Abstract
A theoretical study on the binding conformations and the quantitative structure–activity relationship (QSAR) of combretastatin A4 (CA-4) analogs as inhibitors toward tubulin has been carried out using docking analysis and comparative molecular field analysis (CoMFA). The appropriate binding orientations and conformations of these compounds interacting with tubulin were revealed by the docking study; and a 3D-QSAR model showing significant statistical quality and satisfactory predictive ability was established, in which the correlation coefficient (R2) and cross-validation coefficient (q2) were 0.955 and 0.66, respectively. The same model was further applied to predict the pIC50 values for 16 congeneric compounds as external test set, and the predictive correlation coefficient R2pred reached 0.883. Other tests on additional validations further confirmed the satisfactory predictive power of the model. In this work, it was very interesting to find that the 3D topology structure of the active site of tubulin from the docking analysis was in good agreement with the 3D-QSAR model from CoMFA for this series of compounds. Some key structural factors of the compounds responsible for cytotoxicity were reasonably presented. These theoretical results can offer useful references for understanding the action mechanism and directing the molecular design of this kind of inhibitor with improved activity.
Introduction
Microtubules, formed by polymerization of α- and β-tubulin heterodimers, are important for the formation of the mitotic spindle during the process of mitosisCitation1. Interference with the dynamic equilibrium between the tubulin and microtubules leads to an arrest of cell division and eventually to apoptosisCitation2. The success of tubulin polymerization inhibitors as antitumor agents has stimulated significant interest in the development of new compounds with more potent cytotoxicities.
Combretastatin A4 (CA-4), a low molecular weight natural product, strongly inhibits the polymerization of tubulin by binding to the colchicine site. A water soluble phosphate prodrug of combretastatin A4 (CA-4P) is currently in clinical trials for the treatment of solid tumors (see )Citation3. Because of its simple structure and potent cytotoxicity, a great number of CA-4 analogs have been developed. Recently, Romagnoli and coworkers synthesized a series of analogs of CA-4 and assessed their cytotoxicitiesCitation4–6. They discovered that these compounds have potent growth inhibition on several human cancer cell lines, demonstrating the great potential of developing these compounds as a novel class of drug for cancer therapy. Although some progress has been made in experimental research, so far the theoretical studies on the mechanism of these compounds toward tubulin as well as the structural factors responsible for the cytotoxicity remain few.
Figure 1. Structural diagrams of CA-4 and CA-4P
The quantitative structure–activity relationship (QSAR), which quantitatively correlates the variations in biological activity with the properties or molecular structures, is one of the most effective approaches for understanding the action mechanism of drugs and in designing new drugsCitation7–9. Nowadays, comparative molecular field analysis (CoMFA) is a useful technique in understanding the pharmacological properties of studied compounds, because not only is the CoMFA model visualized, but also the obtained steric and electrostatic maps may help in understanding the detailed 3D structure of the active site of the receptorCitation10–12. Docking analysis is also a useful methodology to further study the action mechanism, since it can offer a vivid interaction picture between a ligand and a receptorCitation13–15.
In this work, 32 CA-4 analogs as tubulin inhibitors with potential cytotoxicity against the human T-lymphocyte Molt4/C8 cell line were selected for performance of a theoretical study using the docking and CoMFA methods. The orientations and conformations of these compounds as inhibitors interacting with tubulin were located, a 3D-QSAR model was established, and some key structural factors responsible for inhibition toward tubulin were revealed. We expect the obtained theoretical results to offer some useful references for experimental work.
Computational methods
Studied compounds and their biological activity data
The general structural diagram of 32 studied compounds, CA-4 analogs with well-expressed cytotoxicity, is shown in . We used the literature data for IC50, defined as the value of the necessary molar concentration of compound to cause 50% growth inhibition against the human T-lymphocyte Molt4/C8 cell lineCitation4–6. The corresponding values are listed in . All original IC50 values were converted to the negative logarithm of IC50 (pIC50), used as a dependent variable in the CoMFA study.
Table 1. Structures and experimental cytotoxicities (against the human T-lymphocyte Molt4/C8 cell line) of studied CA-4 analogs.
Figure 2. Molecular structures and numbering of studied CA-4 analogs.
Molecular docking
To locate the appropriate binding orientations and conformations of these CA-4 analogs interacting with tubulin, a docking study for all studied compounds was performed with the DOCK 6.0 program. All parameters used in docking were default except where explained.
The X-ray crystal structure of tubulin taken from the Protein Data Bank (pdb Id: 1SA0) was used to dock. At the beginning of docking, all the water molecules and subunits were removed, and hydrogen atoms and AMBER7FF99 charges were added to the protein. Next, only hydrogen positions were energy-minimized in 10,000 cycles using the Powell method in SYBYL 6.9Citation16. Then the 3D structures of all studied analogs of CA-4 were constructed in Chem3D software; energy minimizations were performed using the semi-empirical quantum-chemical PM3 method, and partial atomic charges were calculated by the Gasteiger–Hückel method. At last, all compounds were rigidly docked into the binding sites where the protein was considered as rigid. The box size, grid space, energy cutoff distance, and maximum orientation were set as 8 å, 0.3 å, 9999 å, and 100,000, respectively.
Molecular modeling
The CoMFA study was performed using SYBYL 6.9 molecular modeling software running on an SGI R2400 workstation. All parameters used in CoMFA were default except where explained.
Active conformation selection is a key step for CoMFA analysis. Since the crystal structure of the complex of tubulin with one of these compounds is not available, molecular docking was used to simulate the active conformation, and the docked conformation of the most active compound 12 was used as the template to construct the structures of the remaining compounds and was then further optimized by the molecular mechanics MM2 method in Chem3D software. The rest of the molecules were minimized in the same way.
Alignment
Structural alignment is considered one of the most sensitive parameters in CoMFA analysis. The accuracy of the prediction of the CoMFA model and reliability of the contour maps are directly dependent on the structural alignment ruleCitation17. The most active compound 12 was used as a template for superimposition, and the common fragment (i.e. atoms numbered from 2 to 21) was selected for alignment and all the molecules were aligned on it. The aligned compounds are shown in .
Figure 3. Alignment of the 32 studied molecules.
Generation of CoMFA field
Models of steric and electrostatic fields for CoMFA were based on both Lennard-Jones and Coulombic potentialsCitation18. Steric and electrostatic energies were calculated using an spCitation3 carbon atom with van der Waals radius of 1.52 å, charge of +1.0, and grid spacing of 2.0 å. The CoMFA cutoff values were set to 30 kcal/mol for both steric and electrostatic fields.
Partial least squares analysis
Partial least squares (PLS) analysis was used to construct a linear correlation between the 3D fields (independent variables) and the activity values (dependent variables). To select the best model, the optimum number of components was determined by SAMPLSCitation19 analysis implemented in SYBYL 6.9 with a leave-one-out (LOO) cross-validation in which one compound was removed from the data set and its activity was predicted using the model built from rest of the data setCitation20. This educes the cross-validation correlation coefficient (q2) and the optimum number of components N. Non-cross-validation was performed with a column filter value of 2.0 to speed up the analysis and reduce the noise. To further assess the robustness and statistical validity of the obtained models, bootstrapping analysis for 100 runs was performed.
Results and discussion
Validation of docking reliability
It is well known that combretastatin A4 (CA-4) can strongly inhibit the polymerization of tubulin by binding to the colchicine site. These studied compounds have similar structures to CA-4 and are well reported as tubulin inhibitorsCitation4–6, so we can reasonably assume that these compounds exhibit the same activity site with CA-4 as tubulin inhibitor based on the experiments and references, although we do not directly identify the active site in docking analysis. The docking study may offer more insight into understanding the protein–inhibitor interactions and the structural features of the active sites.
First of all, it is necessary to validate the docking reliability. We adopted the known X-ray structure of tubulin in complex with the molecular ligand CN2 (2-mercapto-N-[1,2,3,10-tetramethoxy-9-oxo-5,6,7,9-tetrahydro-benzo[A]heptalen- 7-yl] acetamide) to perform this validation. The ligand CN2 was flexibly docked to the binding site of tubulin and the docking conformation corresponding to the lowest energy score was selected as the most probable binding conformation. As a result, the docked CN2 and crystal CN2 are almost located on the same position in the active site of tubulin (see ) with an acceptable root-mean-square deviation (RMSD = 1.52 å) between the conformation of crystal CN2 and docked CN2, suggesting an acceptable docking reliability of DOCK 6.0. Therefore, the DOCK 6.0 docking protocol and the used parameters could be extended to search the tubulin-binding conformations of other inhibitors.
Figure 4. Binding conformations of the docked CN2 (cyan) and crystal CN2 (magenta) at the active site of tubulin.
Docking results
All 32 studied inhibitors were docked into the binding site of tubulin and the energy scores (E) of the inhibitors are shown in , where no precise correlations could be found between docking scores and pIC50 values. This observation is not surprising, because experimental pIC50 values are very complicated, and depend on not only the binding energy, but also some other factors. Moreover, the docked results of all the studied compounds in the active site are clearly demonstrated in , and these results show that all compounds are adequately docked, and the orientations are similar as a whole. However, we repeated the CoMFA analysis based on the docked alignment instead of the common structure alignment, finding that the resulting leave-one-out (LOO) cross-validation coefficient q2 value was poor and too low (–0.029) to be accepted, and thus the CoMFA methodology cannot be effectively performed on the docked alignment for this series of compounds. In order to probe into the reason, we carefully compared with , and found that the docked conformations were not fully superposed on the above common structure alignment conformations, resulting in the unsatisfactory q2 value. It is not surprising, because any one methodology is not omnipotent and there are some inherent errors in DOCK; meanwhile, there is some difference between DOCK and CoMFA methodologies. Therefore, in this limited work, in order to effectively study the 3D-QSAR, we adopted conventional but acceptable CoMFA methodology, i.e. CoMFA was performed on common structure alignment instead of docked alignment. We found that such a treatment can result in acceptable CoMFA and docking consistency, at least at present.
Figure 5. (a, b) Binding conformations of 32 docked compounds at the active site of tubulin.
Table 2. CoMFA and docking results of the studied compounds.
In order to further understand the interaction between these compounds with tubulin in the active site, we selected the most potent inhibitor 12 in the experiment to perform the deeper docking study, which is discussed below.
Here, a complete overview of receptor–inhibitor binding interactions is presented as shown in , which represents the interaction model of the most potent inhibitor 12 with tubulin. Inhibitor 12 is suitably situated at the colchicine-binding site and results in various interactions with the active regions of the enzyme.
Figure 6. (a) Docking conformation of the most potent inhibitor 12 and corresponding surface of tubulin at the colchicine-binding site, in which the red and blue regions represent oxygen and nitrogen atoms, respectively, whereas white regions represent carbon or hydrogen atoms. (b) Interactions between the colchicine-binding site and compound 12.
The indolyl group, surrounded by the side chains of Asn258, Met259, Thr314, Asn349, Asn350, and Lys352, may form a π–cation interaction with the NH3+ group of Lys352 and a p–π interaction with the N and O atoms of Asn258. The trimethoxybenzoyl is located in a large hydrophobic pocket created by side chains of Leu242, Leu248, Ala250, Lys254, Leu255, Val318, Ala354, and Ile378. In addition, the electron-rich O atom of para methoxyl on the benzoyl ring can form a weak hydrogen bond (1.805 å) with the H atom linking to the S atom of Cys241.
3D-QSAR model
The 3D-QSAR model was established from CoMFA analysis and its statistical parameters are listed in . For a reliable predictive model, the cross-validation coefficient q2 should be greater than 0.5.
Table 3. Statistical parameters of 3D-QSAR model from CoMFA.
This 3D-QSAR model has high R2 (0.955), F (87.8), and q2 (0.66), as well as small SEE (0.204), suggesting that the established 3D-QSAR model is reliable and predictive. The R2bs of 0.961 and SDbs of 0.183 obtained from bootstrapping analysis (100 runs) further confirm the statistical validity and robustness of the established 3D-QSAR model. The steric and electrostatic contributions were found to be 66.3% and 33.7%, respectively. Therefore, the steric field has a greater influence than the electrostatic field, indicating that the steric interaction of the ligand with the receptor may be a more important factor for the cytotoxicity. The calculated activity values and the residual values of compounds for the 3D-QSAR model are also listed in .
A reliable QSAR model should be able to accurately predict activities of new compounds. The high q2 (0.66) in the above-mentioned training set only shows good internal validation, but we cannot automatically infer its high predictive ability for an external test setCitation21. In order to obtain a well-accepted QSAR model with high predictive ability, external validation is also necessary. Therefore, the 3D-QSAR model was further used to predict the activities of 16 congeneric compounds (see ) as an external test set from the literatureCitation22. The 16 compounds, also reported by Romagnoli and coworkers, were selected as the test set because they have specific bioactivity against the human T-lymphocyte Molt4/C8 cell line and have the same skeleton as the compounds in the training set. The predictive ability of the model is expressed by the predictive correlation coefficient R2pred, calculated by the formula: R2pred = (SD – PRESS)/SD, where SD is the sum of the squared deviations between the biological activities of the test set compounds and mean activity of the training set compounds, and PRESS is the sum of the squared deviations between the experimental and predicted activities of the test set compounds. From , the predictive correlation coefficient R2pred reaches 0.883 and the predicted deviations lie in a range of −0.659 to 0.428. Hence, the obtained results from an external test set further demonstrate that the established 3D-QSAR model has a satisfactory predictive ability. A plot of the calculated (predicted) pIC50 values by CoMFA versus the experimental values is shown in , in which most points are evenly distributed along the line Y = X, suggesting that the 3D-QSAR model is of good quality.
Table 4. Structures and computational results for 16 congeneric compounds as the test set.
Figure 7. Plot of calculated (predicted) activities vs. experimental ones for CoMFA analysis, in which 32 compounds in the training set are expressed as dots and 16 compounds in the test set are expressed as triangles.
The difference between R2 and q2 in the CoMFA model is quite high. The R2pred value may not be considered as the only criterion to indicate the external predictability of a modelCitation23. Hence, further tests on external validation are required.
Golbraikh and TropshaCitation24 have recommended that the correlation coefficient r between the predicted and observed activities of compounds from an external test set should be close to 1. At least one (but better both) of the correlation coefficients for regressions through the origin (predicted versus observed activities, or observed versus predicted activities), i.e. r02 or r0’2 should be close to r2. Furthermore, at least one slope of regression lines (k or k’) through the origin should be close to 1. Models are considered to be acceptable if they satisfy all of the following conditions: q2 > 0.5, r2 > 0.6, r02 or r0’2 close to r2, i.e. [(r2 – r02)/r2] or [(r2 – r0’2)/r2] < 0.1, and the corresponding 0.85 ≤ k ≤ 1.15 or 0.85 ≤ k’ ≤ 1.15. In this work, the established CoMFA model for the training set has higher R2 (0.955) and acceptable q2 (0.66). Meanwhile, it was used to predict activities of the 16 compounds in the test set, resulting in satisfactory statistical parameters: r2 = 0.806, r02 = 0.7974, [(r2 – r02)/r2] = 0.0107, k = 1.022 (see ). These parameters are in good agreement with the above criteria. In addition, according to Roy and RoyCitation23, an additional statistic for external validation rm2 is calculated −as rm2 = r2(1−(r2 – r02)1/2). The parameters r2 and r02 are squared correlation coefficient values between observed and predicted values of the test set compounds with and without intercept, respectively. For a model with good external predictability, the rm2 value should be greater than 0.5. However, the parameter rm2 is only applied to the test set; r2m(LOO) can be applied to the training set, considering the correlation between the observed and leave-one-out (LOO) predicted values of the training set compounds (see ). More interestingly, r2m(overall) can be applied to the whole set, considering the LOO predicted values of the training set compounds and predicted values of the test set compounds. The parameter r2m(overall) appears to be advantageous over other internal and external validation parameters since it is based on the prediction of both training and test set compounds and thus involves more compounds in the prediction processCitation25,Citation26. In our established CoMFA model, the additional external validation value rm2 reaches 0.7313 (see ), the training set validation value r2m(LOO) reaches 0.6543 (see ), and the whole set validation value r2m(overall) reaches 0.7127 (see ).
Figure 8. A regression of observed vs. predicted activities for 16 compounds from the external test set, in which the red solid line is not through the origin and the black dotted line is through the origin.
Figure 9. A regression of observed vs. LOO predicted activities for 32 compounds from the training set, in which the red solid line is not through the origin and the black dotted line is through the origin.
Figure 10. A regression of observed vs. predicted activities for 48 compounds from the whole set, in which the red solid line is not through the origin and the black dotted line is through the origin.
In short, all of the statistical and validating parameters show that the established model is satisfactory.
Main factors affecting the activity based on a combined docking and CoMFA study
The results of CoMFA can be displayed as vivid 3D contour maps, providing an opportunity to explain the observed variance in the cytotoxicity (expressed by pIC50). The steric interactions are represented by green and yellow contours, and green contours characterize the regions where bulky substituents would increase the biological activity, whereas yellow contours indicate regions where bulky substituents are detrimental to the biological activity. The electrostatic interactions are represented by blue and red contours, and blue contours indicate regions where positive charge increases the activity, whereas red contours indicate regions where negative charge increases the activity.
The steric contour map of CoMFA is displayed in . There is a green contour near substituent X, suggesting that the compound having a large group at this region can increase the activity. This can be used to explain the experimental fact that compounds 28 (8.000) and 29 (8.071) with bigger S as substituent X show higher activities than corresponding compounds 16 (6.854) and 17 (5.357) with O as substituent X. A large yellow contour embedding in the interspace between substituents R1 and X shows that introducing bulky groups in this region is unfavorable, because it can be blocked by near residue Met259. A large green contour is found close to the -CH3 part of -OCH3 as substituent R1. So, it is not strange that compounds 4 (6.194), 12 (8.420), and 29 (8.071) with larger -OCH3 as substituent R1 have rather higher activities than corresponding compounds 1 (5.854), 6 (5.678) and 9 (6.569), and 25 (6.538) and 27 (7.081) with H or -CH3 as substituent R1. Meanwhile, a small green contour is also found at some distance from the middle site of substituents R2 and R3; this suggests that a large substituent reaching this area is favorable, which is consistent with the fact that there is a big cavum in docking. In addition, there is a green contour near the substituents R5 and R6. In docking, the two green contours are located on the edge of the entrance of the large active pocket, so bigger groups in the two regions are favorable. This may be the reason why compounds 19 (7.319) and 20 (6.174) with -CH3 as substituents R5 and R6, respectively, have higher activities than compounds 16 (6.854) and 17 (5.357) with H atoms as substituents R5 and R6. Moreover, there is a yellow contour under the two green contours, and it is in good agreement with docking, in which the residue Asn258 is located on this region, blocking the prolongation of substituents R5 and R6.
Figure 11. CoMFA steric contour map for compound 12 with the highest activity.
The electrostatic contour map of CoMFA () shows two blue contours. The larger is near the -CH3 part of substituent X; this may result in electrostatic interactions between the electropositive part of substituent X and the electron-rich S atom of residue Met259. The other is near the substituent R2; this may result in electrostatic interactions between the electropositive part of substituent X and the electron-rich O atoms of residue Asn349. These indicate that compounds having high electropositive (i.e. low electronegative) groups on these positions exhibit good activity, and is also in satisfactory agreement with DOCK. For example, compounds 8 (6.959), 10–12 (5.678, 7.244, 8.420), and 13 (5.301) with -NCH3 as substituent X exhibit higher activities than the corresponding compounds 1 (5.854), 2–4 (5.161, 6.201, 6.194), and 5 (5.066) with -NH as substituent X, respectively, because the former terminal H atoms with positive electronic charges just (or almost) fall into the blue area. Likewise, the fact that compounds 11 (7.244) and 28 (8.000) with -OCH3 as substituent R2 are more active than corresponding compounds 8 (6.959) and 26 (7.469) with -CH3 as substituent R2 can also be interpreted.
Figure 12. CoMFA electrostatic contour map for compound 12 with the highest activity.
Conclusions
The docking and CoMFA methods were synthetically applied to study a series of CA-4 analogs as inhibitors toward tubulin at the colchicine-binding site. The appropriate binding orientations and conformations of these compounds interacting with tubulin were located by docking study, and a satisfactory 3D-QSAR model in terms of high R2 (0.955) and q2 (0.66) as well as small SEE (0.204) was established by CoMFA. It is very interesting to find consistency between the 3D topology structure of the active site of tubulin from docking analysis and the CoMFA field distribution for this series of compounds. Moreover, the predictive correlation coefficient R2pred for the test set reached 0.883. Other tests on additional validations further confirm the satisfactory predictive power of the model. Some key factors responsible for the cytotoxicity can be summarized as follows:
(1) Introducing a stronger electropositive and bulky group as substituent X, which can be suitably situated at the cavum of the active site and can interact with Met259, may increase the activity.
(2) The substituent R1 should be selected as a bulky group, because there is a large green contour around it in the 3D-QSAR model.
(3) Higher electropositivity on the terminal part of substituent R2 is favorable to the activity, as it may easily result in electrostatic interaction with Asn349.
(4) Selecting substituents R5 and R6 to be moderate-sized groups may enhance the activity of the compound.
This work further shows that a combined CoMFA and docking study can provide more useful information to understand the structural features of targets and ligands. The obtained results will help with the action mechanism analysis and molecular design of new inhibitors with higher activity.
Acknowledgments
We are pleased to acknowledge financial support from the National Natural Science Foundation of China (No.20673148). We heartily thank Molecular Discovery Ltd. for giving us the DOCK 6.0 program as a freeware and the College of Life Sciences, Sun Yat-Sen University, for SYBYL 6.9 computation environment support. We are also sincerely thankful to the referees for their valuable suggestions to improve this work.
Declaration of interest: Except for the above support and funding received by the authors to carry out the study, no conflict of interest is required.
References
- Bacher G, Beckers T, Emig P, Klenner T, Kutscher B, Nickel B. New small-molecule tubulin inhibitors. Pure Appl Chem 2001;73:1459–64.
- Jordan A, Hadfield JA, Lawrence NJ, McGown AT. Tubulin as a target for anticancer drugs: agents which interact with the mitotic spindle. Med Res Rev 1998;18:259–96.
- Pettit GR, Rhodes MR. Antineoplastic agents 389. New syntheses of the combretastatin A-4 prodrug. Anticancer Drug Des 1998;13:183–91.
- Romagnoli R, Baraldi PG, Sarkar T, Carrion MD, Lopez-Cara C, Cruz-Lopez O, et al. Synthesis and biological evaluation of 1-methyl-2-(3′,4′,5′-trimethoxybenzoyl)-3-aminoindoles as a new class of antimitotic agents and tubulin inhibitors. J Med Chem 2008;51:1464–8.
- Romagnoli R, Baraldi PG, Sarkar T, Carrion MD, Cruz-Lopez O, Lopez-Cara C, et al. Synthesis and biological evaluation of 2-(3′,4′,5′-trimethoxybenzoyl)-3-N,N-dimethylamino benzo[b]furan derivatives as inhibitors of tubulin polymerization. Bioorg Med Chem 2008;16:8419–26.
- Romagnoli R, Baraldi PG, Carrion MD, Lopez-Cara C, Preti D, Fruttarolo F, et al. Synthesis and biological evaluation of 2- and 3-aminobenzo[b]thiophene derivatives as antimitotic agents and inhibitors of tubulin polymerization. J Med Chem 2007;50:2273–7.
- Liao SY, Chen JC, Qian L, Shen Y, Zheng KC. QSAR studies and molecular design of phenanthrene-based tylophorine derivatives with anticancer activity. QSAR Comb Sci 2008;27:280–8.
- Liao SY, Xu LC, Qian L, Zheng KC. QSAR and action mechanism of troxacitabine prodrugs with antitumor activity. J Theor Comput Chem 2007;6:947–58.
- Liao SY, Chen JC, Qian L, Shen Y, Zheng KC. QSAR, action mechanism and molecular design of flavone and isoflavone derivatives with cytotoxicity against HeLa. Eur J Med Chem 2008;43:2159–70.
- Liao SY, Qian L, Lu HL, Shen Y, Zheng KC. A combined 2D- and 3D-QSAR study on analogues of ARC-111 with antitumor activity. QSAR Comb Sci 2008;27:740–9.
- Liao SY, Qian L, Chen JC, Lu HL, Zheng KC. 2D and 3D-QSAR studies on antiproliferative thiazolidine analogs. Int J Quantum Chem 2008;108:1380–90.
- Liao SY, Qian L, Chen JC, Shen Y, Zheng KC. 2D/3D-QSAR study on analogues of 2-methoxyestradiol with anticancer activity. J Theor Comput Chem 2008;7:287–301.
- Liao SY, Qian L, Miao TF, Lu HL, Zheng KC. Theoretical studies on QSAR and mechanism of 2-indolinone derivatives as tubulin inhibitors. Int J Quantum Chem 2009;109:999–1008.
- Liao SY, Qian L, Miao TF, Lu HL, Zheng KC. CoMFA and docking studies of 2-phenylindole derivatives with anticancer activity. Eur J Med Chem 2009;44:2822–7.
- Farce A, Loge C, Gallet S, Lebegue N, Carato P, Chavatte P, et al. Docking study of ligands into the colchicine binding site of tubulin. J Enzyme Inhib Med Chem 2004;19:541–7.
- SYBYL 6.9 [CP]. St. Louis: Tripos Associates, Inc., 2001.
- Cho SJ, Tropsha A. Cross-validated R2-guided region selection for comparative molecular field analysis: a simple method to achieve consistent results. J Med Chem 1995;38:1060–6.
- Cramer RD III, Patterson DE, Bunce JD. Comparative molecular field analysis (CoMFA). 1. Effect of shape on binding of steroids to carrier proteins. J Am Chem Soc 1988;110:5959–67.
- Bush BL, Nachbar RB Jr. Sample-distance partial least squares: PLS optimized for many variables, with application to CoMFA. J Comput Aided Mol Des 1993;7:587–619.
- Tetko IV, Tanchuk VY, Villa AE. Prediction of n-octanol/water partition coefficients from PHYSPROP database using artificial neural networks and E-state indices. J Chem Inf Comput Sci 2001;41: 1407–21.
- Roy DR, Sarkar U, Chattaraj PK, Mitra A, Padmanabhan J, Parthasarathi R, et al. Analyzing toxicity through electrophilicity. Mol Divers 2006;10:119–31.
- Romagnoli R, Baraldi PG, Remusat V, Carrion MD, Cara CL, Preti D, et al. Synthesis and biological evaluation of 2-(3′,4′,5′-trimethoxybenzoyl)-3-amino 5-aryl thiophenes as a new class of tubulin inhibitors. J Med Chem 2006;49:6425–8.
- Roy PP, Roy K. On some aspects of variable selection for partial least squares regression models. QSAR Comb Sci 2008;27:302–13.
- Golbraikh A, Tropsha A. Beware of q2! J Mol Graphics Mod 2002;20:269–76.
- Roy K, Roy PP. Comparative chemometric modeling of cytochrome 3A4 inhibitory activity of structurally diverse compounds using stepwise MLR, FA-MLR, PLS, GFA, G/PLS and ANN techniques. Eur J Med Chem 2009;44:2913–22.
- Roy PP, Paul S, Mitra I, Roy K. On two novel parameters for validation of predictive QSAR models. Molecules 2009;141660–701.