ResearchArticle 2D-QSAR and 3D-QSAR Analyses for EGFR...

Research Article2D-QSAR and 3D-QSAR Analyses for EGFR Inhibitors

Manman Zhao1 LinWang2 Linfeng Zheng3 Mengying Zhang1

Chun Qiu2 Yuhui Zhang4 Dongshu Du15 and Bing Niu1

1Shanghai Key Laboratory of Bio-Energy Crops College of Life Science and Shanghai University High Performance Computing CenterShanghai University Shanghai 200444 China2Department of Oncology Hainan General Hospital Haikou Hainan 570311 China3Department of Radiology Shanghai General Hospital Shanghai Jiao Tong University School of Medicine Shanghai 200080 China4Changhai Hospital Second Military Medical University Shanghai 200433 China5Department of Life Science Heze University Heze Shandong 274500 China

Correspondence should be addressed to Yuhui Zhang gong_chang2008126com Dongshu Du dsdushueducnand Bing Niu phycocy163com

Received 6 January 2017 Revised 19 February 2017 Accepted 12 March 2017 Published 29 May 2017

Academic Editor Vladimir Bajic

Copyright copy 2017 Manman Zhao et alThis is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Epidermal growth factor receptor (EGFR) is an important target for cancer therapy In this study EGFR inhibitors were investigatedto build a two-dimensional quantitative structure-activity relationship (2D-QSAR) model and a three-dimensional quantitativestructure-activity relationship (3D-QSAR) model In the 2D-QSARmodel the support vector machine (SVM) classifier combinedwith the feature selection method was applied to predict whether a compound was an EGFR inhibitor As a result the predictionaccuracy of the 2D-QSARmodel was 9899 by using tenfold cross-validation test and 9767 by using independent set testThenin the 3D-QSAR model the model with 1199022 = 0565 (cross-validated correlation coefficient) and 1199032 = 0888 (non-cross-validatedcorrelation coefficient) was built to predict the activity of EGFR inhibitors The mean absolute error (MAE) of the training set andtest set was 0308 log units and 0526 log units respectively In addition molecular docking was also employed to investigate theinteraction between EGFR inhibitors and EGFR

1 Introduction

Epidermal growth factor receptor (EGFR) a transmembraneglycoprotein is classified to the prototype of receptor tyrosinekinases (TKs) family that includes EGFR ErbB-2 ErbB-3and ErbB-4 EGFR is activated by its cognate ligands viaforming a homodimer or heterodimer with othermembers ofthe EGFR family such as epidermal growth factor (EGF) andtransforming growth factor alpha (TGF-120572) [1] Several signaltransduction cascades are initiated when EGFR is activatedand then lead to DNA synthesis and cell proliferation [2 3]While EGFR is amplified or mutated DNA synthesis and cellproliferation will be abnormal and lead to cancer Currentlythe amplification or mutation of EGFR has been found inhuman solid tumors such as glioma lung cancer ovariancancer and breast cancer Hence EGFR is also considered tobe a potential anticancer target in this disease [4ndash8] ManyEGFR inhibitors have been developed and approved by the

FDA such as lapatinib which has been applied for the treat-ment of breast cancer [9] Moreover other EGFR inhibitorslike temozolomide lomustine erlotinib and gefitinib areapproved by the FDA for the treatment of glioma [10 11]However the existing EGFR inhibitors are beyond peoplersquosexpectation due to selectivity toxicity and side effect Henceit is necessary to design and synthesize new potential EGFRinhibitors

Quantitative structure-activity relationship (QSAR)was avaluable tool for many different applications including drugdiscovery predictive toxicology and risk assessment [12ndash14]The applicability domain of QSAR models defined by theOrganization for Economic Co-operation and Development(OECD) according to Principle 3 includes the physicochem-ical the structural and the biological domain [15ndash17] Ini-tially two-dimensional quantitative structure-activity rela-tionship (2D-QSAR) was widely explored and used inmedic-inal chemistry study However some limitations spurred

HindawiBioMed Research InternationalVolume 2017 Article ID 4649191 11 pageshttpsdoiorg10115520174649191

2 BioMed Research International

the appearance of three-dimensional quantitative structure-activity relationship (3D-QSAR) In the 3D-QSAR study thecorrelation between 3D steric and electrostatic fields andbiologically activity draws attention For the molecular fieldstudy CoMFA was widely used preliminarily However thetime-consuming limit stimulates the advent of TopCoMFATopCoMFA overcomes the weakness and uses an objectivemethod to fragment and align the molecules In additionthe fragmentation process is automated except for somespecific bonds that should be cleaved manually Of courseTopCoMFA and CoMFA also have similarity that they bothshare QSAR PLS analysis The details about TopCoMFA andCoMFA are in [18]

Drug development is a long process and it requiresa vast amount of material and financial resources QSARand molecular docking technology have been extensivelyemployed in drug virtual screening and potential moleculartargets prediction which may shorten the cycle of the drugdevelopment [19ndash22] In this work 2D-QSAR model wasemployed to determine EGFR inhibitor and the 3D-QSARmodel was used to predict the activity Finally moleculardocking was applied to investigate the binding sites

2 Materials and Methods

21 CfsSubsetEval Method and Greedy Stepwise Algorithm Adata set containing 119899 vectors has 2119899 possible combinations offeatures for the subset A useful subset which can correctlypredict other compounds is one of 2119899 combinationsThe bestway to find an optimal subset is to try all the possible featurecombinations However this strategy is difficult to carry outdue to the huge computation In this study the CfsSubsetEval(CFS) search method combined with Greedy Stepwise (GS)algorithmwas employed to search the optimal feature subsetThemain idea of the GS algorithms is to make the best choicewhen selecting good features The CFS method was used toevaluate the attribute Thus the CFS method combined withthe GS algorithm was employed to select the optimal subsetfrom these 2119899 combinations Additional details about the CFSmethod and the GS algorithm could be found in [23ndash25]

22 SVM Support vector machine (SVM) a supervisedlearning algorithm is usually used for pattern recognitionclassification [26] SVM was employed for the classificationand sensitivity analysis in our study due to its high perfor-mance in many studies [25 27 28]

23 Topomer CoMFA Topomer CoMFA possessing boththe topomer technique and CoMFA technology can over-come the alignment problem of CoMFA [18 29] Partialleast squares (PLS) regression is employed to build thetopomer CoMFAmodel and the leave-one-out (LOO) cross-validation is used to evaluate the model Additional detailsabout the topomer CoMFA can be found in [29ndash31]

24 Data Preparation 100 inhibitors derived from the lit-erature and 185 noninhibitors downloaded from the DUDdatabase (httpduddockingorg) were collected [32ndash41] For

2D-QSAR study the data set containing inhibitors andnoninhibitors was randomly divided into three training setswhich accounted for 75 70 and 50of thewhole data setrespectively (see Supplementary Material 1 available onlineat httpsdoiorg10115520174649191) For 3D-QSAR studythe 100 inhibitors were randomly divided into a training set(77 molecules) and an independent test set (23 molecules)

25 Molecular Descriptor Calculation Molecular descriptorcan reflect physicochemical and geometric properties of thecompounds In this study forty-five molecular descriptorscalculated by the ChemOffice were applied to representcompounds [42] First three-dimensional structures of themolecules were optimized by MM+ force field with thePolak-Ribiere algorithm until the root-mean-square gradientbecame less than 01 Kcalmol Then quantum chemicalparameters were obtained for the most stable conformationof each molecule by using PM3 semiempirical molecularorbital method at the restricted Hartree-Fock level with noconfiguration interaction

26 Validation Methods for Prediction Results In this studytenfold cross-validation test and independent set test wereapplied to evaluate the prediction ability of the 2D-QSARmodel For the tenfold cross-validation test the data setwas divided into ten subsets Nine subsets were used as thetraining set and the left subset was predicted In turn eachsubset was omitted in order to be predicted and the correctrate was obtained from each trial The average of the correctrate from ten trials was used to estimate the accuracy of thealgorithm [43ndash45]

27 Prediction Measurement Sensitivity (SN) specificity(SP) overall accuracy (ACC) andMatthewrsquos correlation coef-ficient (MCC) were employed to evaluate the 2D predictionmodel The SN SP ACC and MCC can be represented as

SN = TP[TP + FN]

SP = TN[TN + FP]

ACC = [TP + TN][TP + TN + FP + FN]

MCC

=TP times TN minus FP times FN

radic(TN + FN) times (TN + FP) times (TP + FN) times (TP + FP)

(1)

TP TN FP and FN are true positives true negatives falsepositives and false negatives respectively

In the topomer CoMFA model 1199022 1199032 and MAE wereapplied to evaluate the model [46] The cut-off value of 1199022 is05 The MAE of the test set was less than 01 times training setrange and MAE + 3 times 120590 according to the MAE based criteriaThe optimized model was determined by the highest 1199022 andthe validity of the model depends on 1199032 value [47]

BioMed Research International 3

Table 1The results of prediction accuracy for different data sets containing 9 molecular descriptors using SVM classifier DS and EP presentdata set and evaluation parameters respectively

EP DSTrain set (75) Train set (70) Train set (50) Test set (30)

SN () 9722 9855 9194 9677SP () 9859 9923 9067 9818ACC () 9813 9899 9124 9767MCC 0958 0978 0824 0950

Table 2 Results from two topomer CoMFA model studies

Dataset Topomer CoMFA model 1 Topomer CoMFA model 2

Cutting model

1199022 0483 05651199032 0773 0888

28 Steric and Electrostatic Field Analysis Topomer CoMFAanalysis is an effective approach which has been applied indrug design for HIV central nervous system diseases andother tumors [48ndash50] In the topomer CoMFA model thereare two different ways to calculate the molecular field Oneway is to reduce the field contributions of fragmenting atomsthe other way is to calculate the steric and electrostatic fieldson a regularly spaced grid For detailed information see[51] Topomer CoMFA analysis is used to calculate the stericfield and electrostatic fields of R1 and R2 groups Steric andelectrostatic field analysismay help design novel EGFRdrugs

29 Molecular Docking SYBYL X-20 was used for molec-ular docking based on its Surflex-Dock module [52] Thecrystal structure of EGFR with the resolution of 26 A wasdownloaded from the Protein Data Bank (PDB ID 1M17)[53] Protein was prepared with protein structure preparationmodule of the SYBYL X-20 All the water molecules andligands were deleted and hydrogen atoms were added to thecrystal structure In addition positive and negative chargeswere added to N-terminal and C-terminal regions of theEGFR which became NH3+ and COOminus EGFR inhibitorswere minimized at physiological pH 70 with hydrogen atomsand charge by using Powell energy gradient method and theGasteiger-Huckel system

3 Results

31 Feature Selection and the 2D-QSAR Prediction Model Afeature subset containing nine molecular descriptors (DPLLHHFHOMOMR Pc TIndx VP andWIndx) was obtained

based on CFS combined with GS algorithms Sensitivityanalysis was applied to these nine descriptors to evaluate howthey affected the activity of EGFR inhibitors (see Figure 1)

Based on the optimal features subset the SVM classifiermethod was used to build the 2D-QSAR prediction modelAs a result the prediction accuracy of these models whosedata set accounted for 75 70 and 50 of the wholedata set was 9813 9899 and 9124 respectively bytenfold cross-validation test The sensitivity specificity andoverall accuracy of these three models were more than 90which indicated that changing the size of the training sethad a little impact on the quality of the 2D-SAR models(see Table 1) The model built via the data set accountingfor 70 of the whole data set was chosen finally due to itshigher prediction accuracy and smaller size Although theresult of the tenfold cross-validation test was well it was notgood enough for evaluating the classifier as the SVMclassifiermight be overfitted To validate the reliability of the classifieran independent test set was further employed in this studyAs a result the prediction accuracy of the independent settest was 9767

32 3D-QSAR Prediction Model The training set wasemployed to build the topomer CoMFA model by fragment-ing EGFR inhibitors into R1 and R2 groups Two topomerCoMFA models were generated by two cutting ways Thetopomer CoMFA model 2 with higher 1199022 and 1199032 values wasselected to analyze and predict EGFR inhibitorsrsquo activities(see Table 2)

The experimental and predicted activities of the trainingset and the independent test set were listed in Table 3 and

4 BioMed Research InternationalAc

tivity

DPLL

80

78

76

74

72

70

68

66

64

62

60

0 1 2 3 4 5 6 7 8

(a)

Activ

ity

H

90

85

80

75

70

65

60

55

50

0 5 10 15 20 25 30

(b)HF

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

minus800 minus600 minus400 minus200 0 200 400 600

(c)

HOMO

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

minus95 minus90 minus85 minus80 minus75 minus70

(d)MR

Activ

ity80

78

76

74

72

70

68

66

64

62

60

6 7 8 9 10 11 12 13 14 15 16

(e)Pc

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

0 10 20 30 40 50 60

(f)

TIndx

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

0 100 200 300 400 500 600

times102

(g)VP

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

00 02 04 06 08 10 12 14

(h)WIndx

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

0 10 20 30 40 50 60 70 80 90

times102

(i)

Figure 1 (a) Activity value versus DPLL (b) Activity value versus H (c) Activity value versus HF (d) Activity value versus HOMO (e)Activity value versus MR (f) Activity value versus Pc (g) Activity value versus TIndx (h) Activity value versus VP (i) Activity value versusWIndx

Figure 2 As a result the MAE and 1199032 of the training set were0308 and 0888 respectively The training set range was 732To estimate the reliability of model 2 the independent set testwas used to evaluate the model The MAE and 1199032 of the testset were 0526 and 0681 respectively The MAE of the testset was less than 0732 (01 times training set range) and 1903(MAE(training set) + 3 times 120590)

Additionally steric and electrostatic contour maps of R1and R2 groups were obtained Compound 33 was selectedto study how to redesign EGFR inhibitors due to the highlyactivity (see Figure 3) From Figure 3 it could be concludedthat large volume and positively charged groups were addedwhich can increase compound activity

33 Molecular Docking Compounds 27 28 30 31 32 and33 were used for molecular docking with EGFR As a

result these compounds have hydrogen bonds at Thr766and Met769 which were in ATP binding sites (see Figure 4)These compounds interact with EGFR kinase at binding sitesand the quinolone ring bound to the hydrophobic pocket ofEGFR instead of the purine ring of ATP

4 Discussion

41 2D-QSAR Model Feature selection via removal of someunnecessary features is required for a precise predictionmodel [25 54 55] A subset containing nine features wasobtained to build the 2D-QSAR prediction model Theprediction accuracy of the model was well for the trainingset and independent test set This result indicated that theoriginal data contained some redundant features and featureselection was a helpful step in building a prediction model


Table 3 Experimental and predicted PIC50 for topomer CoMFAmodel 2

Compound Exp PreTraining set

2 764 6624 624 625 604 6457 6 6168 8 81510 725 70511 611 66213 7 63115 609 60616 626 61417 753 80218 95 90620 839 82822 792 80123 832 75924 815 80525 792 82226 795 77827 916 86429 842 88730 818 8331 782 80332 76 72633 976 9634 901 80536 811 79437 774 74338 735 73140 801 85941 836 84642 745 77143 788 7745 66 63646 739 78447 8 7548 704 68750 688 68251 617 60853 574 63654 531 57255 607 72156 692 7457 739 6958 729 71460 69 71561 858 84763 616 585

Table 3 Continued

Compound Exp Pre64 602 63665 728 68666 648 65467 658 769 708 75770 882 83871 911 89772 902 89773 842 89675 853 9276 863 83577 642 69778 776 77879 836 83480 863 83981 619 6882 852 79783 805 80485 71 71686 75 75787 726 75288 604 60690 433 43591 466 46292 5 55294 719 71795 623 58997 414 39898 805 75499 697 679

Test set1 646 5583 757 7726 645 6169 725 64412 624 72414 521 58719 905 91421 707 74128 679 73335 746 72339 85 85344 74 83149 543 6452 527 63659 739 72562 863 841


Table 3 Continued

Compound Exp Pre68 788 78174 909 89884 672 76989 594 56793 717 66896 501 682100 62 618

10

9

8

7

6

5

4

10987654

Pred

icte

d ac

tivity

Experimental activity

Training set (－２ = 00427 N = 77)Test set (－２ = 00856 N = 23)

Figure 2 Scatterplot of experimental data versus predicted datafrom topomer CoMFA model 2

Although the accuracy of the prediction model with asubset containing nine features (DPLL H HF HOMO MRPc TIndx VP and WIndx) was reliable it was difficultto analyze the relationship between these descriptors andthe activity of EGFR inhibitors as the prediction modelis nonlinear Thus sensitivity was further applied for thisproblem [56] Figure 1(a) shows the relationship betweenthe Dipole length and activity When the Dipole length isapproximately 2 and 65 the activity levels are at minimumand maximum respectively Figure 1(b) shows the relation-ship between Henryrsquos law constant and activity The activityincreases alongwithHenryrsquos law constant from 0 to 30WhenHenryrsquos law constant is more than 30 the activity has arising trend Figure 1(c) shows the relationship between theHeat of Formation and activity When the Heat of Formationranges from minus700 to 600 the activity increases When theHeat of Formation is more than 600 the activity has a risingtrend Figure 1(d) shows the relationship between theHOMOenergy and activity When the HOMO energy ranges fromminus925 to minus825 the activity increases When the HOMOenergy is approximately minus825 the activity peaks When theHOMO energy is greater than minus825 the activity decreasesWhen the HOMO energy is more than minus725 the activityhas a decreasing trend Figure 1(e) shows the relationshipbetween the Molar refractivity and activity When the Molar

refractivity is approximately 10 and 14 the activity levelsare at minimum and maximum respectively Figure 1(f)shows the relationship between the critical pressure andactivity When the critical pressure ranges from 0 to 60the activity increases When the critical pressure is morethan 60 the activity has a rising trend Figure 1(g) showsthe relationship between the molecular topological indexand activity When the molecular topological index rangesfrom 0 to 60000 the activity decreases When the moleculartopological index is more than 60000 the activity has adecreasing trend Figure 1(h) shows the relationship betweenthe Vapor pressure and activity When the Vapor pressureranges from 0 to 14 the activity decreases When the Vaporpressure was more than 14 the activity had a decreasingtrend Figure 1(i) shows the relationship between the Wienerindex and activityWhen theWiener index and activity rangefrom 0 to 9000 the activity decreases When the Wienerindex is more than 9000 the activity has a decreasing trend

42 3D-QSAR Model Molecules in the topomer CoMFAmodels can be split into two three four and more groups asneeded [51 57] In this study compounds were divided intotwo groups (R1 and R2) EGFR inhibitorsrsquo activity was relatedto the completeness of the pharmacophore In topomerCoMFA models the pharmacophore is related to cutting[44 48 58] which plays an important role in the modelrsquospredictive performance of the model [58] In the topomerCoMFA analysis all molecules of the training set are cut intotwo fragments While the fragmentation was complete theinput structures were standardized and the topomers weregenerated They all shared the same identical substructure Ifthe same identical substructure was recognized by the test setthe modelrsquos predictive ability was promising

It could be found that model 2 added an 119873 elementin R1 based on model 1 which contributed to the modelrsquospredictive ability (see Table 2) Thus it is speculated that R1and R2 in model 2 are the same identical substructures Theindependent set test was used for evaluating model 2 (seeFigure 2) It was observed that the predicted pIC50 of somecompounds was poor such as compound 9 and compound34 (see Table 3) We guess this is because the same identicalsubstructures of the two compounds (see Figure 5) weredifferent from the other compounds The poor predictedpIC50 of compounds may cause high MAE According toRoy et alrsquos report [46] the 3D-QSAR model in our studywas reliable as the MAE of the external validation was bothless than 01 times training set range and MAE (training set) +3 times 120590 It is well known that the presence of systematic errorin predictions may easily be identified from the differencein mean error and mean absolute error It is importantto analyze prediction errors of compounds in test set inorder to search any possible systematic error In Roy etalrsquos study [59] various metrics including the number ofpositive prediction errors (NPE) the number of negativeprediction errors (NNP) the absolute value for average ofprediction errors (AE) the average of absolute predictionerrors (AAE) the mean of positive prediction errors (MPE)and the absolute value for mean of negative prediction errors


(a) (b)

(c) (d)

Figure 3 3D contour maps of topomer CoMFA model for R1 and R2 of compound 33 (a) and (c) present steric contour map (b) and (d)present electrostatic field map Green yellow blue and red represent large volume small volume positively charged and negatively chargedgroups respectively

(MNE) were employed to analyze the predictionrsquos error Ifprediction error is complied with principles IndashV defined byRoy the results were recommended In our study the NPENNPAEAAEMPE andMNEwere 12 11 0219 0526 0713and minus0321 respectively ABS (MPEMNE) and 1198772 (119884 versusresiduals) were 22 (threshold = 2) and 067 (threshold = 05)respectively Hence it was regarded that our 3D-QSARmodelis reliable

In addition topomer CoMFA model provides opinionson modifying EGFR inhibitors in order to design potentialhighly selective and highly active EGFR inhibitors Com-pound 33 (see Figure 5) was chosen to study the effect of R1and R2 groups on activity due to its high activity In R1 grouplarge group with a positive-charge in the yloxyethyl increasesthe compoundrsquos bioactivity (see Figure 3) In R2 group smallgroups with a positive-charge in the benzene ring may alsoincrease the compoundrsquos bioactivity

43 Molecular Docking Analysis Molecular docking wasapplied to predict the interaction sites between compoundsand EGFR As the structure of compound 33 is similarto erlotinib EGFR also interacts with compound 33 atThr766 and Met769 [50] Interestingly it is observed thatthe binding modes of compound 33-EGFR and erlotinib-EGFR were different despite the similar structure after calcu-lation Quinolone ring of erlotinib competitively binds to thehydrophobic pocket of EGFR kinase For erlotinib the anilinegroup reached into the pocket and substituent groups of site6 and site 7 were located outside of the hydrophobic pocketFor compound 33 it interacts with the EGFR by substituentgroups of site 6 and site 7 in the hydrophobic pocket Inthe steric and electrostatic fields large volume group andpositively charged group in site 6 and site 7 of compound33 may increase inhibitor activity (see Figure 3) Then thesimilar chemical series of compound 33was selected to study


(a) (b)

(c) (d)

(e) (f)

Figure 4The docking result of the EGFR inhibitors with EGFR (a)The binding site of compound 27 with EGFR isThr766 (b)The bindingsite of compound 28 with EGFR is Met769 (c) The binding site of compound 30 with EGFR is Met769 (d) The binding site of compound 31with EGFR is Met769 (e)The binding site of compound 32with EGFR is Met769 (f)The binding sites of compound 33with EGFR isThr766and Met769

the docking site As a result compounds 28 30 31 and 32interact with EGFR at Met769 and compound 27 interactswith EGFR at Thr766 Thus we considered that the Thr766and Met769 played a crucial role in the EGFR activity

Many studies performed the QSAR on kinase inhibitorsand the result was helpful for drugs design In Farghaly etalrsquos study [60] QSAR model was built and the RMSE and1199032 were applied to evaluate the model After calculatingthey selected out three predominant descriptors affectingthe anticancer activity and five anticancer agents werescreened finally Sharma showed the 2D-QSAR studies of c-Src tyrosine kinase inhibitors with 1199022 = 0755 and 1199032 =0832 [61] Sharma et al reported QSAR studies of Aurora

A kinase inhibitors [62] 1199022 is 0762 and 1199032 is 0806 Thedifference in the number of samples causes the differencein 1199022 and 1199032 When 1199022 and 1199032 are more than 05 and 08respectively the model has statistical significance In ourQSAR study 1199022 is 0565 lower than these two studies but 1199032is higher (see Table 4) In addition steric and electrostaticfield and molecular docking analysis were applied in ourstudy to explore the activity development and predict theinteraction between inhibitors and protein which is notshowed in these studies In conclusion QSAR combinedwith molecular docking provides better insight into thefuture design of more potent EGFR inhibitors prior tosynthesis


Compound 9 Compound 27 Compound 28 Compound 30

Compound 34Compound 33Compound 31 Compound 32

HN HNHN

HN

HNHNHNHN

F

N

N N N

NN

NN

N

N

NNNN

N N N

N

BrBr

Br

BrBrBr

H Cl

Cl

OO

O O O

22

（2（2

（2

Figure 5 Structures of compounds 9 27 28 30 31 32 33 and 34

Table 4The comparison of metrics between other studies and oursin QSAR study of the kinase inhibitors

MetricQSAR study

c-src tyrosine kinaseinhibitors [61]

Aurora inhibitors[62] Our study

1199022 0755 0762 05651199032 0832 0806 0888

5 Conclusion

In this study 2D-QSAR and 3D-QSAR prediction modelswere built to analyze EGFR inhibitors Firstly the 2D-QSARmodel was built to predict whether a compound was aninhibitor or a noninhibitor The accuracy of the 2D-QSARmodel using the tenfold cross-validation test and indepen-dent set test was 9899 and 9767 respectively Then thetopomer CoMFAmodel was built based on EGFR inhibitorsTwo models were obtained by cutting different molecularbonds As a result model 2 with higher 1199022 value and 1199032values was selected to predict EGFR inhibitors Finally aseries of similar chemical inhibitors were selected to study theinteracting sites between EGFR and EGFR inhibitors usingmolecular docking tool As a result Thr766 andMet769 werereceived by studying the docking result Thus we consideredthat Thr766 and Met769 played a crucial role in the EGFRactivity

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Authorsrsquo Contributions

Manman Zhao Lin Wang and Linfeng Zheng contributedequally to this work

Acknowledgments

The authors would like to express gratitude towards schol-arship from Natural Science Foundation of Shanghai Sci-ence and Technology Commission (no 17ZR1422500)the Shanghai Jiao Tong University Medical EngineeringCrossover Fund Project (no YG2016MS26) the ShanghaiUniversity High Performance Computing Shanghai Munic-ipal Education Commission the National Natural ScienceFoundation of China (81271384 81371623 31571171 and31100838) Shanghai Key Laboratory of Bio-Energy Crops(13DZ2272100) the Shanghai Natural Science Foundation(Grant no 15ZR1414900) the Key Laboratory of MedicalElectrophysiology (Southwest Medical University) of Min-istry of Education of China (Grant no 201502) and the YoungTeachers of Shanghai Universities Training Program Theyalso would like to thank Professor Mingyue Zheng from theState Laboratory of Drug Research of Chinese Academy ofScience for helping calculate the molecular descriptors

References

[1] F Ciardiello and G Tortora ldquoEGFR antagonists in cancertreatmentrdquo The New England Journal of Medicine vol 358 no11 pp 1160ndash1174 2008

[2] Y He B S Harrington and J D Hooper ldquoNew crossroadsfor potential therapeutic intervention in cancermdashintersectionsbetween CDCP1 EGFR family members and downstreamsignaling pathwaysrdquo Oncoscience vol 3 no 1 pp 5ndash8 2016

[3] R Wang X Wang J Q Wu et al ldquoEfficient porcine repro-ductive and respiratory syndrome virus entry in MARC-145 cells requires EGFR-PI3K-AKT-LIMK1-COFILIN signalingpathwayrdquo Virus Research vol 225 pp 23ndash32 2016

[4] F Imamura J Uchida Y Kukita et al ldquoMonitoring of treatmentresponses and clonal evolution of tumor cells by circulatingtumor DNA of heterogeneous mutant EGFR genes in lungcancerrdquo Lung Cancer vol 94 pp 68ndash73 2016

[5] K Wang D Li and L Sun ldquoHigh levels of EGFR expression intumor stroma are associated with aggressive clinical features in


epithelial ovarian cancerrdquo OncoTargets and Therapy vol 9 pp377ndash386 2016

[6] A Cho J Hur Y W Moon et al ldquoCorrelation between EGFRgene mutation cytologic tumor markers 18F-FDG uptake innon-small cell lung cancerrdquo BMC Cancer vol 16 no 1 article224 2016

[7] X B Holdman T Welte K Rajapakshe et al ldquoUpregulationof EGFR signaling is correlated with tumor stroma remodelingand tumor recurrence in FGFR1-driven breast cancerrdquo BreastCancer Research vol 17 no 1 article 141 2015

[8] C Sarkar ldquoEpidermal growth factor receptor (EGFR) geneamplification in high grade gliomasrdquo Neurology India vol 64no 1 pp 27-28 2016

[9] K Oda Y Matsuoka A Funahashi and H Kitano ldquoA com-prehensive pathway map of epidermal growth factor receptorsignalingrdquoMolecular Systems Biology vol 1 p E1 2005

[10] J C Baer A A Freeman E S Newlands A JWatson J A Raf-ferty and G P Margison ldquoDepletion of O6-alkylguanine-DNAalkyltransferase correlates with potentiation of temozolomideand CCNU toxicity in human tumour cellsrdquo British Journal ofCancer vol 67 no 6 pp 1299ndash1302 1993

[11] N Minkovsky and A Berezov ldquoBIBW-2992 a dual receptortyrosine kinase inhibitor for the treatment of solid tumorsrdquoCurrent Opinion in Investigational Drugs vol 9 no 12 pp 1336ndash1346 2008

[12] J C Dearden ldquoThe history and development of quantitativestructure-activity relationships (QSARs)rdquo International Journalof Quantitative Structure-Property Relationships vol 1 no 1 pp1ndash44 2016

[13] A Cherkasov E N Muratov D Fourches et al ldquoQSARmodeling where have you been Where are you going tordquoJournal of Medicinal Chemistry vol 57 no 12 pp 4977ndash50102014

[14] K Roy S Kar and R N DasUnderstanding the Basics of QSARfor Applications in Pharmaceutical Sciences and Risk AssessmentAcademic Press 2015

[15] D Gadaleta G F Mangiatordi M Catto A Carotti andO Nicolotti ldquoApplicability domain for QSAR models wheretheory meets realityrdquo International Journal of QuantitativeStructure-Property Relationships vol 1 no 1 pp 45ndash63 2016

[16] K Roy S Kar and P Ambure ldquoOn a simple approach for deter-mining applicability domain of QSAR modelsrdquo Chemometricsand Intelligent Laboratory Systems vol 145 pp 22ndash29 2015

[17] S Lee andM G Barron ldquoDevelopment of 3D-QSARmodel foracetylcholinesterase inhibitors using a combination of finger-print molecular docking and structure-based pharmacophoreapproachesrdquo Toxicological Sciences vol 148 no 1 pp 60ndash702015

[18] G Tresadern and D Bemporad ldquoModeling approaches forligand-based 3D similarityrdquo Future Medicinal Chemistry vol 2no 10 pp 1547ndash1561 2010

[19] K-C Chou ldquoStructural bioinformatics and its impact tobiomedical sciencerdquo Current Medicinal Chemistry vol 11 no16 pp 2105ndash2134 2004

[20] J-W Liang T-J Zhang Z-J Li Z-X Chen X-L Yan and F-H Meng ldquoPredicting potential antitumor targets of Aconitumalkaloids by molecular docking and proteinndashligand interactionfingerprintrdquo Medicinal Chemistry Research vol 25 no 6 pp1115ndash1124 2016

[21] L Blake and M E S Soliman ldquoIdentification of irreversibleprotein splicing inhibitors as potential anti-TB drugs insight

from hybrid non-covalentcovalent docking virtual screeningand molecular dynamics simulationsrdquo Medicinal ChemistryResearch vol 23 no 5 pp 2312ndash2323 2014

[22] P Ambure S Kar and K Roy ldquoPharmacophore mapping-based virtual screening followed by molecular docking studiesin search of potential acetylcholinesterase inhibitors as anti-Alzheimerrsquos agentsrdquo BioSystems vol 116 no 1 pp 10ndash20 2014

[23] J Lv J Su FWang Y Qi H Liu andY Zhang ldquoDetecting novelhypermethylated genes in Breast cancer benefiting from featureselectionrdquo Computers in Biology andMedicine vol 40 no 2 pp159ndash167 2010

[24] F R Ajdadi Y A Gilandeh K Mollazade and R P Hasan-zadeh ldquoApplication of machine vision for classification of soilaggregate sizerdquo Soil and Tillage Research vol 162 pp 8ndash17 2016

[25] R Sadeghia R Zarkami K Sabetraftar and P Van DammeldquoApplication of genetic algorithm and greedy stepwise to selectinput variables in classification tree models for the predictionof habitat requirements of Azolla filiculoides (Lam) in Anzaliwetland Iranrdquo Ecological Modelling vol 251 pp 44ndash53 2013

[26] C J C Burges ldquoA tutorial on support vector machines forpattern recognitionrdquo Data Mining and Knowledge Discoveryvol 2 no 2 pp 121ndash167 1998

[27] F Imani F E Boada F S Lieberman D K Davis and J MMountz ldquoMolecular and metabolic pattern classification fordetection of brain glioma progressionrdquo European Journal ofRadiology vol 83 no 2 pp e100ndashe105 2014

[28] K E Emblem F G Zoellner B Tennoe et al ldquoPredictivemodeling in glioma grading from MR perfusion images usingsupport vector machinesrdquoMagnetic Resonance in Medicine vol60 no 4 pp 945ndash952 2008

[29] R D Cramer ldquoTopomer CoMFA a design methodology forrapid lead optimizationrdquo Journal of Medicinal Chemistry vol46 no 3 pp 374ndash388 2003

[30] K Z Myint and X-Q Xie ldquoRecent advances in fragment-basedQSAR and multi-dimensional QSAR methodsrdquo InternationalJournal of Molecular Sciences vol 11 no 10 pp 3846ndash38662010

[31] C G Gadhe ldquoCoMFA vs Topomer CoMFA which one isbetter a case study with 5-lipoxygenase inhibitorsrdquo Journal ofthe Chosun Natural Science vol 4 no 2 pp 91ndash98 2011

[32] G W Rewcastle W A Denny A J Bridges et al ldquoTyrosinekinase inhibitors 5 Synthesis and structure-activity relation-ships for 4-[(phenylmethyl)amino]- and 4-(phenylamino)quin-azolines as potent adenosine 51015840-triphosphate binding site inhib-itors of the tyrosine kinase domain of the epidermal growthfactor receptorrdquo Journal of Medicinal Chemistry vol 38 no 18pp 3482ndash3487 1995

[33] A M Thompson A J Bridges D W Fry A J Krakerand W A Denny ldquoTyrosine kinase inhibitors 7 7-Amino-4-(phenylamino)- and 7-amino-4-[(phenylmethyl)amino]py-rido[43-d]pyrimidines a new class of inhibitors of the tyrosinekinase activity of the epidermal growth factor receptorrdquo Journalof Medicinal Chemistry vol 38 no 19 pp 3780ndash3788 1995

[34] G W Rewcastle A J Bridges D W Fry J R Rubin and W ADenny ldquoTyrosine kinase inhibitors 12 Synthesis and structure-activity relationships for 6-substituted 4-(phenylamino)pyr-imido[54-d]pyrimidines designed as inhibitors of the epider-mal growth factor receptorrdquo Journal ofMedicinal Chemistry vol40 no 12 pp 1820ndash1826 1997

[35] A M Thompson D K Murray W L Elliott et al ldquoTyrosinekinase inhibitors 13 Structuremdashactivity relationships for


soluble 7-substituted 4-[(3-bromophenyl)amino]pyrido[43-d]pyrimidines designed as inhibitors of the tyrosine kinaseactivity of the epidermal growth factor receptorrdquo Journal ofMedicinal Chemistry vol 40 no 24 pp 3915ndash3925 1997

[36] A J Bridges H Zhou D R Cody et al ldquoTyrosine kinaseinhibitors 8 An unusually steep structure-activity relation-ship for analogues of 4-(3-bromoanilino)-67-dimethoxyquin-azoline (PD 153035) a potent inhibitor of the epidermal growthfactor receptorrdquo Journal of Medicinal Chemistry vol 39 no 1pp 267ndash276 1996

[37] GW Rewcastle B D Palmer AMThompson et al ldquoTyrosinekinase inhibitors 10 Isomeric 4-[(3-bromophenyl)amino]py-rido[d]-pyrimidines are potent ATP binding site inhibitors ofthe tyrosine kinase function of the epidermal growth factorreceptorrdquo Journal of Medicinal Chemistry vol 39 no 9 pp1823ndash1835 1996

[38] S Li C Guo H Zhao Y Tang and M Lan ldquoSynthesis andbiological evaluation of 4-[3-chloro-4-(3-fluorobenzyloxy)anilino]-6-(3-substituted-phenoxy)pyrimidines as dualEGFRErbB-2 kinase inhibitorsrdquo Bioorganic and MedicinalChemistry vol 20 no 2 pp 877ndash885 2012

[39] A GWaterson K G Petrov K R Hornberger et al ldquoSynthesisand evaluation of aniline headgroups for alkynyl thienopy-rimidine dual EGFRErbB-2 kinase inhibitorsrdquo Bioorganic andMedicinal Chemistry Letters vol 19 no 5 pp 1332ndash1336 2009

[40] N Suzuki T Shiota F Watanabe et al ldquoSynthesis and evalu-ation of novel pyrimidine-based dual EGFRHer-2 inhibitorsrdquoBioorganic and Medicinal Chemistry Letters vol 21 no 6 pp1601ndash1606 2011

[41] N Suzuki T Shiota F Watanabe et al ldquoDiscovery of novel5-alkynyl-4-anilinopyrimidines as potent orally active dualinhibitors of EGFR and Her-2 tyrosine kinasesrdquo Bioorganic andMedicinal Chemistry Letters vol 22 no 1 pp 456ndash460 2012

[42] J J Irwin ldquoSoftware review ChemOffice 2005 Pro by Cam-bridgesoftrdquo Journal of Chemical Information and Modeling vol45 no 5 pp 1468-1469 2005

[43] S Varma and R Simon ldquoBias in error estimation when usingcross-validation for model selectionrdquo BMC Bioinformatics vol7 supplement 5 pp 91ndash98 2006

[44] L Wang H Shen B Li and D Hu ldquoClassification ofschizophrenic patients and healthy controls using multiplespatially independent components of structural MRI datardquoFrontiers of Electrical and Electronic Engineering in China vol6 no 2 pp 353ndash362 2011

[45] Y P Zhang N Sussman G Klopman and H S RosenkranzldquoDevelopment of methods to ascertain the predictivity andconsistency of SAR models application to the US Nationaltoxicology program rodent carcinogenicity bioassaysrdquo Quan-titative Structure-Activity Relationships vol 16 no 4 pp 290ndash295 1997

[46] K Roy R N Das P Ambure and R B Aher ldquoBe aware of errormeasures Further studies on validation of predictive QSARmodelsrdquo Chemometrics and Intelligent Laboratory Systems vol152 pp 18ndash33 2016

[47] S Yu J Yuan J Shi et al ldquoHQSAR and topomer CoMFA forpredicting melanocortin-4 receptor binding affinities of trans-4-(4-chlorophenyl) pyrrolidine-3-carboxamidesrdquo Chemomet-rics and Intelligent Laboratory Systems vol 146 pp 34ndash41 2015

[48] S Kumar and M Tiwari ldquoTopomer-CoMFA-based predic-tive modelling on 23-diaryl-substituted-13-thiazolidin-4-onesas non-nucleoside reverse transcriptase inhibitorsrdquo MedicinalChemistry Research vol 24 no 1 pp 245ndash257 2015

[49] Y Tian Y Shen X Zhang et al ldquoDesign some new type-I c-met inhibitors based onmolecular docking and topomer comfaresearchrdquo Molecular Informatics vol 33 no 8 pp 536ndash5432014

[50] G Tresadern J-M Cid and A A Trabanco ldquoQSAR designof triazolopyridine mGlu2 receptor positive allosteric modula-torsrdquo Journal of Molecular Graphics and Modelling vol 53 pp82ndash91 2014

[51] H Tang L Yang J Li and J Chen ldquoMolecular modellingstudies of 35-dipyridyl-124-triazole derivatives as xanthineoxidoreductase inhibitors using 3D-QSAR Topomer CoMFAmolecular docking and molecular dynamic simulationsrdquo Jour-nal of the Taiwan Institute of Chemical Engineers vol 68 pp64ndash73 2016

[52] S D Joshi U A More D Koli M S Kulkarni M NNadagouda and T M Aminabhavi ldquoSynthesis evaluationand in silico molecular modeling of pyrroyl-134-thiadiazoleinhibitors of InhArdquo Bioorganic Chemistry vol 59 pp 151ndash1672015

[53] J Stamos M X Sliwkowski and C Eigenbrot ldquoStructure ofthe epidermal growth factor receptor kinase domain alone andin complex with a 4-anilinoquinazoline inhibitorrdquo Journal ofBiological Chemistry vol 277 no 48 pp 46265ndash46272 2002

[54] J Miao and L Niu ldquoA survey on feature selectionrdquo ProcediaComputer Science vol 91 pp 919ndash926 2016

[55] V Bolon-Canedo N Sanchez-Marono and A Alonso-Betanzos ldquoFeature selection for high-dimensional datardquoProgress in Artificial Intelligence vol 5 no 2 pp 65ndash75 2016

[56] B Niu Q Su X Yuan W Lu and J Ding ldquoQSAR study on5-lipoxygenase inhibitors based on support vector machinerdquoMedicinal Chemistry vol 8 no 6 pp 1108ndash1116 2012

[57] W Ding M Sun S Luo et al ldquoA 3D QSAR study ofbetulinic acid derivatives as anti-tumor agents using topomerCoMFAmodel building studies and experimental verificationrdquoMolecules vol 18 no 9 pp 10228ndash10241 2013

[58] Y Xiang J Song and Z Zhang ldquoTopomer CoMFA andvirtual screening studies of azaindole class renin inhibitorsrdquoCombinatorial Chemistry and High Throughput Screening vol17 no 5 pp 458ndash472 2014

[59] K Roy P Ambure and R B Aher ldquoHow important is todetect systematic error in predictions and understand statisticalapplicability domain of QSAR modelsrdquo Chemometrics andIntelligent Laboratory Systems vol 162 pp 44ndash54 2017

[60] T A Farghaly H M E Hassaneen and H S A Elzahabi ldquoEco-friendly synthesis and 2D-QSAR study of novel pyrazolinesas potential anticolon cancer agentsrdquo Medicinal ChemistryResearch vol 24 no 2 pp 652ndash668 2015

[61] M C Sharma ldquo2D QSAR studies of the inhibitory activity ofa series of substituted purine derivatives against c-Src tyrosinekinaserdquo Journal of Taibah University for Science vol 10 no 4pp 563ndash570 2016

[62] M C Sharma S Sharma and K Bhadoriya ldquoQSAR studieson pyrazole-4-carboxamide derivatives as Aurora A kinaseinhibitorsrdquo Journal of Taibah University for Science vol 10 no1 pp 107ndash114 2016

Submit your manuscripts athttpswwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of


Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 201


Molecular Biology International

GenomicsInternational Journal of


The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


BioinformaticsAdvances in

Marine BiologyJournal of



Signal TransductionJournal of


BioMed Research International

Evolutionary BiologyInternational Journal of



Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Genetics Research International


Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational



Enzyme Research



Microbiology


the appearance of three-dimensional quantitative structure-activity relationship (3D-QSAR) In the 3D-QSAR study thecorrelation between 3D steric and electrostatic fields andbiologically activity draws attention For the molecular fieldstudy CoMFA was widely used preliminarily However thetime-consuming limit stimulates the advent of TopCoMFATopCoMFA overcomes the weakness and uses an objectivemethod to fragment and align the molecules In additionthe fragmentation process is automated except for somespecific bonds that should be cleaved manually Of courseTopCoMFA and CoMFA also have similarity that they bothshare QSAR PLS analysis The details about TopCoMFA andCoMFA are in [18]

Drug development is a long process and it requiresa vast amount of material and financial resources QSARand molecular docking technology have been extensivelyemployed in drug virtual screening and potential moleculartargets prediction which may shorten the cycle of the drugdevelopment [19ndash22] In this work 2D-QSAR model wasemployed to determine EGFR inhibitor and the 3D-QSARmodel was used to predict the activity Finally moleculardocking was applied to investigate the binding sites

2 Materials and Methods

21 CfsSubsetEval Method and Greedy Stepwise Algorithm Adata set containing 119899 vectors has 2119899 possible combinations offeatures for the subset A useful subset which can correctlypredict other compounds is one of 2119899 combinationsThe bestway to find an optimal subset is to try all the possible featurecombinations However this strategy is difficult to carry outdue to the huge computation In this study the CfsSubsetEval(CFS) search method combined with Greedy Stepwise (GS)algorithmwas employed to search the optimal feature subsetThemain idea of the GS algorithms is to make the best choicewhen selecting good features The CFS method was used toevaluate the attribute Thus the CFS method combined withthe GS algorithm was employed to select the optimal subsetfrom these 2119899 combinations Additional details about the CFSmethod and the GS algorithm could be found in [23ndash25]

22 SVM Support vector machine (SVM) a supervisedlearning algorithm is usually used for pattern recognitionclassification [26] SVM was employed for the classificationand sensitivity analysis in our study due to its high perfor-mance in many studies [25 27 28]

23 Topomer CoMFA Topomer CoMFA possessing boththe topomer technique and CoMFA technology can over-come the alignment problem of CoMFA [18 29] Partialleast squares (PLS) regression is employed to build thetopomer CoMFAmodel and the leave-one-out (LOO) cross-validation is used to evaluate the model Additional detailsabout the topomer CoMFA can be found in [29ndash31]

24 Data Preparation 100 inhibitors derived from the lit-erature and 185 noninhibitors downloaded from the DUDdatabase (httpduddockingorg) were collected [32ndash41] For

2D-QSAR study the data set containing inhibitors andnoninhibitors was randomly divided into three training setswhich accounted for 75 70 and 50of thewhole data setrespectively (see Supplementary Material 1 available onlineat httpsdoiorg10115520174649191) For 3D-QSAR studythe 100 inhibitors were randomly divided into a training set(77 molecules) and an independent test set (23 molecules)

25 Molecular Descriptor Calculation Molecular descriptorcan reflect physicochemical and geometric properties of thecompounds In this study forty-five molecular descriptorscalculated by the ChemOffice were applied to representcompounds [42] First three-dimensional structures of themolecules were optimized by MM+ force field with thePolak-Ribiere algorithm until the root-mean-square gradientbecame less than 01 Kcalmol Then quantum chemicalparameters were obtained for the most stable conformationof each molecule by using PM3 semiempirical molecularorbital method at the restricted Hartree-Fock level with noconfiguration interaction

26 Validation Methods for Prediction Results In this studytenfold cross-validation test and independent set test wereapplied to evaluate the prediction ability of the 2D-QSARmodel For the tenfold cross-validation test the data setwas divided into ten subsets Nine subsets were used as thetraining set and the left subset was predicted In turn eachsubset was omitted in order to be predicted and the correctrate was obtained from each trial The average of the correctrate from ten trials was used to estimate the accuracy of thealgorithm [43ndash45]

27 Prediction Measurement Sensitivity (SN) specificity(SP) overall accuracy (ACC) andMatthewrsquos correlation coef-ficient (MCC) were employed to evaluate the 2D predictionmodel The SN SP ACC and MCC can be represented as

SN = TP[TP + FN]

SP = TN[TN + FP]

ACC = [TP + TN][TP + TN + FP + FN]

MCC

=TP times TN minus FP times FN

radic(TN + FN) times (TN + FP) times (TP + FN) times (TP + FP)

(1)

TP TN FP and FN are true positives true negatives falsepositives and false negatives respectively

In the topomer CoMFA model 1199022 1199032 and MAE wereapplied to evaluate the model [46] The cut-off value of 1199022 is05 The MAE of the test set was less than 01 times training setrange and MAE + 3 times 120590 according to the MAE based criteriaThe optimized model was determined by the highest 1199022 andthe validity of the model depends on 1199032 value [47]




SN () 9722 9855 9194 9677SP () 9859 9923 9067 9818ACC () 9813 9899 9124 9767MCC 0958 0978 0824 0950



Cutting model

1199022 0483 05651199032 0773 0888



3 Results







tivity

DPLL

80

78

76

74

72

70

68

66

64

62

60

0 1 2 3 4 5 6 7 8

(a)

Activ

ity

H

90

85

80

75

70

65

60

55

50

0 5 10 15 20 25 30

(b)HF

Activ

ity

80

78

76

74

72

70

68

66

64

62

60


(c)

HOMO

Activ

ity

80

78

76

74

72

70

68

66

64

62

60


(d)MR

Activ

ity80

78

76

74

72

70

68

66

64

62

60

6 7 8 9 10 11 12 13 14 15 16

(e)Pc

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

0 10 20 30 40 50 60

(f)

TIndx

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

0 100 200 300 400 500 600

times102

(g)VP

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

00 02 04 06 08 10 12 14

(h)WIndx

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

0 10 20 30 40 50 60 70 80 90

times102

(i)






4 Discussion





2 764 6624 624 625 604 6457 6 6168 8 81510 725 70511 611 66213 7 63115 609 60616 626 61417 753 80218 95 90620 839 82822 792 80123 832 75924 815 80525 792 82226 795 77827 916 86429 842 88730 818 8331 782 80332 76 72633 976 9634 901 80536 811 79437 774 74338 735 73140 801 85941 836 84642 745 77143 788 7745 66 63646 739 78447 8 7548 704 68750 688 68251 617 60853 574 63654 531 57255 607 72156 692 7457 739 6958 729 71460 69 71561 858 84763 616 585

Table 3 Continued

Compound Exp Pre64 602 63665 728 68666 648 65467 658 769 708 75770 882 83871 911 89772 902 89773 842 89675 853 9276 863 83577 642 69778 776 77879 836 83480 863 83981 619 6882 852 79783 805 80485 71 71686 75 75787 726 75288 604 60690 433 43591 466 46292 5 55294 719 71795 623 58997 414 39898 805 75499 697 679

Test set1 646 5583 757 7726 645 6169 725 64412 624 72414 521 58719 905 91421 707 74128 679 73335 746 72339 85 85344 74 83149 543 6452 527 63659 739 72562 863 841


Table 3 Continued

Compound Exp Pre68 788 78174 909 89884 672 76989 594 56793 717 66896 501 682100 62 618

10

9

8

7

6

5

4

10987654

Pred

icte

d ac

tivity









(a) (b)

(c) (d)






(a) (b)

(c) (d)

(e) (f)








HN HNHN

HN

HNHNHNHN

F

N

N N N

NN

NN

N

N

NNNN

N N N

N

BrBr

Br

BrBrBr

H Cl

Cl

OO

O O O

22

（2（2

（2



MetricQSAR study



1199022 0755 0762 05651199032 0832 0806 0888

5 Conclusion






Acknowledgments


References











































































Volume 201






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology




SN () 9722 9855 9194 9677SP () 9859 9923 9067 9818ACC () 9813 9899 9124 9767MCC 0958 0978 0824 0950



Cutting model

1199022 0483 05651199032 0773 0888



3 Results







tivity

DPLL

80

78

76

74

72

70

68

66

64

62

60

0 1 2 3 4 5 6 7 8

(a)

Activ

ity

H

90

85

80

75

70

65

60

55

50

0 5 10 15 20 25 30

(b)HF

Activ

ity

80

78

76

74

72

70

68

66

64

62

60


(c)

HOMO

Activ

ity

80

78

76

74

72

70

68

66

64

62

60


(d)MR

Activ

ity80

78

76

74

72

70

68

66

64

62

60

6 7 8 9 10 11 12 13 14 15 16

(e)Pc

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

0 10 20 30 40 50 60

(f)

TIndx

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

0 100 200 300 400 500 600

times102

(g)VP

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

00 02 04 06 08 10 12 14

(h)WIndx

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

0 10 20 30 40 50 60 70 80 90

times102

(i)






4 Discussion





2 764 6624 624 625 604 6457 6 6168 8 81510 725 70511 611 66213 7 63115 609 60616 626 61417 753 80218 95 90620 839 82822 792 80123 832 75924 815 80525 792 82226 795 77827 916 86429 842 88730 818 8331 782 80332 76 72633 976 9634 901 80536 811 79437 774 74338 735 73140 801 85941 836 84642 745 77143 788 7745 66 63646 739 78447 8 7548 704 68750 688 68251 617 60853 574 63654 531 57255 607 72156 692 7457 739 6958 729 71460 69 71561 858 84763 616 585

Table 3 Continued

Compound Exp Pre64 602 63665 728 68666 648 65467 658 769 708 75770 882 83871 911 89772 902 89773 842 89675 853 9276 863 83577 642 69778 776 77879 836 83480 863 83981 619 6882 852 79783 805 80485 71 71686 75 75787 726 75288 604 60690 433 43591 466 46292 5 55294 719 71795 623 58997 414 39898 805 75499 697 679

Test set1 646 5583 757 7726 645 6169 725 64412 624 72414 521 58719 905 91421 707 74128 679 73335 746 72339 85 85344 74 83149 543 6452 527 63659 739 72562 863 841


Table 3 Continued

Compound Exp Pre68 788 78174 909 89884 672 76989 594 56793 717 66896 501 682100 62 618

10

9

8

7

6

5

4

10987654

Pred

icte

d ac

tivity









(a) (b)

(c) (d)






(a) (b)

(c) (d)

(e) (f)








HN HNHN

HN

HNHNHNHN

F

N

N N N

NN

NN

N

N

NNNN

N N N

N

BrBr

Br

BrBrBr

H Cl

Cl

OO

O O O

22

（2（2

（2



MetricQSAR study



1199022 0755 0762 05651199032 0832 0806 0888

5 Conclusion






Acknowledgments


References











































































Volume 201






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology


tivity

DPLL

80

78

76

74

72

70

68

66

64

62

60

0 1 2 3 4 5 6 7 8

(a)

Activ

ity

H

90

85

80

75

70

65

60

55

50

0 5 10 15 20 25 30

(b)HF

Activ

ity

80

78

76

74

72

70

68

66

64

62

60


(c)

HOMO

Activ

ity

80

78

76

74

72

70

68

66

64

62

60


(d)MR

Activ

ity80

78

76

74

72

70

68

66

64

62

60

6 7 8 9 10 11 12 13 14 15 16

(e)Pc

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

0 10 20 30 40 50 60

(f)

TIndx

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

0 100 200 300 400 500 600

times102

(g)VP

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

00 02 04 06 08 10 12 14

(h)WIndx

Activ

ity

80

78

76

74

72

70

68

66

64

62

60

0 10 20 30 40 50 60 70 80 90

times102

(i)






4 Discussion





2 764 6624 624 625 604 6457 6 6168 8 81510 725 70511 611 66213 7 63115 609 60616 626 61417 753 80218 95 90620 839 82822 792 80123 832 75924 815 80525 792 82226 795 77827 916 86429 842 88730 818 8331 782 80332 76 72633 976 9634 901 80536 811 79437 774 74338 735 73140 801 85941 836 84642 745 77143 788 7745 66 63646 739 78447 8 7548 704 68750 688 68251 617 60853 574 63654 531 57255 607 72156 692 7457 739 6958 729 71460 69 71561 858 84763 616 585

Table 3 Continued

Compound Exp Pre64 602 63665 728 68666 648 65467 658 769 708 75770 882 83871 911 89772 902 89773 842 89675 853 9276 863 83577 642 69778 776 77879 836 83480 863 83981 619 6882 852 79783 805 80485 71 71686 75 75787 726 75288 604 60690 433 43591 466 46292 5 55294 719 71795 623 58997 414 39898 805 75499 697 679

Test set1 646 5583 757 7726 645 6169 725 64412 624 72414 521 58719 905 91421 707 74128 679 73335 746 72339 85 85344 74 83149 543 6452 527 63659 739 72562 863 841


Table 3 Continued

Compound Exp Pre68 788 78174 909 89884 672 76989 594 56793 717 66896 501 682100 62 618

10

9

8

7

6

5

4

10987654

Pred

icte

d ac

tivity









(a) (b)

(c) (d)






(a) (b)

(c) (d)

(e) (f)








HN HNHN

HN

HNHNHNHN

F

N

N N N

NN

NN

N

N

NNNN

N N N

N

BrBr

Br

BrBrBr

H Cl

Cl

OO

O O O

22

（2（2

（2



MetricQSAR study



1199022 0755 0762 05651199032 0832 0806 0888

5 Conclusion






Acknowledgments


References











































































Volume 201






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology




2 764 6624 624 625 604 6457 6 6168 8 81510 725 70511 611 66213 7 63115 609 60616 626 61417 753 80218 95 90620 839 82822 792 80123 832 75924 815 80525 792 82226 795 77827 916 86429 842 88730 818 8331 782 80332 76 72633 976 9634 901 80536 811 79437 774 74338 735 73140 801 85941 836 84642 745 77143 788 7745 66 63646 739 78447 8 7548 704 68750 688 68251 617 60853 574 63654 531 57255 607 72156 692 7457 739 6958 729 71460 69 71561 858 84763 616 585

Table 3 Continued

Compound Exp Pre64 602 63665 728 68666 648 65467 658 769 708 75770 882 83871 911 89772 902 89773 842 89675 853 9276 863 83577 642 69778 776 77879 836 83480 863 83981 619 6882 852 79783 805 80485 71 71686 75 75787 726 75288 604 60690 433 43591 466 46292 5 55294 719 71795 623 58997 414 39898 805 75499 697 679

Test set1 646 5583 757 7726 645 6169 725 64412 624 72414 521 58719 905 91421 707 74128 679 73335 746 72339 85 85344 74 83149 543 6452 527 63659 739 72562 863 841


Table 3 Continued

Compound Exp Pre68 788 78174 909 89884 672 76989 594 56793 717 66896 501 682100 62 618

10

9

8

7

6

5

4

10987654

Pred

icte

d ac

tivity









(a) (b)

(c) (d)






(a) (b)

(c) (d)

(e) (f)








HN HNHN

HN

HNHNHNHN

F

N

N N N

NN

NN

N

N

NNNN

N N N

N

BrBr

Br

BrBrBr

H Cl

Cl

OO

O O O

22

（2（2

（2



MetricQSAR study



1199022 0755 0762 05651199032 0832 0806 0888

5 Conclusion






Acknowledgments


References











































































Volume 201






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology


Table 3 Continued

Compound Exp Pre68 788 78174 909 89884 672 76989 594 56793 717 66896 501 682100 62 618

10

9

8

7

6

5

4

10987654

Pred

icte

d ac

tivity









(a) (b)

(c) (d)






(a) (b)

(c) (d)

(e) (f)








HN HNHN

HN

HNHNHNHN

F

N

N N N

NN

NN

N

N

NNNN

N N N

N

BrBr

Br

BrBrBr

H Cl

Cl

OO

O O O

22

（2（2

（2



MetricQSAR study



1199022 0755 0762 05651199032 0832 0806 0888

5 Conclusion






Acknowledgments


References











































































Volume 201






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology


(a) (b)

(c) (d)






(a) (b)

(c) (d)

(e) (f)








HN HNHN

HN

HNHNHNHN

F

N

N N N

NN

NN

N

N

NNNN

N N N

N

BrBr

Br

BrBrBr

H Cl

Cl

OO

O O O

22

（2（2

（2



MetricQSAR study



1199022 0755 0762 05651199032 0832 0806 0888

5 Conclusion






Acknowledgments


References











































































Volume 201






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology


(a) (b)

(c) (d)

(e) (f)








HN HNHN

HN

HNHNHNHN

F

N

N N N

NN

NN

N

N

NNNN

N N N

N

BrBr

Br

BrBrBr

H Cl

Cl

OO

O O O

22

（2（2

（2



MetricQSAR study



1199022 0755 0762 05651199032 0832 0806 0888

5 Conclusion






Acknowledgments


References











































































Volume 201






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology




HN HNHN

HN

HNHNHNHN

F

N

N N N

NN

NN

N

N

NNNN

N N N

N

BrBr

Br

BrBrBr

H Cl

Cl

OO

O O O

22

（2（2

（2



MetricQSAR study



1199022 0755 0762 05651199032 0832 0806 0888

5 Conclusion






Acknowledgments


References











































































Volume 201






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology






































































Volume 201






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology

ResearchArticle 2D-QSAR and 3D-QSAR Analyses for EGFR...

Documents

Transcript of ResearchArticle 2D-QSAR and 3D-QSAR Analyses for EGFR...