External Test Set Validation Research Articles

Development of highly predictive 3D-QSAR CoMSIA models for anthraquinone and acridone derivatives as telomerase inhibitors targeting G-quadruplex DNA telomere

G-quadruplex structures of DNA represent a potentially useful target for anticancer drugs. Telomerase enzyme, involved in immortalization of cancer cells is inhibited by stabilization of G-quadruplex at the ends of chromosomes. Anthraquinone and acridone derivatives are promising G-quadruplex ligands as telomerase inhibitors. So far, optimization of these ligands remained hampered due to the lack of creditable quantitative structure–activity relationships. To understand the structural basis of anthraquinone and acridone derivatives, a predictive 3D-QSAR model has been developed for the first time for telomerase inhibitory activity of G4 ligands, employing comparative molecular similarity indices analysis (CoMSIA). Considering the proposition that the basic nitrogens in these compounds should exist in protonated form at physiological pH the protonated forms of the reported compounds were analyzed and investigated. The QSAR model from conformational template Conf1 exhibited best correlative and predictive properties. The actual predictive abilities of the QSAR model were thoroughly validated through an external validation test set of compounds. The statistics indicate a significantly high prediction power of the best model ( r 2, 0.721), supporting the proposed molecular mechanism of DNA G-quadruplex ligands.

INTERNAL TEST SET (ITS) METHOD: A NEW CROSS-VALIDATION TECHNIQUE TO ASSESS THE PREDICTIVE CAPABILITY OF QSAR MODELS. APPLICATION TO A BENCHMARK SET OF STEROIDS

A new internal cross-validation method is presented for assessing the true predictive capability of QSAR models. The test is general and can be applied in many QSAR/QSPR approaches. In this work, the method is tested on a well-known benchmark set of steroids. In order to make the calculations, Topological Quantum Similarity Indices and Multiple Linear Regression models were considered. warned that these simple aspects of paramount importance for method reliability must be well taken care of. In this work an internal validation method inspired by the concept of the external validation test set is described. The method is general and applicable to many QSAR approaches. It can be easily generalized to any LMO cross-validation protocol but, for the sake of simplicity, the algorithm presented here lies within the context of a simpler approach: the LOO procedure. In its basic formulation, the LOO-CV procedure can be applied when a fixed set of descriptors is considered for a set of n molecules. For each molecule, a model is built considering the experimental property value of the n- 1 remaining compounds and the respective descriptors. Then, this model is used to make a prediction for the molecule left out. The above primary algorithm is named here the kernel of the standard LOO-CV method. This kernel algorithm can be applied using linear or non-linear methods such as Neural Networks 5,6

Predictive QSAR modeling of HIV reverse transcriptase inhibitor TIBO derivatives

Comparative quantitative structure–activity relationship (QSAR) studies have been carried out on tetrahydroimidazo[4,5,1- jk][1,4]benzodiazepine (TIBO) derivatives as reverse transcriptase inhibitors ( n = 70) using topological, structural, physicochemical, electronic and spatial descriptors. The data set was divided into training and test sets using a cluster-based method. Linear models were developed using multiple regression (with stepwise regression, factor analysis and genetic function approximation (GFA) as variable selection tools) and partial least squares (PLS) and combination of factor analysis and partial least squares (FA–PLS). Genetic function approximation (spline) and artificial neural networks (ANN) were used for the development of non-linear models. Using topological and structural descriptors, the best equation was obtained from GFA (spline) based on internal validation ( Q 2 = 0.737), but the model with the best external validation characteristics was obtained with FA–PLS ( R pred 2 = 0.707). When structural, physicochemical, electronic and spatial descriptors were used, the best Q 2 (0.740) value was obtained from GFA (spline) whereas PLS provided the best R pred 2 (0.784) value. When all descriptors were used in combination, the best R pred 2 (0.760) value and the best Q 2 (0.800) value were obtained from ANN and GFA (spline), respectively. The majority of the models satisfied the criteria of external validation recommended by Golbraikh and Tropsha (2002) and the criteria of modified r 2 ( r m 2) values of the test set for external validation as suggested by Roy and Roy (2008). In order to further validate selected models, an external set of 10 TIBO derivatives, which fall within the applicability domain of the models and are not shared with the compounds of the present data set, was taken from a different source, and reverse transcriptase inhibitory activity of these compounds was predicted. Acceptable values of squared correlation coefficients between the observed and predicted values of the external set compounds were obtained from the selected models suggesting true predictive potential of the models.

Combined Ligand and Structure Based Approaches for Narrowing on the Essential Physicochemical Characteristics for CDK4 Inhibition

In the absence of an experimentally determined 3D structure of CDK4 (Cyclin-Dependent Kinase 4), QSARs (Quantitative Structure Activity Relationship) have been explored to rationalize binding affinity in terms of physicochemical and structural parameters. Further, docking on a homology model of CDK4 validated the derived QSARs and predicted the binding mode of this series of inhibitors. Relevant parameters and leave-one-out (LOO) cross-validation (q (2)) as well as an external test set validation (r (2) pred) judged the statistical significance and predictive ability of the models. Docking enabled a better understanding of protein-ligand interaction and provided a mechanistic interpretation in terms of physicochemical characteristics. It identified a unique hydrogen bonding between the imidazole of His-95 and the pyridine nitrogen in the ligand. It rationalized the need for R 2 substituents to be bulky and polar, while the substituent at R 8 to be hydrophobic and comparatively less steric. It also explained why at R 6 a variety of substituents are tolerated and how the presence of methyl at R 5 enhances binding affinity.

Constrained NBMPR analogue synthesis, pharmacophore mapping and 3D-QSAR modeling of equilibrative nucleoside transporter 1 (ENT1) inhibitory activity

Conformationally constrained analogue synthesis was undertaken to aid in pharmacophore mapping and 3D-QSAR analysis of nitrobenzylmercaptopurine riboside (NBMPR) congeners as equilibriative nucleoside transporter 1 (ENT1) inhibitors. In our previous study [ J. Med. Chem. 2003, 46, 831–837], novel regioisomeric nitro-1,2,3,4-tetrahydroisoquinoline conformationally constrained analogues of NBMPR were synthesized and evaluated as ENT1 ligands. 7-NO 2-1,2,3,4-Tetrahydroisoquino-2-yl purine riboside was identified as the analogue with the nitro group in the best orientation at the NBMPR binding site of ENT1. In the present study, further conformational constraining was introduced by synthesizing 5′- O,8-cyclo derivatives. The flow cytometrically determined binding affinities indicated that the additional 5′- O,8-cyclo constraining was unfavorable for binding to the ENT1 transporter. The structure–activity relationship (SAR) acquired was applied to pharmacophore mapping using the PHASE program. The best pharmacophore hypothesis obtained embodied an anti-conformation with three hydrogen-bond acceptors, one hydrophobic center, and two aromatic rings involving the 3′-OH, 4′-oxygen, the NO 2 group, the benzyl phenyl and the imidazole and pyrimidine portions of the purine ring, respectively. A PHASE 3D-QSAR model derived with this pharmacophore yielded an r 2 of 0.916 for four ( 4) PLS components, and an excellent external test set predictive r 2 of 0.78 for 39 compounds. This pharmacophore was used for molecular alignment in a comparative molecular field analysis (CoMFA) 3D-QSAR study that also afforded a predictive model with external test set validation predictive r 2 of 0.73. Thus, although limited, this study suggests that the bioactive conformation for NBMPR at the ENT1 transporter could be anti. The study has also suggested an ENT1 inhibitory pharmacophore, and established a predictive CoMFA 3D-QSAR model that might be useful for novel ENT1 inhibitor discovery and optimization.

QSAR Modeling of Human Serum Protein Binding with Several Modeling Techniques Utilizing Structure−Information Representation

Four modeling techniques, using topological descriptors to represent molecular structure, were employed to produce models of human serum protein binding (% bound) on a data set of 1008 experimental values, carefully screened from publicly available sources. To our knowledge, this data is the largest set on human serum protein binding reported for QSAR modeling. The data was partitioned into a training set of 808 compounds and an external validation test set of 200 compounds. Partitioning was accomplished by clustering the compounds in a structure descriptor space so that random sampling of 20% of the whole data set produced an external test set that is a good representative of the training set with respect to both structure and protein binding values. The four modeling techniques include multiple linear regression (MLR), artificial neural networks (ANN), k-nearest neighbors (kNN), and support vector machines (SVM). With the exception of the MLR model, the ANN, kNN, and SVM QSARs were ensemble models. Training set correlation coefficients and mean absolute error ranged from r2=0.90 and MAE=7.6 for ANN to r2=0.61 and MAE=16.2 for MLR. Prediction results from the validation set yielded correlation coefficients and mean absolute errors which ranged from r2=0.70 and MAE=14.1 for ANN to a low of r2=0.59 and MAE=18.3 for the SVM model. Structure descriptors that contribute significantly to the models are discussed and compared with those found in other published models. For the ANN model, structure descriptor trends with respect to their affects on predicted protein binding can assist the chemist in structure modification during the drug design process.

QSAR Modeling of Carcinogenic Risk Using Discriminant Analysis and Topological Molecular Descriptors

A discriminant analysis model is presented for carcinogenic risk. The data set is obtained from the two-year rodent study FDA/CDER database and was divided into a training set of 1022 organic compounds and an external validation test set of 50 compounds. The model is designed to use as a decision support tool for a defined decision threshold, and is thus a binary discrimination into "high risk" and "low risk" categories. The carcinogenic risk classification is based on the method for estimating human risk from two-year rodent studies developed at the FDA/CDER/ICSAS. The paradigm chosen for this model allows a straightforward risk analysis based on historic information, as well as the computation of coverage, probability and confidence metrics that can further qualify the computed result. The molecular structures were represented as MDL mol files. The molecular structure information was obtained as topological structure descriptors, including atom-type and group-type E-State and hydrogen E-State indices, molecular connectivity chi indices, topological polarity, and counts of molecular features. The MDL QSAR software computed all these descriptors. Furthermore, the discriminant analyses were all performed with the MDL QSAR software. The reported model is based on fifty-three descriptors, using the nonparametric normal kernel method and the Mahalanobis distance to determine proximity. The model performed very well on the fifty compounds of the test set, yielding the following statistics: 76% correctly classified "high risk" (carcinogenic) and 84% correctly classified as "low risk" (non-carcinogenic).

ADME evaluation in drug discovery. 5. Correlation of Caco-2 permeation with simple molecular properties.

The correlations between Caco-2 permeability (logPapp) and molecular properties have been investigated. A training set of 77 structurally diverse organic molecules was used to construct significant QSAR models for Caco-2 cell permeation. Cellular permeation was found to depend primarily upon experimental distribution coefficient (logD) at pH = 7.4, high charged polar surface area (HCPSA), and radius of gyration (rgyr). Among these three descriptors, logD may have the largest impact on diffusion through Caco-2 cell because logD shows obvious linear correlation with logPapp (r=0.703) when logD is smaller than 2.0. High polar surface area will be unfavorable to achieve good Caco-2 permeability because higher polar surface area will introduce stronger H-bonding interactions between Caco-2 cells and drugs. The comparison among HCPSA, PSA (polar surface area), and TPSA (topological polar surface area) implies that high-charged atoms may be more important to the interactions between Caco-2 cell and drugs. Besides logD and HCPSA, rgyr is also closely connected with Caco-2 permeabilities. The molecules with larger rgyr are more difficult to cross Caco-2 monolayers than those with smaller rgyr. The descriptors included in the prediction models permit the interpretation in structural terms of the passive permeability process, evidencing the main role of lipholiphicity, H-bonding, and bulk properties. Besides these three molecular descriptors, the influence of other molecular descriptors was also investigated. From the calculated results, it can be found that introducing descriptors concerned with molecular flexibility can improve the linear correlation. The resulting model with four descriptors bears good statistical significance, n = 77, r = 0.82, q = 0.79, s = 0.45, F = 35.7. The actual predictive abilities of the QSAR model were validated through an external validation test set of 23 diverse compounds. The predictions for the tested compounds are as the same accuracy as the compounds of the training set and significantly better than those predicted by using the model reported. The good predictive ability suggests that the proposed model may be a good tool for fast screening of logPapp for compound libraries or large sets of new chemical entities via combinatorial chemistry synthesis.

E-State Modeling of Fish Toxicity Independent of 3D Structure Information

Topological structure methods are used to model fish toxicity against three classes of organic chemicals. The models were obtained independent of 3D structure information. Further, no mechanism of partitioning was assumed, thus avoiding the problems associated with selection of partitioning system for computation of log P . QSAR models were developed for a set of 92 compounds, including phenols, anilines and substituted aromatic hydrocarbons, yielding excellent statistics: r 2 =0.87, s =0.25 and q 2 =0.85 leave-one-out (LOO), that are better than those reported in the literature. The model is based on molecular connectivity valence chi-1 index [ 1 h v ], the atom type E-State indices for chlorine [ S T (-Cl)] and for ether oxygen [ S T (-O-)], and the maximum hydrogen E-State atom value in a molecule [H max ]. Each of the subgroups was also separately well modeled. The model for the full set is validated through use of external validation test sets and ten-fold cross-validation (repeated three times). The quality of the validation statistics supports the claim that the model may be used for estimation of pLC 50 values for similar molecules. Detailed structure interpretation is given for the descriptors in the model. These four structure descriptors encode influence of molecular context of groups as well as counts of those groups, in addition to molecular skeletal structure.

Prediction of vapour pressures for halogenated diphenyl ether congeners from molecular descriptors.

Polychlorinated diphenyl ethers (PCDE) and polybrominated diphenyl ethers (PBDE) have both been identified as environmental contaminants. The physical properties are important in determining the distribution and fate of organic contaminants in the environment. The purpose of the present investigation was to characterise halogenated diphenyl ethers using computationally derived descriptors, and to develop calibration models for the vapour pressure from published experimental data. Experimental data for vapour pressures were obtained from the literature. The chemical structure of each PCDE and PBDE congener was optimised prior to descriptor generation. The data analysis was performed using principal component analysis (PCA) and partial least squares regression (PLSR). The calibration models were validated with external test sets. All congeners of PCDEs and PBDEs were characterised by 795 molecular descriptors and two principal components could account for about two thirds of the variance within each group. Bilinear calibration models were developed that could explain 99.4% of the variance in the external validation test sets. Vapour pressures were subsequently predicted for all congeners that were adequately described by these calibration models. The type and number of halogen atoms in the molecule were the main factors influencing the vapour pressures of halogen substituted diphenyl ethers, but the variations in substitution pattern was also shown to be a significant factor. The molecular descriptor patterns of halogenated aromatic compounds such as diphenyl ethers can be described and interpreted using principal component analysis (PCA). The major sources of variation in the descriptor spaces for PCDEs and PBDEs are the same as those contributing to the differences in vapour pressure, similar to what has previously been reported for the PCBs. The bilinear calibration models for vapour pressure presented here, has a standard error of prediction that is lower than what is reported as the experimental uncertainty or observed as deviations between experimental investigations. The estimated prediction errors are expected to be within the reported boundaries when the models are applied to new objects within the same molecular descriptor space, and model predictions can hence extend the current database of experimental values. The results from this investigation and others show that the establishment of quantitative structure-property relationships (QSPR) is a viable approach to estimate physical properties for halogenated diphenyl ethers. It is easy to foresee an increased need for using QSPR estimation methods in the future, for evaluation of the environmental fate for organic pollutants. Despite method developments and automation, it is unlikely that laboratory determinations can cope with the pace that new pollutants are identified.

Modeling blood-brain barrier partitioning using the electrotopological state.

The challenging problem of modeling blood-brain barrier partitioning is approached through topological representation of molecular structure. A QSAR model is developed for in vivo blood-brain partitioning data treated as the logarithm of the blood-brain concentration ratio. The model consists of three structure descriptors: the hydrogen E-State index for hydrogen bond donors, HS(T)(HBd); the hydrogen E-State index for aromatic CHs, HS(T)(arom); and the second order difference valence molecular connectivity index, d(2)chi(v) (q(2) = 0.62.) The model for the set of 106 compounds is validated through use of an external validation test set (20 compounds of the 106, MAE = 0.33, rms = 0.38), 5-fold cross-validation (MAE = 0.38, rms = 0.47), prediction of +/- values for an external test set (27/28 correct), and estimation of logBB values for a large data set of 20 039 drugs and drug-like compounds. Because no 3D structure information is used, computation of logBB by the model is very fast. The quality of the validation statistics supports the claim that the model may be used for estimation of logBB values for drug and drug-like molecules. Detailed structure interpretation is given for the structure indices in the model. The model indicates that molecules that penetrate the blood-brain barrier have large HS(T)(arom) values (presence of aromatic groups) but small values of HS(T)(HBd) (fewer or weaker H-Bond donors) and smaller d(2)chi(v) values (less branched molecules with fewer electronegative atoms). These three structure descriptors encode influence of molecular context of groups as well as counts of those groups.

Use of near-infrared spectrometry in control analyses of acrylic fibre manufacturing processes

Near-infrared (NIR) spectroscopy was used in control analyses for sodium thiocyanate and 90 : 10 (w/w) acrylonitrile–vinyl acetate copolymer in spinning solutions used for manufacturing textile acrylic fibres. NIR spectra were recorded in the transflectance mode, using fibre optics and a ceramic disk as reflecting surface. Partial least-squares regression (PLSR) was used to construct models for both analytes, using the test set for external validation. Second-derivative and standard normal variate (SNV) spectra led to simplified and more robust models than conventional spectra. Two PLS1 models for the individual determination of the analytes and a PLS2 model for their joint simultaneous determination are proposed. The errors obtained for the validation samples were 1.6% for NaSCN and 2.6% for the copolymer. The robustness of the proposed calibration models was confirmed by using them in control analyses of samples manufactured several months after the models were constructed.

Multivariate Free-Wilson analysis of α-chymotrypsin inhibitors using PLS

PLS modelling of the inhibitory activity of carboxylic esters in α-chymotrypsin based on Free-Wilson descriptors was described. The resulting PLS model was robust and predictive for designing new potent inhibitors of α-chymotrypsin. D-optimal designs were used for selecting the training and test set for external validation of the PLS model. The contributions of structural fragments to the inhibitory activity were examined by the MLR-like model transformed from the PLS model. Finally, the best candidate compound using the PLS model was synthesized and its inhibitory activity was evaluated.