Abstract
Ureido-substituted benzene sulfonamides are the most important class of CA inhibitors which significantly inhibited the formation of primary tumors and metastases. Here, we present quantitative structure activity relationships (QSAR) study on a pool of 27 such inhibitors. A heuristic algorithm selected the best multiple linear regression (MLR) equation, showing the correlation between the observed values and the calculated values of activity. The calculated values of activity were in good agreement with the experimental values. The novelty of this work consists in not only exploring the structural attributes of bioactive molecules but also in predicting in silico the structures of new compounds which may show antimetastatic activity. The not yet synthesized such molecules (i.e. the prediction set) included many compounds showing a higher computed activity compared to the reported such derivatives, but they have been however not yet assayed.
Introduction
Human carbonic anhydrase (hCA, EC 4.2.1.1) IX is one of the three trans-membrane such isozymes (together with hCA XII and XIV) which has been shown to be involved in tumorigenesis (together with hCA XII but not hCA XIV)Citation1,Citation2. It has recently been demonstrated that CA IX is a druggable target for imaging and treatment of hypoxic tumorsCitation1–9, as it is overexpressed in many types of cancersCitation3–10. Several studies showed that a clear-cut relationship between high hCA IX levels in tumors and a poor prognosisCitation11–13. The hCA IX is involved in the tumor acidification processes, providing H+ ions to the extracellular milieu by means of the CO2 hydration reaction to bicarbonate and protons. The pH of tumors is in fact more acidic by 0.5–1.0 pH units than that of the surrounding normal tissueCitation14, and this acidic environment plays a crucial role both in the growth, dissemination and propagation of tumor cells and in their non-responsiveness to chemo- and radio-therapyCitation14–19. Hypoxia is known to be a distinct feature of the tumor microenvironment, being the result of uncontrolled tumor growth outpacing the rate of vascular proliferation and of architecturally defective microcirculation. The process compromises oxygen diffusion to tumor cells adjacent to neo-vesselsCitation20,Citation24. In hypoxia, the cells are deprived of sufficient O2 for maintaining oxidative phosphorylation, and thus they die either by necrosis or programmed cell death, apoptosisCitation20. The most prominent alterations observed in cancer are after their exposure to chronic and/or acute hypoxia. This exposure to hypoxia affects tumor progression in multiple ways, including the induction of angiogenesis and a metabolic switch towards elevated glucose uptake and elevated glycolysis, which may generate the acidic disbalance observed in tumors and mentioned above, together with the contribute of acidification coming from the CA-catalyzed reactionCitation22–23,Citation25.
Recently, the synthesis of a large series of ureido-substituted sulfonamides that possess strong affinity for CA IX, and excellent in vivo antimetastatic effects in breast cancer xenograft models, has been reportedCitation26. It was thus reported that 4-{[(3-nitrophenyl) carbamoyl] amino} benzene sulfonamide (one of the compounds of this series), significantly inhibited the formation of metastases by the highly aggressive 4T1 mammary tumor cells at pharmacologic concentrations of 45 mg/kg, constituting an interesting candidate for the development of conceptually novel anti-cancer/anti-metastatic drugsCitation26. The compound also inhibited the growth of primary tumorsCitation26.
QSARs, or Quantitative structure–property relationships (QSPRs), are mathematical models that to relate the structure-derived features of a compound to its biological or physico-chemical activity. QSAR (also QSPR) works on the assumption that structurally similar compounds have similar activities. Therefore, these methods have predictive and diagnostic ability. They can be used to predict the biological activity of compounds before the actual biological testing. They can also be used in the analysis of structural characteristics that can give rise to the properties of interestCitation27. We have attempted to build QSAR models using the MOPAC and PRECLAV software to explore the correlations between the calculated molecular descriptors of some ureido-substituted benzene sulfonamides compounds and their experimental inhibitory activity against the CA IX inhibitors.
QSAR model of CA IX inhibitory activity is important because they summarize and organize information, and assist in the prediction of inhibitory activity for new compounds which may be important for novel antimetastatic compounds design. Thus, QSAR approach saves resources and shortens the process of development of new drugs.
Materials and methods
Calibration set (experimental data)
Recently one of our groupsCitation26 reported the inhibition study against newly CA IX carbonic anhydrase (CA, EC 4.2.1.1) with a series of ureido-substituted benzene sulfonamides derivatives. The ureido-substituted benzene sulfonamides investigated for the inhibition of the CA IX of types 1–27 are shown in .
Table 1. Structural detail and CA IX inhibitory activities (in nM and A = log 3 × 104/KI), estimated activities, hat diagonal, standardized residual, |RStudent| of the calibration set molecules 1–27 and of the not yet synthesized ones 28–49.
The inhibition data of ureido-substituted benzene sulfonamides derivatives compounds are shown in exhibiting a wide range of inhibitory activity. The enzyme inhibition data KI values in the nanomolar (nM) were converted in “A” according to the formula A = log (3 × 104/KI) and subsequently, used as the dependent variable for the QSAR study (). The inhibitory activity (A) value of the molecules under the study spanned a wide range from 1 to 5.
Prediction set (design of new compounds)
The prediction set contained 22 other not yet synthesized ureido-substituted benzene sulfonamides generated by the BroodCitation28 software, having unknown observed values of activity presented in (compound no. 28–49). Brood uses the shape and attachment geometry of the query fragment to identify a family of similar fragments. The structures of the prediction set molecules were selected mainly due to their possibility to be synthesized in laboratory conditions and taking into account the commercial availability of the raw materials.
Descriptor calculation and quality of the model
The minimum energy geometry, for each compound was obtained by the conformational search ability of the Omega v.2.4.3Citation29–31 program. The isomeric SMILES notation was used as program input in order to avoid any influences on conformational model generation by presenting 3D seed structures. The omega employs a rule-based algorithmCitation29,Citation30 in combination with variants of the Merck molecular force field 94. The force field used was the 94s variant of the MMMF_NoEstat (Merck Molecular force field)Citation29–31 includes all MMFF terms except coulomb interactions. The conformations of minimum energy obtained by molecular mechanics calculations were further minimized by quantum chemical calculations. The semi empirical PM6 methodCitation32 included in the MOPAC 2009 softwareCitation33 optimized the geometry more rigorously.
The energy minimized structure was used to calculate different molecular properties, including virtual fragmentation descriptors, Whole molecule quantum chemical (global) descriptors. For each molecule, over 470 descriptors were calculated using programs such as MOPACCitation29–33 and PRECLAVCitation34,Citation35. The parameters to be calculated are various descriptors that are indicative of molecular structure and used as independent variables. The DRAGON topological descriptors were not used because the physical sense of these descriptors is difficult to emphasize.
Several criteria were used to reduce the descriptors while optimizing the information content of the descriptors set. First, descriptors for which no value was available for all the compounds were disregarded. Second, descriptors of which the value is constant (or near-constant) inside each group of descriptors were excluded. Identification of the “significant” descriptors uses specific criteriaCitation34. The “significant” descriptors are those which are sufficiently correlated with the dependent property. The variables having high enough diversity of values are considered significant only if their quality q is high enough.
where
(Here min r2 = 0.01) and r2 is the square of the Pearson linear correlation between the values of the analyzed descriptor and the values of the dependent property.
The experimental information associated with biological activity was used as dependent variables in building a QSAR model. The parameters to be calculated were various descriptors that are indicative of molecular structure and used as independent variables. The PRECLAV algorithmCitation34,Citation35 was used for obtaining the parameters and for the statistical analysis as reported earlierCitation36–44.
Stepwise multiple linear regression (MLR) technique was used for the QSAR model development using the entire dataset. Using only the “significant” descriptorsCitation34,Citation35 PRECLAV computes thousands of QSAR equations, i.e. multilinear formulas of the dependent property.
The program combines successively sets with 2, 3 , … , k significant descriptors (1 < k < 11). A set of descriptors contains only descriptors that are sufficiently low intercorrelated and fulfill criteria (Equation 2).
where
is the square of Pearson linear correlation between the values of two descriptors present in the same set, N is the number of molecules in the calibration set (here N = 27).
Each set of descriptors has been used to calculate a multilinear QSAR equation of type (Equation 3).
where A is represents a dependent property (here the inhibitory activity defined above), C0 is the free term (intercept), Ck are the coefficients (weighting factors) of the descriptors, Dk are some significant descriptors and k is the number of descriptors in the set.
The relative utility (U) of a certain descriptor on dependent property values was computed by the Specific procedureCitation34. The descriptor which present a high value for U, within the range [0, 1000], may be considered very useful in estimating the activity, because they correlate very well with activity and do not correlate with other predictors. Each “useful” descriptor offers ample information about the variation in activity from molecule to molecule.
After computing the Acalc values of the inhibitory activity for the prediction set molecules, PRECLAV arranged these molecules according to the estimated values. It is computed average value for the estimated values and standard deviation (σ) of the estimated values. The program considers “high values” as the values fulfilling the criterion (Equation 4) and “low values” as the values fulfilling the criterion (Equation 5). Here, the molecules having “high” computed value of inhibitory activity have been taken as recommended for “synthesis”Citation38.
The “quality” of each QSPR was computed using usual statistical formulas that are a measure of agreement of observed/computed values of activity: standard error of estimation Se, Pearson square correlation r2, Fisher function F and cross-validated Pearson square correlation . The concordance between the calculated/observed values has been calculated using the quality function QCitation34 which possesses values in the interval {−1, 1}.
where r2 is Pearson square linear correlation between computed/observed values and N is the number of molecules in the calibration set (here N = 27). By increasing the number of descriptors k, the quality Q of the equations increases, reaches a maximum and then decreases. For predictions, the equation with the highest quality was used, the descriptors present in this equation being called “predictors”. The best way to evaluate quality of regression model is leave one out (LOO) cross validation, one object (one biological activity value) is eliminated from dataset and dataset is divided into subsets (number of subsets = number of data points) of equal size. The cross-validated function
is a measure of homogeneity of calibration set from the point of view of predictors' set, i.e. from the point of view of structure–property relationship. The rank correlation KendallCitation45 and predictive squared correlation coefficient (
)Citation46 are also used to validate the model. Predictive potential of a model on the new dataset is influenced by the similarity of chemical nature between calibration set and prediction setCitation47.
Applicability of domain and detection of outliers
A QSAR model can be used for screening new compounds if its domain of application is definedCitation46,Citation48. The need to characterize the model applicability domain is also reflected in the OECD guidelines for QSAR model validationCitation49,Citation50. QSAR model should only be used for making predictions of compounds fall within the specified domain may be considered reliable. Extent of extrapolationCitation51,Citation52 is one simple approach to define the applicability of the domain. It is based on the calculation of the hat diagonal (leverage) hi for each chemical, where the QSAR model is used to predict its activity:
In Equation (7), xi is the descriptor-row vector of the query molecule and X is the k × n matrix containing the k descriptor values for each one of the n training molecules. A hat diagonal (leverage) value >3(k + 1)/n leverage warning limitCitation50 is considered large.
Outliers are compounds that are poorly fit by the regression model. Outlying compounds should not be removed unless a good reason for their removal can be given. The variance of the observed residuals is not constant. This makes comparisons among the residuals difficult. One solution is to standardize the residualsCitation53,Citation54 by dividing by their standard deviations. This gives a set of residuals |RStudent| (cross-validated LOO standardized residuals) is a standardized residual that has the impact of a single observation removed from the mean square error. A molecule is defined as an outlier in which |RStudent| > 2Citation54.
To visualize the applicability of domain of a developed QSAR model, William plot was used. In the William plot, |RStudent| versus leverage values (hi) are plotted. This plot could be used for an immediate and simple graphical detection of both the response outliers and structurally influential compounds in a model. It must be noted that compounds with high value of leverage and good fitting in the developed model can stabilize the model. On the other hand, compounds with bad fitting in the developed model may be outliers. Thus, combination of leverage and the |RStudent| could be used for assigning the applicability of domain.
Results and discussions
The statistical computations were conducted using the specific formulas and procedures of PRECLAV program algorithm. Using only the “significant” descriptors PRECLAV computes 10 000 QSPR typeCitation3 multilinear equations. The quality of the obtained equations is reflected by the value of the Q function and also by values of some usual statistical functions. During the PRECLAV MLR analysis, we observed that the equation with highest value of the Q function is five parametric model and this model also has the highest predictive power which is as follows:
Dependent property: hCA IX inhibitory activity (A).
Molecules number in calibration set: 27.
Number of “significant” descriptors in presence of prediction set = 195.
D1 xhc (Maximum charge of H in H–C bonds; U = 988), D2 miz Dipole moment (Z component; U = 617), D3 = atg (percentage of atoms in aromatic circuits; U = 1000), D4 = rnc (Number of CN aromatic bonds/Number of bonds ratio; U = 932), D5 = axh (Average distance between OH/NH functional groups; U = 472).
Whereas the quality of correlation is described by the statistical indices:
where Se = standard error of values, r2 = Pearson square correlation, F = Fisher function,
= Pearson cross-validated square correlation (LOO method), K = Kendall rank correlation.
The negative correlation of xhc in the equation shows that increase in the maximum charge of H in H–C bonds like property will decrease the inhibitory activity. It is confirmed with the high value of xhc for compounds 1, 9, 14, 15, 21 and 23 which decreases the inhibitory activity. The combined effect of the absence of substitution in benzene and high value of xhc makes the compound no. 1 is lowest in all the inhibitory activity. The negative correlation of miz descriptor indicates that increasing the dipole moment (Z component) also decreases the inhibitory activity.
Compounds 1, 3, 20 and 21 have high value of atg (Percentage of atoms in aromatic circuits) and due to the high atg value it has a decreased inhibitory activity. The low value of atg in compound 25 indicate that the existence of cyclopentyl in this compound has improved the inhibitory activity. The rnc represents the number of CN aromatic bonds/number of bonds ratio, so the CN bonds play a dominant role for the inhibitory activity. The negative correlation of axh in the equation shows that increase in the value of this predictor decreases the inhibitory activity.
|RStudent| (cross-validated LOO standardized residuals) is one of the best single diagnostics for capturing large residuals. This diagnostics confirm that the zero compounds are outliers in calibration set. The minimum correlation descriptor/activity is computed for D5 (r2 = 0.032). The minimum intercorrelation between descriptors D1 and D4 (r2 = 0) and the maximum intercorrelation between descriptors is computed for D2/D5 pair (r2 = 0.2978).
The correlation matrix for selected parameters is given in . The correlation matrix shows no intercorrelation of selected descriptors. To further check the intercorrelation of the descriptors variance inflation factor (VIF) and Tolerance was calculatedCitation55. In practice, when VIF > 5 or if the tolerance remains less than 0.20, then this would indicate multicollinearity among the descriptors. The calculated VIF and tolerance values are given against each descriptor in . The VIF values are all less than 2.5 and tolerance greater than 0.2 indicating that here is no multicollinearity and the model is stable.
Table 2. Inter correlation of predictors and collinearity statistics.
Using the above QSAR equation, the maximum activity computed for calibration set molecules is 4.75, the average activity computed for calibration set molecules is 3.132 ± 0.8683 and the average activity computed for prediction set molecules is 4.22 ± 0.821. According to criterion (Equation 4), this equation identified nine molecules in prediction set having high values of CA IX inhibitory activity “suggested for synthesis”. In , the calculated values of not yet synthesized compound no. 28–49 identified by the program as high have been marked in bold letters, while the values identified as low have been underlined.
The validation set was extracted from the homogenized calibration set. In this study, the selection of the validation set is based on the hierarchical clustering techniqueCitation56. Cluster analysisCitation56 is a method of arranging objects into groups. In this study, the molecules with rank 04, 10, 14, 17, 22 and 27 constituted the validation set and the remaining molecules form the reduced calibration set. The validation set of six molecules (22% of database) captures all the features and span the activity range of the entire dataset. We can assume that the reduced calibration set (training set) obtained in this way is a representative sample for the calibration set. The remaining 21 molecules formed the reduced calibration set. The value of predictive r2 () greater than the stipulated value of 0.5 reflects efficient prediction for the validation set molecules by the developed modelCitation46. In the presence of the validation set, we obtained the four parametric QSAR model for the reduced calibration set (for 21 molecules) with the predictors (miz, xhc, atg and rnc) used in the above QSAR study and obtained results (r2 = 0.8688, F = 26.5024, Se = 0.37349,
= 0.7848 and
= 0.6652). The predictive r2 (
> 0.5) parameter indicates significant ability of the developed model to predict the inhibitory activity of new compounds. We can state that the estimated value for the molecules in the validation set are close to the experimental ones and has ordered the molecules in a sequence similar enough to the real one CA IX inhibitory value. This one was confirmed by graph () between observed and estimated value of calibration set and validation set. The result, qualitatively correct suggests that QSAR model obtained from the above study have sufficient accuracy for the prediction of the activity of new not yet synthesized molecules.
As discussed earlier, we used |RStudent| of observed inhibitory activity calculated by the obtained models and hat diagonal (leverage) for assigning applicability of domain (AD). Values of leverage could be calculated for both calibration set and prediction set compounds shown in (). Applicability of domain for the developed model is shown in William plot . Influential compounds are points with leverage value higher than the warning leverage limit. It can be seen in the William plot; all molecules in calibration set lie in the application domain of the developed model. None of the molecules have a hat diagonal (leverage) value higher than warning leverage limit (0.666), and also, none of the molecules have higher |RStudent| (cross-validated LOO standardized residuals) than threshold limit |RStudent| < 2.
Figure 1. |RStudent| of observed versus Hat Diagonal.
Figure 2. Graphs of observed versus estimated activity in the calibration set and validation set.
We have developed a computer representation of the pharmacophore model; this also includes information on the available space at important substituent positions. represent pharmacophore models with most active compound (Compound no. 19) which is generated by BroodCitation28. The model displays seven pharmacophore elements (three hydrogen bond donors and four hydrogen bond acceptors) which are used to develop and describe the interaction between ligands and the target receptor from the ligand point of view.
Figure 3. Pharmacophore model.
Conclusions
Statistically significant linear QSAR models imply the proposal of CA IX inhibitors for: data representation, data modeling and data prediction. Based on the correlations, we have identified the key properties. These properties can subsequently be utilized for the manipulation of structural features in the ureido-substituted moiety in a bid to achieve ureido-substituted benzene sulfonamides derivatives with higher potency than those already reported in the literature.
The QSAR model conclude that the maximum charge of H in H–C bonds are not favorable to inhibitory activity, low percentage of atoms in aromatic circuits are favorable for inhibitory activity, polarity also influences the activity and the carbon nitrogen bond play a dominant role for inhibitory activity. The predicted compounds having high activity belong to the carbon nitrogen bonds. The analyzed prediction set includes many molecules having high estimated inhibitory activity and their hat diagonal (leverage) are within threshold limit is recommended for synthesis. Thus, attempts have been made to design and develop new drugs against CA IX inhibitory activity on a rational basis so as to decreases the trial and error factor and predict the biological activity before synthesis.
Declaration of interest
One author (S. S.) declares no conflict of interest, whereas CTS declares conflict of interest as he is author on a patent claiming these derivatives as anticancer/antimetastatic agents. One of the author (S. S.) expresses her thanks to the University Grants Commission, New Delhi, India for providing financial support under UGC Research Award No. F.30-29 /2011(SA-II). The work was also financed by an EU FP7 project (Metoxia) to CTS.
Supplementary material available online
Value of the predictors (xhc, miz, atg, rnc, axh) used in calibration set and prediction set, observed, estimated and residual values of h CA IX inhibitory activity (A) for the molecules used in the reduced calibration set (training set) and validation set can be found in Supplementary Material associated with this article.
Supplementary Material
Download PDF (49.6 KB)Acknowledgements
This article is dedicated to the memory of the late Prof. Padmakar V. Khadikar (1936–2012).
References
- Supuran CT. Carbonic anhydrases: novel therapeutic applications for inhibitors and activators. Nat Rev Drug Discov 2008;7:168–81
- Ozensoy O, De Simone G, Supuran CT. Drug design studies of the novel antitumor targets carbonic anhydrase IX and XII. Curr Med Chem 2010;17:1516–26
- Alterio V, Hilvo M, Di Fiore A, et al. Crystal structure of the extracellular catalytic domain of the tumor-associated human carbonic anhydrase IX. Proc Natl Acad Sci USA 2009;106:16233–8
- Neri D, Supuran CT. Interfering with pH regulation in tumours as a therapeutic strategy. Nat Rev Drug Discov 2011;10:767–77
- Dubois L, Douma K, Supuran CT, et al. Imaging the hypoxia surrogate marker CA IX requires expression and catalytic activity for binding fluorescent sulfonamide inhibitors. Radiother Oncol 2007;83:367–73
- Swietach P, Wigfield S, Cobden P, et al. Tumor-associated carbonic anhydrase 9 spatially coordinates intracellular pH in three-dimensional multicellular growth. Biol Chem 2008;283:20473–83
- Swietach P, Wigfield S, Supuran CT, et al. Cancer-associated, hypoxia-inducible carbonic anhydrase IX facilitates CO2 diffusion. BJU Int 2008;101:22–4
- De Simone G, Alterio V, Supuran CT. Exploiting the hydrophobic and hydrophilic binding sites for designing carbonic anhydrase inhibitors. Expert Opin Drug Discov 2013;8:793–810
- Ahlskog JKJ, Dumelin CE, Trüssel S, et al. In vivo targeting of tumor-associated carbonic anhydrases using acetazolamide derivatives. Bioorg Med Chem Lett 2009;19:4851–6
- Ahlskog JK, Schliemann C, Mårlind J, et al. Human monoclonal antibodies targeting carbonic anhydrase IX for the molecular imaging of hypoxic regions in solid tumours. Cancer 2009;101:645–57
- Monti SM, Supuran CT, De Simone G. Anticancer carbonic anhydrase inhibitors: a patent review (2008–2013). Expert Opin Ther Pat 2013;23:737–49
- Chia SK, Wykoff CC, Watson PH, et al. Prognostic significance of a novel hypoxia-regulated marker, carbonic anhydrase IX, in invasive breast carcinoma. J Clin Oncol 2001;19:3660–8
- Scozzafava A, Owa T, Mastrolorenzo A, Supuran CT. Anticancer and antiviral sulfonamides. Curr Med Chem 2003;10:925–53
- Supuran CT. Carbonic anhydrases: from biomedical applications of the inhibitors and activators to biotechnologic use for CO2 capture. J Enzyme Inhib Med Chem 2013;28:229–30
- Pastorekova S, Parkkila S, Pastorek J, Supuran CT. Carbonic anhydrases: current state of the art, therapeutic applications and future prospects. J Enzyme Inhib Med Chem 2004;19:199–229
- Svastova E, Hulikova A, Rafajova M, et al. Hypoxia activates the capacity of tumor-associated carbonic anhydrase IX to acidify extracellular pH. FEBS Lett 2004;577:439–45
- Alp C, Ozsoy S, Alp NA, et al. Sulfapyridine-like benzenesulfonamide derivatives as inhibitors of carbonic anhydrase isoenzymes I, II and VI. J Enzyme Inhib Med Chem 2012;27:818–24
- Liu F, Martin-Mingot A, Lecornué F, et al. Carbonic anhydrases inhibitory effects of new benzenesulfonamides synthesized by using superacid chemistry. J Enzyme Inhib Med Chem 2012;27:886–91
- Supuran CT. Structure-based drug discovery of carbonic anhydrase inhibitors. J Enzyme Inhib Med Chem 2012;27:759–72
- Ekinci D, Kurbanoglu NI, Salamcı E, et al. Carbonic anhydrase inhibitors: inhibition of human and bovine isoenzymes by benzenesulphonamides, cyclitols and phenolic compounds. J Enzyme Inhib Med Chem 2012;27:845–8
- Chohan ZH, Shad HA, Supuran CT. Synthesis, characterization and biological studies of sulfonamide Schiff's bases and some of their metal derivatives. J Enzyme Inhib Med Chem 2012;27:58–68
- Sentürk M, Ekinci D, Göksu S, Supuran CT. Effects of dopaminergic compounds on carbonic anhydrase isozymes I, II, and VI. J Enzyme Inhib Med Chem 2012;27:365–9
- McDonald PC, Winum JY, Supuran CT, Dedhar S. Recent developments in targeting carbonic anhydrase IX for cancer therapeutics. Oncotarget 2012;3:84–97
- Garaj V, Puccetti L, Fasolis G, et al. Carbonic anhydrase inhibitors: novel sulfonamides incorporating 1,3,5-triazine moieties as inhibitors of the cytosolic and tumour-associated carbonic anhydrase isozymes I, II and IX. Bioorg Med Chem Lett 2005;15:3102–8
- Singh S. A QSAR Study on novel series of carbonic anhydrase inhibitors hCA IX – tumor-associated (Hypoxia). Med Chem 2012;8:656–72
- Pacchiano F, Carta F, McDonald PC, et al. Ureido-substituted benzenesulfonamides 1-potently inhibit carbonic anhydrase IX and show antimetastatic activity in a model of breast cancer metastasis. J Med Chem 2011;54:1896–902
- Dhmer M, Steib EF. Quantitative and network biology. In: Dehmer M, Varmuza K, Bonchev D, eds. Statistical modelling of molecular descriptors in QSAR/QSPR, Vol. 2. Germany: Wiley-Blackwell; 2012:1–2
- BROOD, version 2.0.0. OpenEye Science Software, 3600 Cerrillos Road, Suite 1107, Santa Fe, USA; 2010
- OMEGA, version 2.4.3. OpenEye Science Software, 3600 Cerrillos Road, Suite 1107, Santa Fe, USA; 2010
- Tresadern G, Bemporad D, Howe TA. Comparison of ligand based virtual screening methods and application to corticotropin releasing factor 1 receptor. J Mol Graph Model 2009;27:860–70
- Halgre TA. MMFF VI. MMFF94s option for energy minimization studies. J Comput Chem 1999;20:720–9
- Stewart JJP. Optimization of parameters for semiempirical methods. V. Modification of NDDO approximations and application to 70 elements. J Mol Model 2007;13:1173–213
- Stewart JJP. MOPAC2009, Stewart Computational Chemistry, Colorado Springs, CO, USA. Available from: http://OpenMOPAC.net; 2008
- PRECLAV v. 1203 (documentation included) is available from Center of Organic Chemistry, Bucharest; 2012
- Tarko L. QSPR/QSAR computations by PRECLAV software. Rev Chim (Bucuresti) 2005;56:639–48
- Tarko L, Stecoza CE, Ilie C, Chifiriuc MC. QSAR studies on antibacterial activity of some substituted dihydrodibenzothiepins. Rev Chim (Bucuresti) 2009;60:476–9
- Tarko L. QSAR studies regarding the inhibition of the carbonic anhydrase by the sulfonamides containing a picolinoyl Group. Rev Chim (Bucuresti) 2007;58:191–4
- Tarko L, Avram S, Mihailescu D. Prediction for antidepressants activity using QSAR Study. Rev Chim (Bucuresti) 2011;62:371–5
- Done R, Mandrila G, Tarko L. QSAR study concerning toxicity and threshold limit value VL8 of chlorine containing compounds. Rev Chim (Bucuresti) 2009;60:992–5
- Singh S, Khadikar PV, Scozzafava A, Supuran CT. QSAR studies for the inhibition of the trans membrane carbonic anhydrase isozyme XIV with sulfonamides using PRECLAV software. J Enzym Inhib Med Chem 2009;24:337–49
- Singh S. Comparative QSAR studies on the novel series of thiazolones and tetrazole derivatives as HCV NS5B polymerase allosteric inhibitors. Lett Drug Des Discov 2009;6:286–97
- Singh S, Singh S, Shukla P. Modeling of novel HIV-1 protease inhibitors incorporating N-aryl-oxazolidinone-5-carboxamides as P2 ligands using quantum chemical and topological finger print descriptors. Med Chem Res 2011;20:1556–65
- Singh S, Supuran CT. Chemometric QSAR modeling and in silico design of carbonic anhydrase inhibition of a coral secretory isoform by sulfonamide. Bioorg Med Chem 2013;21:1495–502
- Singh S, Supuran CT. 3D-QSAR CoMFA studies on sulfonamide inhibitors of the Rv3588c b-carbonic anhydrase from Mycobacterium tuberculosis and design of not yet synthesized new molecules. J Enzyme Inhib Med Chem 2013. [Epub ahead of print]. doi: 10.3109/ 14756366.2013.800059
- Andrei T, Stancu S. Statistica. 1st ed. Bucharest: Editura Economica; 1995:341
- Golbraikh A, Tropsha A. Beware of q2!. J Mol Graph Model 2002;20:269–76
- Eriksson L, Jaworska J, Worth AP, et al. Methods for reliability and uncertainty assessment and for applicability evaluations of classification and regression-based QSARs. Environ Health Perspect 2003;111:1361–75
- Osten DW. Selection of optimal regression models via cross-validation. J Chemom 1998;2:39–48
- Organization for Economic Co-operation and Development Guidance. Document on the validation of (quantitative) structure-activity relationship QSAR models. OECD Document ENV/JM/MONO(2007)2; 2007
- Worth AP, Aldenberg T, Benjamin I, et al. Quantitative structure--activity relationships – the report and recommendations of ECVAM workshop 52. ATLA 2005;33:155–73
- Tropsha A, Gramatica P, Gombar VK. The importance of being earnest: validation is the absolute essential for successful application and interpretation of QSPR models. Comb Sci 2003;22:69–77
- Weaver S, Gleeson MP. The importance of the domain of applicability in QSAR modeling. J Mol Graph Model 2008;26:1315–26
- NCSS (Statistical Software Delux package), 329 North 1000 East; Kaysville, UT, USA; 2004
- Cook RD. Residuals and influence in regression. New York: Chapman and Hall; 1982
- Draper NR, Smith H. Applied regression analysis. New York: Wiley; 1981
- Everitt B, Landau S, Leese M. Cluster analysis. London: Arnold Press; 2001