cosmetic, agricultural and medicinal fields,

to avoid the undesirable effects of abnormal melanin overproduction However, the research of novel chemical with

anti-tyrosinase activity demands the use of more efficient tools to speed up the anti-tyrosinase inhibitors discovery process This

chapter is focused in the different components of a predictive modeling workflow for the identification and prioritization

of potential new compounds with activity against the tyrosinase enzyme In this case, two structure chemical libraries

Spectrum Collection and Drugbank are used in this attempt to combine different virtual screening data mining

tech-niques, in a sequential manner helping to avoid the usually expensive and time consuming traditional methods Some of

the sequential steps summarize here comprise the use of drug-likeness filters, similarity searching, classification and

po-tency QSAR multiclassifier systems, modeling molecular interactions systems, and similarity/diversity analysis Finally,

the methodologies showed here provide a rational workflow for virtual screening hit analysis and selection as a

promis-sory drug discovery strategy for use in target identification phase

Keywords: Drug-likeness filtering, molecular docking, QSAR modeling, similarity searching, tyrosinase inhibitor, virtual

screening

1 INTRODUCTION

Tyrosinase (monophenol monooxygenasse; EC

1.1.4.18.1) is a metalloenzyme oxidase widely distributed in

the phylogenetic tree This enzyme catalyze the two first

steps of melanin synthesis pathway, by the hydroxylation of

the L-tyrosine to 3,4-dihydroxyphenylalanine L-DOPA

(mo-nophenolase activity), and the posterior oxidation to

do-paquinone (diphenolase activity) [1] Because its main role

in melanogenesis, abnormal tyrosinase regulation it is related

with some skin diseases such as hyperpigmentation,

melasma (acquired hyperpigmentation), post in ammatory

melanoderma, solar lentigo, etc [2-4] Hence, tyrosinase

in-hibitors have been used largely as depigmenting agents for

treatment of these pigmentation disorders [5-7]

*Address correspondence to this author at the School of Medicine and

Pharmacy, Vietnam National University, Hanoi (VNU) 144 Xuan Thuy,

Cau Giay, Hanoi, Vietnam;

Tels: 53-33-223066 (Cuba) and 963543156 (València);

Faxes: 53-33-223066 (Cuba) and 963543156 (València);

E-mails: gmaikelc@gmail.com or gmaikelc@yahoo.es

In recent times, besides QSAR methodologies, there are other data mining techniques introduced in drug discovery with high accuracy levels [8] This successful data integra-tion is a complex theme in the desktop today´s researchers

In this sense the current drug discovery scenarios are intro-ducing standard workflow for screening structural chemical libraries [9] This issue presents advantages because com-pounds could be obtained by the direct purchase from the owners avoiding the time-consuming of synthesis or isola-tion process [10] Moreover, in the last times, the campaigns associated with massive virtual HTS are continuously in-creasing are gaining on accuracy in the prioritization process

of chemicals, due to the introduction of several ligand and structure based methodologies [11-12]

Therefore, to manage larger compounds libraries and

cover the expectations of the drug discovery process, in

silico virtual screening and computer-aided drug design have

become increasingly important [13] In our research group, this approach has been applied in the discovery of novel ty-rosinase inhibitors (TIs) as cycloartane [14-15], diterpenoi-dal alkaloids [16], tetraketones [17], coumarin [18], and so

1873-5294/14 $58.00+.00 © 2014 Bentham Science Publishers

on In these studies, the classification QSAR-based virtual

screening (VS) has been employed for in-house congeneric

chemical libraries of different laboratories to identify TIs

from inactives Latter, new class of QSAR models named

potency models [19-20] were developed The last models

could be used as a cascade system together with the

classifi-cation models for more complete description of tyrosinase

inhibitory activity Besides, this type of models helps to

identify true positives and make an adequate process of

pri-oritization of compounds In recent work, all these models

were assembled in different multi-classifier systems (MCSs)

that improved the performance of QSAR methods [21]

By this means, in this chapter we present a procedure of

combining these and others different VS strategies in the

computational research for the selection/identification of

novel tyrosinase inhibitors This framework was employed

with efficacy to discover new chemical entities with

anti-tyrosinase activity Finally, is important to stand out that the

different virtual screening approaches mentioned comprises:

drug-likeness filters, similarity searching, classification and

potency QSAR multi-classifier systems, molecular docking

studies, and post-processing procedures as strategies; that

were assessed in a sequential manner over the Spectrum

Collection and Drugbank databases

2 TWO STRUCTURE CHEMICAL LIBRARIES:

SPECTRUM COLLECTION AND DRUGBANK

The ascending grown of computational resources have

brought a rapid increasing of structural chemical databases

either online or repository company The more interesting

examples are the huge ZINC and ChemSpider databases,

comprising 13 million and 26 million of compounds,

respec-tively This two chemical data sources are included between

the main sixty-four free databases nowadays [22] Therefore,

the huge structural chemical database screening is becoming

one of the hotter topics in compound retrieve using any

QSAR, ligand or structure data mining procedures In this

sense, some authors have included interesting updated

re-views about this topic [23-24] By this mean here we

pre-sented the results obtained over the Spectrum collection

(http://www.msdiscovery.com/spectrum.html) and

Drug-bank (www.drugbank.ca), which consists of 2 000 and 6 827

compounds, respectively, that were screened using a

sequen-tial strategy for virtual screening looking for potensequen-tial

thera-peutic chemicals for the treatment of hyper-pigmentation disorders A owchart depicting the various steps of virtual

screening including database ltration, similarity searching,

QSAR modeling, docking and clustering studies to prioritize

the virtual hits is shown in (Fig 1)

This Fig (1) displays the virtual screening stepwise

workflow which resulted in the discovery of novel scaffolds against the tyrosinase enzyme The protocol was based in the computational hierarchy of each filter, the consuming CPU time and the complexity of input information for each step This hierarchical procedure allows reducing the number of selected compounds (retrieved as novel TIs) gradually after each filter This mentioned strategy was employed to screen

virtually two databases (Spectrum Collection and

Drug-bank )

By other way, many of the chemical libraries as the case

of PubChem [25] on-line database are web-based systems with well recognize facilities to do some pre-processing

tasks in an easy way This is the also the case Spectrum

collection and Drugbank were some drug-likeness filters or

similarity searching methods are implemented as tools for search and retrieving Therefore some of these services were used in these studies

3 DRUG-LIKENESS FILTERS

The term “druglike” [26-29] is used for pharmaceutical

research to describe molecules with properties that fall within the boundaries delineated by the wide majority of pharmaceutical agents This process is associated with the

many possible molecular properties that most directly influ-ence the drug-like properties of a molecule in some specific type of research Lipinski et al.[30] defined the so-called

“rule of five” (sometimes abbreviated as RoF) in an effort to solve this question The main steps of this concept is the ex-amination of different parameters such as the number of

ro-tatable bonds (nRotB), polar surface area (PSA), log D, and

counts of nitrogen and oxygen atoms in an effort to define easily calculated properties that will be predictive of a

favor-able outcome and established mayor cutoff for these

physi-Fig (1). Sequential virtual screening workflow used in the identification of promissory TIs and the filtering of compounds involved in each

one of different steps from the Spectrum Collection and DrugBank databases

cal-chemical properties and others [31-35] However the

threshold of Lipinski seems very rigid in occasions Hence

some scientist in this field have stand out and proposed other

diverse boundaries and criteria of drug-likeness filters [36]

Taking this into consideration, in our work we applied

supe-rior limits of all these filters In our case, a compound was

not taken into consideration in the next steps if it has the

molecular weight (MW) above 700 g/mol; the computed

octanol–water partition coefficient CLogP higher than 7;

the number of hydrogen bond donors (nHBDon) above 5

and acceptors (nHBAc) above 10; the number of rotatable

bonds (nRotB) higher than 10 and a polar surface area

(PSA) above 140 Å2 All these descriptors were calculated

with our in house TOMOCOMD-CARDD (acronym for

TO pological MOlecular COMputational Design -

Com-puted-Aided ‘Rational’ Drug Design) software These

mo-lecular descriptors are implemented in a new module

(DE-SPOOLs , acronym of DEScriptor POOLs) of our program

[37] that offers calculations of the several 0-3D indices,

which are calculated mainly using The Chemistry

Devel-opment Kit [38] By using the defined way above we

pro-ceed to the first filtering consisting in the application of the

criteria describe above on the Spectrum collection

data-base (http://www.msdiscovery.com/spectrum.html) This

first step also consists of reducing the number of chemicals

(negative design) employing the Drug-likeness filters These

are simple, fast and also allow “optimizing” in some way

simultaneously the potency and the pharmacokinetic [39]

So, we further sorted these 2 000 compounds using the

supe-rior boundaries of all filters reported in the literature and

nally 1 394 compounds were further considered for the next

step

4 SIMILARITY SEARCHING

Similarity searching identifies those database molecules

that are most similar to reference structures, using some

quantitative definition of intermolecular structural similarity

The reference structures and the molecules in the database

are characterized by one or more molecular descriptors

Their comparisons allow the calculation of a similarity

measurement between the reference structure and each of the

database structures, and the latter ones are then sorted into

order of decreasing similarity with the target The output

from the search is a ranked list in which the structures that

are calculated to be most similar to the reference structure

are located at the top of the list These chemicals will be

those that have the greatest probability of being of interest to

the user, given an appropriate measure of intermolecular

structural similarity The similarity methods are extremely

useful at the beginning of a drug discovery project, because

it needs little information about the target and only few

known active compounds Moreover, the implementations of

similarity methods are generally computationally

inexpen-sive, so searching large databases can be routinely

per-formed The result of this step is a focused library, since all

included compounds present common features a reference

compounds In our case the data fusion method [40] was

applied In Table 1, the structures of reference compounds

were given A hierarchical cluster analysis, k-NNCA, was

executed to visualize the distribution of reference

com-pounds in different groups In Fig (2), a dendogram for

these compounds is shown It can be seen, there is great structural diversity among these chemicals, which represent different molecular subsystems important for the tyrosinase activity

The set of 15 strong tyrosinase inhibitors of diverse struc-tures was selected as reference compounds The molecular

structures of these chemicals are given in Table 1

The MACCS fingerprints [41] were calculated to charac-terize reference structures and the ones of the database

em-ploying the program TOMOCOMD-CARDD software

These fingerprint are implemented in a new module

(MOLFIP, acronym of MOLecular FIngerPrints) of our

program [37] that offers calculations of the several finger-prints, which are calculated mainly using The Chemistry

Development Kit [38] The Tanimoto coefficient [42],

com-monly used for binary data, was computed to establish the metrics of intermolecular comparison (each compound with every other in its activity class) A specific database

mole-cule appears at rank position r ij (1
i
n, 1
j
15) with a similarity measurement (scores), s j against every reference structure We used the fusion rule MAX for combining the similarity scores, so the final fused score was established as

shown in Equation 1

s f = Maximum {s 1 , s 2 …s 15} (1)

Later, each molecule of the database was sorted by its

fused score, s f The similarly active compounds in the top 30% of highest ranked data set compounds were retrieved

for the next step For the case of the Spectrum collection

database structures this procedure was applied resulting in the elimination of 1285 molecules The remaining 109 pounds are similar in some way to one of 15 reference com-pounds (positive design)

The Drugbank database (www.drugbank.ca) of 6 827

drugs was also screened using a similar procedure as above

In this case, first the similarity searching (data fusion by maximum score using 15 strong TIs as reference structures)

was applied, because DrugBank offers this option in the

management of its search database By this procedure were eliminated 6659 compounds representing the 97.54% of the chemicals in the database The repeated or reported against the tyrosinase compounds were removed and the remaining

ones were ltered using Druglikeness criteria mentioned in

the section above From this, 131 compounds were selected

and considered for the next step

5 MULTICLASSIFIERS GUIDED BY QSAR MODELS

In recent times, Quantitative Structure-Activity Relation-ships (QSARs) are the most widely used approach in drug design and have been applied successfully in the discovery

of novel tyrosinase inhibitors [18, 43-49] Hence, this method could constitute the principal “switch” for sequential workflow aiding to new lead compounds identification The binary QSARs for tyrosinase inhibitors are described in pre-vious reports [18, 43-49], therefore a brief approximation will be discussed here A first training set of 1072 com-pounds was collected with 526 chemicals classi ed as “ac-tive” (TIs) and 546 compounds as “inac“ac-tive” (non-TIs) The molecular structures and properties were correlated with

biological activity using TOMOCOMD-CARDD descriptors,

Table 1 Structures of reference compounds in similarity study

O

O HO

1 Kojic acid

NH2 N

HO

2 L-mimosine

O OH

3 L-Tropolone

N OH

N O

4 N-cyclopenthyl-N-nitrosohydroxyl-amine

HO

OH O

O

5 Methyl ester of gentisic acid

O O

HO

6 Kurarinone

HN S

H2N

7 Phenyl-thiourea

N N

8 BP4

O O

O

O

OH O

H

O

HO

9 8´-epi-cleomiscosin A

O

O

10

HO

OH HO

HO

11 4-Prenyloxy- resveratrol

HN

S O

12 Alkyl-thiocarbamate E

NH

O

OH HO

OH HO

13

Benzylbenzamide 15

N NH S

NH2

14 3-Hydroxy -4-methoxy benzaldehyde

thiosemicarbazone

O

O

O O

NH 2

15 TK21

and different classification models were generated These

models enable the identification of TIs from inactive ones In

second place, the potency models were obtained using a

learning set of 257 strong TIs and 141 moderate-to-weak

compounds [19-20] The last ones would be used

hierarchi-cally with the models adjusted on the first database, for more

complete description of tyrosinase inhibitory activity

Afterward, we introduced other statistical techniques

[quadratic discriminant analysis (QDA), binary logistic

re-gression (BLR) and classification tree (CT) [20]] and many

machine learning approaches [support vector machine

(SVM), artificial neural network (ANN), Bayesian networks

(BNs), k-nearest neighbors (KNN) [19]], which enhanced

the performance of previous LDA-QSAR models in both

database Theses single classifiers can be used to make

ty-rosinase inhibitory activity depictions for new chemicals

However, many factors can affect the performance of those

classifiers Selecting the best available classifier is an option, but because the distribution of new chemicals that the classi-fier may meet during operation may vary (slightly or signifi-cantly depending on the application), this approach does not provide the best solution in all cases Furthermore, because many classifiers are generally tried before a single classifier

is selected, this approach also discards valuable information

by ignoring the performance of all the other classifiers [50]

By this aim, the combination of multiple classifiers has been proposed in the field of machine learning to improve the performance of the single classifier approaches [51-53] These multiple classifier systems (MCS) are based on the combination of several classifiers such that their union achieves higher performance than the stand-alone classifiers Hence, an ensemble of classifiers is a set of classifiers, whose individual classification decisions are combined in some way [54] Many studies have demonstrated that

Fig (2). Dendrogram illustrating the results of the hierarchical k-NNCA of strong TIs used as reference compounds

ensembles often outperform their base models (the

compo-nent models of the ensemble) if the base models perform

well on novel examples and tend to make errors on different

examples [55]

In the case of tyrosinase inhibitor QSAR equations, to

in-crease performance demands in modeling tyrosinase activity

the individual models obtained were assembled in different

multi-classifier systems (MCSs) to improve their

perform-ance classifiers for tyrosinase inhibitory activity prediction

[21]

For the Spectrum Collection, the compounds found by

similarity searching were screened by QSAR (another

posi-tive design approach) using classification MCS based on

average probability (AP) [21] to identify new TIs Thus,

65/109 compounds were identified as active against

ty-rosinase enzyme It is important to highlight that most of

inactive compounds identified by QSAR were the last ones

of the list of 109 compounds The same occurs in the case of

the Drugbank database were 119/131 were identified as

ac-tives This justifies the selection of the cutoff value of

simi-larity searching The compounds identified by classification

QSAR were sequentially screened by potency QSAR using

boosting ensemble based on support vector machine [21]

This potency MCS identified 25 and 107 compounds from

Spectrum Collection and Drugbank, respectively as strong

TIs

6 MODELING MOLECULAR INTERACTIONS

SYS-TEMS

The next step was to use the molecular docking, that

con-sists of posing each ligand into the binding site of the target

This gives a predicted binding mode for each database

com-pound, together with a measure of the quality of the fit of the compound in the target binding site This information is used

to rank the compounds with a view to selecting and experi-mentally testing a small subset for biological activity [56] The docking calculations of strong TIs identified by QSARs

in the mentioned above studies were performed using the ICM™ docking module with the default setup as earlier mentioned [57-59]

6.1 Preparations of the Inhibitors and Target Molecules

The 2D structure of the compound (in mol file format) was converted to 3D and energy minimized at the 3D space

of ICM environment The atom types using local chemical environment, Merck Molecular Force Field (MMFF) [60-66] formal charges and 3D topology were assigned The lowest energy conformers of the compounds were then docked into the 3D space of the active site of the three dimensional struc-ture of Tyrosinase (PDB ID: 3NQ1)

All the docking calculations were performed using the

“interactive docking” menu at the ICM environment After docking the stack of docking poses were checked visually Multiple stack conformations were selected based on their docking energies, rmsd values (compared between the docked model and x-ray conformation) and similarities to closely related x-ray crystal structures from PDB Then the best conformations for the compound were finally chosen, and then the binding energy was calculated using ICM script For each of the individual docked complexes the free en-ergies of binding (
Gcal) between the protein and ligand was

calculated using ICM script utilizing Equations 2 and 3[67]

Gcal. =
GH +
Gel. +
Gs + C (2)

Gcal. =
GH +
Gcol. +
Gdes-sol. +
Gs + C (3)

Here,
GH is the hydrophobic or cavity term, which

ac-counts for the variation of water/non-water interface area

Gel is the electrostatic term composed of coulombic (
Gcol)

interactions and desolvation (
Gdes-sol) of partial charges

transferred from an aqueous medium to a protein core

envi-ronment The
Gs is the entropic term which results from the

decrease in the conformational freedom of functional groups

buried upon complexation; and finally the C is a constant

accounts for the change of entropy of the system due to the

decrease of free molecules concentration (cratic factor), and

loss of rotational/translational degrees of freedom [67]

After preparation of docking process the strong TIs

iden-tified by QSARs were subsequently docked in the active site

of the tyrosinase (PDB ID: 3NQ1) using the ICM program

[57] In this study, only the chemicals selected of the

Spec-trum Collection were used on the structure-based study

Docking is an effective method for prioritizing ligands with

favorable interactions with the receptor and can also be seen

as a positive design For each case, the binding energy (BE)

was achieved and used as score to binding mode prediction

of the compounds First, we calculated the BR for a set of

strong, moderate-weak and inactive compounds with known

activity against the tyrosinase The docking molecular

inter-action process revealed that only one compound of the total

of 25 selected by QSAR did not complement favorably the

protein binding site This result showed a good

correspon-dence between QSAR and docking approaches

7 POST-PROCESSING ANALYSIS

Some methodologies for post-processing after sequential

virtual screening were assessed In our case, we selected the

k-NNCA (k-nearest neighbors cluster analysis) and k-MCA

(k-means cluster analysis) algorithms [68-69] to study the

similarity/diversity among the retrieved active compounds

and these latter ones with active compounds This two types

of Cluster Analysis (CA) were chosen because are a group of

methods capable to recognize similarities among cases

(ob-jects) or among variables and single out some categories as a

set of similar cases (or variables) Therefore it enables the

selection of novel scaffold for tyrosinase inhibitors Before

carrying out the cluster processes, all the variables were

standardized In standardization, all values of selected

vari-ables (molecular descriptors) were replaced by standardized

values, which are computed as follows: Std score = (raw

score - mean)/Std deviation

Finally, by a CA of the database active compounds and

the retrieved ones plus a detailed visual inspection, for the

case of Spectrum Collection, 19 out of 24 compounds were

selected to be evaluated experimentally It is important to

note that within the six compounds removed, some have

been reported in the literature activity against tyrosinase,

such as hinokitiol [70] and angolensine [71], while the

Spec-trum does not report itself This fact confirmed the

applica-bility of our protocol in the discovery of novel lead

com-pounds anti-tyrosinase from large databases Table 2 shows

traditional uses and values of different "in silico" studies, the

molecular structures of these compounds are given in Fig

(3)

On the case of Drugbank database, after cluster analysis and visual inspection we decided to select 32 compounds of the for enzyme assays The structures of these compounds

are shown in Table 3 and Table 4 shows traditional uses and

values of different "in silico" studies for these drugs

The flowchart in Fig (1) is a schematic representation of

the rational workflow sequential VS process with the number

of hits reduced for each screening step in both databases Using the sequential workflow, a total of fifty one putative novel TIs were successfully identified, which can be pur-chased and further evaluated in enzymatic experimental cor-roborations

As it can be seen in both cases, many compounds identi-fied as new TIs are already known drugs because and this avoids time-consuming to bring new drugs to market be-cause re-discovered drugs that are already in use and its pharmacokinetic and toxicological properties are well-known [72] This novel discovered drugs could be introduced into the market in the shortest time possible, thus accelerating the speed of discovery of new drugs for treating disorders of hyperpigmentation

8 FUTURE TRENDS ON WORKFLOWS FOR BIO-ACTIVITY PREDICTION

Bioactivity or any type of property prediction has always

be one of the challenges on data mining fields In the case of selection of adequate anti-target activity the main arduous task are mainly focused in the correct identification of lead compounds or promissory high activity chemicals that could lead to drug-like compounds after examining its ADMET properties Many predictive workflows has been showed in literature most of them focused on the use of 3D-QSAR COMFA, COMSIA and pharmacophore approaches together with docking studies [73-74] Because the drug discovery is

a highly complex and costly process, the integration in inno-vation, knowledge, information, technologies, expertise, in-vestments and management skills is required In this way, the multistep VS can help identify bioactive substances from

a large screening compound pool with limited experimental effort enabling to focus rapidly on the most promising can-didate structures

In the case of the specific workflow scenarios, some questions that remains unsolved were derived during writing this chapter, like the use of sub-workflows integrated by sev-eral Multi-Sequential Workflow responsible of each step for adequate drug-like properties, that is ADMET Moreover the consideration of other aspects concerning to workflow like the accuracy, sequence of combination, the most better quan-tity of sub-workflows to be used, and thresholds established for any workflow should be examined

Finally, these results offers a suitable alternative to the new era of open on-line chemical databases encouraging its use together with the ascending approaches based on the new technologies development such as massive computer calcula-tions algorithms and cloud computing could have a over-whelming impact on virtual screening procedures based on ensemble workflows to solve several questions that are still

in the route of drug discovery pipeline

Table 2 Results of different in silico filters of VS protocol on the Spectrum Collection database

ID *

Compound Bioactivity MWa

LogP b

nHBDonc

nHBAd

nRotBe

PSA f

S f

Simi-larity Ranking

Pre-dicted class h

Esti-mated Potency i

Docking

BE k

(Kcal/ mol)

1500485 Phenytoin

sodium

Anticonvulsant, antieleptic 251.08 4.89 1.00 4.00 2.00 46.17 0.825 7 Act P -3.7

1503801 Naproxol

Antinflamma-tor, analgesic, antipyretic

468.21 2.35 0.00 6.00 5.00 95.34 0.816 10 Act P -2.2

1505130

3,4-Dimethoxy-cinnamic acid - 208.07 1.75 1.00 2.00 4.00 55.76 0.786 19 Act P -3.1

1505311

Diben-zoylmethane Antineoplastic 224.08 5.72 0.00 2.00 4.00 34.14 0.756 38 Act P -4.1

1505140

2',4-Dihydroxy-

3,4',6'-

trimethoxy-chalcone

- 330.11 2.18 2.00 1.00 6.00 85.22 0.742 49 Act P -6.6

1504152 Nilutamide Antiandrogen 317.06 3.22 1.00 4.00 3.00 92.55 0.735 59 Act P -3.3

1503032 Dipyrocetyl Antirheumatic,

analgesic 238.05 1.59 1.00 4.00 5.00 89.90 0.724 67 Act P -0.7

300610 Acetosyringone

Insect attractant, plant hormone

196.07 0.45 1.00 1.00 3.00 55.76 0.724 69 Act P -1.3

1504209 Diplosalsalate antipyretic analgesic, 300.06 4.12 1.00 4.00 6.00 89.90 0.724 71 Act P -1.5

1504118 Difractaic acid - 374.14 3.37 2.00 3.00 6.00 102.29 0.719 79 Act P -6.5

201448 4,4'-Dimethoxy

dalbergione - 284.10 2.88 0.00 3.00 5.00 52.60 0.710 96 Act P -5.5

2300228 Kainic acid

Glutamate receptor agonist, anthelmintic

213.10 -0.21 3.00 5.00 4.00 86.63 0.708 98 Act P -1.9

1505673 Troglitazone Antidiabetic 441.16 3.52 2.00 3.00 5.00 110.16 0.700 111 Act P -7.2

330032 Dicamba Herbicide 219.97 1.78 1.00 2.00 2.00 46.53 0.700 115 Act P -0.9

* ID=Code of Spectrum Collection;a MW = Molecular weight; b LogP = Computed octanol/water partition coefficient; cnHBDon = Number of hydrogen bond donors; dnHBAc= Num-ber of hydrogen bond acceptors; f PSA = Polar surface area; g S f =Fused Score for the maximum of the similarity; hAct =Active against the tyrosinase identified by Clasiffication MCS

QSAR; iP = Potent inhibitor of tyrosinase identified by Potency MCS QSAR; k BE =Binding energy (PDB ID: 3NQ1)

Fig (3) Molecular structures of the identified virtual hits from Spectrum Collection by VS protocol

Table 3 Molecular structures of virtual hits identified on Drugbank by current VS protocol

H

OH

O O O

O

DB00227

H O

H O HO

DB00486

H

OH

O O O O

DB00641

O O

OH O

HO

DB00769

OH

HO

O O

O

DB00929

H

O O

O HO

DB02205

O

HO O

O

O

OH

O

DB02329

O HO

DB02699

HN

O

F

OH

Br

DB02880

OH O O

DB03007

HO

O

OH

OH

DB03451

H

H O O

H

H H

HO

OH

O

DB03785

H O

OH

H O

O

OH

DB04324

H HO

HO

OH O O

DB04376

H

O O

DB04392

O

O N O

DB04599

(Table 3) contd…

HO

OH

DB04641

F

H

H O O

HO

OH

DB07036

O N

DB07123

H OH O

H

HO O

H O

DB07177

O

O OH

OH

DB07500

H OH

O H

O N H

HO

DB07567

H H

H HO H

H H

O

DB07703

SH O

HN N

DB07734

O

O N

H

O

SH

DB07735

H

H

NH2

O H

DB07883

H H

H O

OH HO

DB07933

H O

OH O

HO

H H

DB08020

H

OH

O O H H

H

H

OH

H

DB08224

O

HO HO

DB08442

H

O O

O HO

OH

DB08517

H

H H O

HO HO

DB08737

Table 4 Results of different in silico filters of virtual screening protocol on the DrugBank database

ID *

LogP b

nHBDonc

nHBAd

nRotBe

PSA f

Similarity fused score

Predicted class h

Esti-mated potency i

DB00227 Lovastatin approved,

DB00486 Nabilone approved, inves-tigational 372.27 6.155 1 1 6 46.53 0.573 Act P

DB02205

6-(1.1-Dimethylallyl)-2-(1-

Hydroxy-1-Methylethyl)-

2.3-Dihydro-7h-Furo[3.2-G]Chromen-7-One

(Table 4) contd…

ID *

LogP b

nHBDonc

nHBAd

nRotBe

PSA f

Similarity fused score

Predicted class h

Esti-mated potency i

DB02880

N-[1-(4-Bromophenyl)-Ethyl]-5-Fluoro Salicylamide experimental 337.01 2.77 2 2 4 49.33 0.551 Act P

DB03451

1alpha.25-Dihydroxyl-20-

Epi-22-Oxa-24.26.27-Trihomovitamin D3

experimental 460.355

DB03785

(3r.5r)-7-((1r.2r.6s.8r.8as)-

2.6-Dimethyl-8-{[(2r)-2-

Methylbutanoyl]Oxy}-

1.2.6.7.8.8a-Hexahydrona-

phthalen-1-Yl)-3.5-Dihydroxyheptanoic Acid

DB04392

Bishydroxy[2h-1-

Benzopyran-2-One.1.2-Benzopyrone] experimental 336.06 0.156 0 4 2 86.74 0.573 Act P

DB04641

3.7-Dihydroxynaphthalene-2-carboxylic acid experimental 204.04 0.914 3 2 1 77.76 0.659 Act P

DB07036

(3aS.4R.9bR)-2.2-difluoro-4-

(4-hydroxyphenyl)-6-

(methoxymethyl)-

1.2.3.3a.4.9b-hexahydro-cyclopenta[c]chromen-8-ol

DB07123

n-(4-methylbenzoyl)-4-benzylpiperidine experimental 293.18 4.138 0 2 4 20.31 0.556 Act P

DB07177

(5e.13e)-11-hydroxy-9.15-dioxoprosta-5.13-dien-1-oic

acid

DB07500

(2E)-1-[2-hydroxy-4-meth-

oxy-5-(3-methylbut-2-en-1-

yl)phenyl]-3-(4-hydroxy-phenyl)prop-2-en-1-one

DB07567

(2r.3r.4s)-3-(4-

hydroxyphenyl)-4-methyl-2-[4-(2-pyrrolidin-1-ylethoxy)

phenyl]chroman-6-ol

DB07703

(3r.4s.5s.7r.9e.11r.12r)-12-

ethyl-4-hydroxy-3.5.7.11-

tetramethyloxacyclododec-9-ene-2.8-dione

DB07734

N-(1-benzylpiperidin-4-yl)-4-sulfanylbutanamide experimental 292.16 2.058 1 3 7 71.14 0.708 Act P

DB07735

N-[1-(2.6-

dimethoxybenzyl)piperidin-4-yl]-4-sulfanylbutanamide