External Prediction Set Research Articles

Evaluating the role of electron-correlation in the external prediction of the toxicity of nitrobenzenes towards Tetrahymena pyriformis

The role of dynamic interactions between electrons (the electron-correlations) in the prediction of inhibitory growth concentration of alkyl- and halogen-substituted nitrobenzenes against Tetrahymena pyriformis is analyzed through the quantitative structure–activity relationships (QSARs) based on the widely used quantum-chemical descriptors. The advanced semi-empirical and ab initio quantum-mechanical methods are employed for the computation of the descriptors and their electron-correlation contribution. The statistical quality of the proposed QSAR models is assessed through the state-of-the-art external validation parameters via employing an external prediction set of compounds. This study revealed that the models based on the quantum-chemical descriptors incorporating mainly the effect of dynamical electron-correlation are significantly robust and externally predictive, in particular, the models based on the electron-correlation energy are found to be highly reliable. Besides this, the electron-correlation contributions to the orbital energies and to the electrophilicity index are also observed to be significant. The present work emphasizes that the biological activities of chemicals are significantly influenced by the “correlated” motion of electrons in the molecule.

Modeling ready biodegradability of fragrance materials.

In the present study, quantitative structure activity relationships were developed for predicting ready biodegradability of approximately 200 heterogeneous fragrance materials. Two classification methods, classification and regression tree (CART) and k-nearest neighbors (kNN), were applied to perform the modeling. The models were validated with multiple external prediction sets, and the structural applicability domain was verified by the leverage approach. The best models had good sensitivity (internal ≥80%; external ≥68%), specificity (internal ≥80%; external 73%), and overall accuracy (≥75%). Results from the comparison with BIOWIN global models, based on group contribution method, show that specific models developed in the present study perform better in prediction than BIOWIN6, in particular for the correct classification of not readily biodegradable fragrance materials.

Externally predictive single-descriptor based QSPRs for physico-chemical properties of polychlorinated-naphthalenes: Exploring relationships of log SW, log KOA, and log KOW with electron-correlation

Quantitative structure–property relationships (QSPRs), based only on a single-parameter, are proposed for the prediction of physico-chemical properties, namely, aqueous solubility (logSW), octanol–water partition coefficient (logKOW) and octanol–air partition coefficient (logKOA) of polychloronaphthalenes (PCNs) including all the 75 chloronaphthalene congeners. The QSPR models are developed using molecular descriptors computed through quantum mechanical methods including ab-initio as well as advanced semi-empirical methods. The predictivity of the developed models is tested through state-of-the-art external validation procedures employing an external prediction set of compounds. To analyse the role of instantaneous interactions between electrons (the electron-correlation), the models are also compared with those developed using only the electron-correlation contribution of the quantum chemical descriptor. The electron-correlation contribution towards the chemical hardness and the LUMO energy are observed to be the best predictors for octanol–water partition coefficient, whereas for the octanol–air partition coefficient, the total electronic energy and electron-correlation energy are found to be reliable descriptors, in fact, even better than the polarisability. For aqueous solubility of PCNs, the absolute electronegativity is observed to be the best predictor. This work suggests that the electron-correlation contribution of a quantum-chemical descriptor can be used as a reliable indicator for physico-chemical properties, particularly the partition coefficients.

Therapeutic index modeling and predictive QSAR of novel thiazolidin-4-one analogs against Toxoplasma gondii

The main idea of this study was to find predictive quantitative structure–activity relationships (QSAR) for the therapeutic index of 68 thiazolidin-4-one analogs against Toxoplasma gondii. Multivariate adaptive regression spline (MARS) together with Monte-Carlo (MC) sampling was proposed as a reliable descriptor subset selection strategy. Basis functions and knot points are also determined for each selected descriptor using generalized cross validation after frequency analysis. Least squares-support vector regression (LS-SVR) with optimized hyper-parameters was employed as mapping tool due to its promising empirical performance. The models were validated and tested through the use of the external prediction set of compounds, leave-one-out and leave-many-out cross validation methods, applicability domain analysis and Y-randomization. The robustness and accuracy of the QSAR models were confirmed by the satisfactory statistical parameters for the experimentally reported dataset (R2p=0.853, Q2LOO=0.785, R2L20%O=0.742 and r2m=0.715) and low standard error values (RMSEp=0.208, RMSELOO=0.321 and RMSEL20%O=0.376). The comprehensive analysis carried out in the present contribution using the proposed strategy can provide a considerable basis for the design and development of novel drug-like molecules against T.gondii.

Role of exchange and correlation in the real external prediction of mutagenicity: performance of hybrid and meta-hybrid exchange–correlation functionals

Quantum-mechanical exchange and correlation interactions between electrons are quite crucial in deciding molecular geometry and properties. Such electronic interactions can have a significant role in the reliability of a quantitative structure–activity relationship (QSAR) because the biological activities of the chemicals can be described as a function of the molecular structure through the QSARs which are routinely based on the quantum-mechanical molecular descriptors. In this work, we present a detailed analysis of the effect of the quantum-mechanical exchange and correlation on the internal stability and external predictivity of a QSAR model based on the quantum-mechanical molecular descriptors while modeling the mutagenic activity of a set of 51 nitrated-polycyclic aromatic hydrocarbons (PAHs). For this, various molecular descriptors are computed using electronic structure methods such as the Hartree–Fock (HF) method, and density functional theory (DFT) employing only the exchange functionals (HFX, B88), pure exchange and correlation functionals (HFX + LYP, BLYP), hybrid (B3LYP), meta (M06-L), and meta-hybrid (M06, M06-2X) exchange–correlation (XC) functionals. To further analyze the role of electron-correlation, QSAR models are also developed using the descriptors incorporating mainly the effect of electron-correlation. The external predictivity of the developed models is assessed through state-of-the-art external validation parameters employing an external prediction set of compounds. A comparison of the quality of the models developed with the descriptors computed using different electronic structure methods revealed that the exchange interactions are quite critical along with the electron-correlation in modeling the mutagenicity. Notably, for most of the models, electron-correlation based descriptors are found to be highly reliable when computed using the hybrid XC functionals, particularly B3LYP and M06-2X.

Benefits of statistical molecular design, covariance analysis, and reference models in QSAR: a case study on acetylcholinesterase.

Scientific disciplines such as medicinal- and environmental chemistry, pharmacology, and toxicology deal with the questions related to the effects small organic compounds exhort on biological targets and the compounds’ physicochemical properties responsible for these effects. A common strategy in this endeavor is to establish structure–activity relationships (SARs). The aim of this work was to illustrate benefits of performing a statistical molecular design (SMD) and proper statistical analysis of the molecules’ properties before SAR and quantitative structure–activity relationship (QSAR) analysis. Our SMD followed by synthesis yielded a set of inhibitors of the enzyme acetylcholinesterase (AChE) that had very few inherent dependencies between the substructures in the molecules. If such dependencies exist, they cause severe errors in SAR interpretation and predictions by QSAR-models, and leave a set of molecules less suitable for future decision-making. In our study, SAR- and QSAR models could show which molecular sub-structures and physicochemical features that were advantageous for the AChE inhibition. Finally, the QSAR model was used for the prediction of the inhibition of AChE by an external prediction set of molecules. The accuracy of these predictions was asserted by statistical significance tests and by comparisons to simple but relevant reference models.Electronic supplementary materialThe online version of this article (doi:10.1007/s10822-014-9808-1) contains supplementary material, which is available to authorized users.

Towards Global QSAR Model Building for Acute Toxicity: Munro Database Case Study

A series of 436 Munro database chemicals were studied with respect to their corresponding experimental LD50 values to investigate the possibility of establishing a global QSAR model for acute toxicity. Dragon molecular descriptors were used for the QSAR model development and genetic algorithms were used to select descriptors better correlated with toxicity data. Toxic values were discretized in a qualitative class on the basis of the Globally Harmonized Scheme: the 436 chemicals were divided into 3 classes based on their experimental LD50 values: highly toxic, intermediate toxic and low to non-toxic. The k-nearest neighbor (k-NN) classification method was calibrated on 25 molecular descriptors and gave a non-error rate (NER) equal to 0.66 and 0.57 for internal and external prediction sets, respectively. Even if the classification performances are not optimal, the subsequent analysis of the selected descriptors and their relationship with toxicity levels constitute a step towards the development of a global QSAR model for acute toxicity.

Single-descriptor based quantum-chemical QSPRs for physico-chemical properties of polychlorinated-dibenzo-p-dioxins and -dibenzo-furans (PCDD/Fs): Exploring the role of electron-correlation

In this work, we propose and analyze single-descriptor based quantitative structure–property relationships (QSPRs) developed using quantum-chemical descriptors and their electron-correlation contribution, for various physico-chemical properties, namely, aqueous solubility, subcooled liquid vapour pressure, n-octanol/water and n-octanol/air partition coefficients of polychlorinated-dibenzo-p-dioxins (PCDDs) and -dibenzo-furans (PCDFs). The predictivity of the developed QSPRs is examined through external validation procedures employing an external prediction set of chemicals not used in the QSPR model development. From the comparison of different models, it is observed that the total energy and the mean polarizability of a molecule are highly significant in determining the predictivity of a QSPR, whereas dipole moment is found to be a poor descriptor in the models developed for all the properties analyzed. Besides these, the electron density based descriptors, namely, absolute electronegativity and electrophilicity index, were also observed to be highly influential in determining the external prediction for partition coefficients.

Electron-correlation based externally predictive QSARs for mutagenicity of nitrated-PAHs in Salmonella typhimurium TA100

In quantitative modeling, there are two major aspects that decide reliability and real external predictivity of a structure–activity relationship (SAR) based on quantum chemical descriptors. First, the information encoded in employed molecular descriptors, computed through a quantum-mechanical method, should be precisely estimated. The accuracy of the quantum-mechanical method, however, is dependent upon the amount of electron-correlation it incorporates. Second, the real external predictivity of a developed quantitative SAR (QSAR) should be validated employing an external prediction set. In this work, to analyze the role of electron-correlation, QSAR models are developed for a set of 51 ubiquitous pollutants, namely, nitrated monocyclic and polycyclic aromatic hydrocarbons (nitrated-AHs and PAHs) having mutagenic activity in TA100 strain of Salmonella typhimurium. The quality of the models, through state-of-the-art external validation procedures employing an external prediction set, is compared to the best models known in the literature for mutagenicity. The molecular descriptors whose electron-correlation contribution is analyzed include total energy, energy of HOMO and LUMO, and commonly employed electron-density based descriptors such as chemical hardness, chemical softness, absolute electronegativity and electrophilicity index. The electron-correlation based QSARs are also compared with those developed using quantum-mechanical descriptors computed with advanced semi-empirical (SE) methods such as PM6, PM7, RM1, and ab initio methods, namely, the Hartree-Fock (HF) and the density functional theory (DFT). The models, developed using electron-correlation contribution of the quantum-mechanical descriptors, are found to be not only reliable but also satisfactorily predictive when compared to the existing robust models. The robustness of the models based on descriptors computed through advanced SE methods, is also observed to be comparable to those developed with the electron-correlation based descriptors. The work emphasizes that the correlation-energy can serve as a reliable descriptor to explore the origin of biological activities at the level of electron-dynamics.

Prediction of glucagon receptor antagonist activities of some substituted imidazoles using combined radial basis function neural network and density functional theory

QSAR study of human glucagon receptor (HGR) ligands has been made with the help of quantum descriptors, such as energy of HOMO, energy of LUMO, softness, hardness using combination of principal component analysis, and radial basis function artificial neural network (ANN). Quantum descriptors have been calculated via the DFT-B3LYP method, with the basis set 6-311G. The developed neural network QSAR model outperformed the principal component regression model in both fitting and predictive abilities. ANN analysis indicated that the estimated activities were in total agreement with the experimentally observed values (R2 = 0.869, RMSD = 0.186; predictive Q2 = 0.732, RMSEcv = 0.346). The developed models were further examined by means of an external prediction set. The modeling study also reflected the important role of quantum properties of molecules when they interact with the target, HGR. The developed neural network model is expected to be useful in the rational design of new chemical entities as ligands of HGR and also for directing the synthesis of potent molecules in the future.

Externally predictive quantitative modeling of supercooled liquid vapor pressure of polychlorinated-naphthalenes through electron-correlation based quantum–mechanical descriptors

For predicting physico-chemical properties related to environmental fate of molecules, quantitative structure–property relationships (QSPRs) are valuable tools in environmental chemistry. For developing a QSPR, molecular descriptors computed through quantum–mechanical methods are generally employed. The accuracy of a quantum–mechanical method, however, rests on the amount of electron-correlation estimated by the method. In this work, single-descriptor QSPRs for supercooled liquid vapor pressure of chloronaphthalenes and polychlorinated-naphthalenes are developed using molecular descriptors based on the electron-correlation contribution of the quantum–mechanical descriptor. The quantum–mechanical descriptors for which the electron-correlation contribution is analyzed include total-energy, mean polarizability, dipole moment, frontier orbital (HOMO/LUMO) energy, and density-functional theory (DFT) based descriptors, namely, absolute electronegativity, chemical hardness, and electrophilicity index. A total of 40 single-descriptor QSPRs were developed using molecular descriptors computed with advanced semi-empirical (SE) methods, namely, RM1, PM7, and ab intio methods, namely, Hartree–Fock and DFT. The developed QSPRs are validated using state-of-the-art external validation procedures employing an external prediction set. From the comparison of external predictivity of the models, it is observed that the single-descriptor QSPRs developed using total energy and correlation energy are found to be far more robust and predictive than those developed using commonly employed descriptors such as HOMO/LUMO energy and dipole moment. The work proposes that if real external predictivity of a QSPR model is desired to be explored, particularly, in terms of intra-molecular interactions, correlation-energy serves as a more appropriate descriptor than the polarizability. However, for developing QSPRs, computationally inexpensive advanced SE methods such as PM7 can be more reliable than the expensive ab inito methods.

Rough set based wavelength selection in near-infrared spectral analysis

Rough set based procedure was proposed as a new methodology to select component-specific wavelengths for near-infrared (NIR) spectral analysis. Information gain (IG) was employed to regulate the size of the discernibility matrix and decrease the memory requirements of rough set based reduction. This procedure involved submitting the resulting subsets of wavelengths to the analytical models in question. The utility of this method was illustrated by an analysis of classification models for phenylalanine (Phe) in plasma. The wavelength selection algorithm was compared with correlation based feature selection (CRFS) method and consistency based feature selection (CSFS) approach. Model fit was assessed using 10-fold cross-validation (10-fold CV) and leave-one out (LOO) approach. The predictability of the model was evaluated by an external prediction set. Furthermore, another two NIR data sets, obtained from the published literatures, were used to develop the quantitative models and validate the rough set based wavelength selection method. This study demonstrates conclusively that reducts of rough set could preserve the spectra–structure relationship and provide reliable model variables for NIR analysis. The results also indicate that rough set algorithm may hold promise for application as an additional feasible technique to NIR band assignment. As a fast, simple and noninvasive measurement, it is hopeful to find a clinical use in the diagnosis of unusual Phe elevation with further research.

The Development of Models Based on Linear and Nonlinear Multivariate Methods to Predict ADME/PK Properties Using Physicochemical Properties of Kinase, Protease Inhibitors, and GPCR Antagonists.

Oral bioavailability of a drug compound is the significant property for potential drug candidates. Measuring this property can be costly and time-consuming. Quantitative structure-property relationships (QSPRs) are used to estimate the percentage of oral bioavailability, and they are an attractive alternative to experimental measurements. A data set of 217 drug and drug-like compounds with measured values of the percentage of oral bioavailability taken from the small molecule ChemBioBase database was used to develop and test a QSPR model. Descriptors were calculated for the compounds using Codessa 2.1 tool. Nonlinear general regression neural network model was generated using the DTREG predictive modeling program software. The calculated percentage of oral bioavailability model performs well, with root-mean-square (rms) errors of 4.55% oral bioavailability units for the training set, 14.32% oral bioavailability units for the test set, and 19.12% oral bioavailability units for the external prediction set. Given the structural diversity and bias of the data set, this is a good first attempt at modeling oral bioavailability using QSPR methods. The model can be used as a potential virtual screen or property estimator. With a larger data supply less biased toward the high end values of the percentage of oral bioavailability, a more successful model could likely be developed.

A novel two-step QSAR modeling work flow to predict selectivity and activity of HDAC inhibitors

A two-step modeling approach was employed to study the selectivity and activity of histone deacetylase inhibitors. First, according to the activity difference against HDAC1 and HDAC6, a binary classification model was established to classify two kinds of inhibitors. Then two continuous models were built for each subclass to predict the activity value of HDAC1 and HDAC6 inhibitors. The three models were all built with the GA-kNN method combined with dragon descriptors. They were external validated by using external prediction set and Y-randomization test. The highly predictive models were generated for all three data sets. For the classification model, the classification accuracies of the models were as high as 100% for the external test set. For HDAC1and HDAC6 inhibitor consecutive models, external R2 values are 0.947 and 0.911, respectively. The results proved the reliability of these models. All models were used to screen 1000 compounds included in PubMed dataset. Virtual screening resulted in 8 and 13 structurally unique consensus hits that were considered novel putative HDAC1 and HDAC6 inhibitors, respectively.

Quantitative Structure–Property Relationship Prediction of Gas Heat Capacity for Organic Compounds

In the present work, a quantitative structure–property relationship study is performed to predict gas heat capacity for a structurally wide variety of organic compounds using the genetic function approximation (GFA) and the adaptive neuro-fuzzy inference system (ANFIS) methods. The simple proposed models contain only three descriptors calculated solely from the molecular structure of compounds which are 3D-independent descriptors. The models were validated by an external prediction set. Good results were obtained from both models which get the squared correlation coefficients of 0.996 and 0.997 for GFA and ANFIS, respectively. This study discloses enhanced correlations of the heat capacity of gases with their molecular structures, wherein the influence of the size of molecules is found to predominate.

Multi-wavelength high-performance liquid chromatographic fingerprints and chemometrics to predict the antioxidant activity of Turnera diffusa as part of its quality control

The determination of the antioxidant activity of Turnera diffusa using partial least squares regression (PLSR) on chromatographic data is presented. The chromatograms were recorded with a diode array detector and, for each sample, an enhanced fingerprint was constructed by compiling into a single data vector the chromatograms at four wavelengths (216, 238, 254 and 345nm). The wavelengths were selected from a contour plot, in order to obtain the greater number of peaks at each of the wavelengths. A further pretreatment of the data that included baseline correction, scaling and correlation optimized warping was performed. Optimal values of the parameters used in the warping were found by means of simplex optimization. A PLSR model with four latent variables (LV) explained 52.5% of X variance and 98.4% of Y, with a root mean square error for cross validation of 6.02. To evaluate its reliability, it was applied to an external prediction set, retrieving a relative standard error for prediction of 7.8%. The study of the most important variables for the regression indicated the chromatographic peaks related to antioxidant activity at the used wavelengths.

Experimental and QSPR Studies on the Effect of Ionic Surfactants on n‐Decane–Water Interfacial Tension

AbstractA quantitative structure property relationship approach was performed to find the relation between the surfactant structure and its effect on water–oil interfacial tension. As a result, a new database has been developed measuring the interfacial tension between n‐decane as the model oil and different aqueous solutions of some ionic compounds. In spite of other reports we selected surfactants by a scientific method that covers all structural information. Twenty four different compounds were selected by the principle component analysis method and their interfacial tensions were measured at their critical micelle concentrations. The geometrical optimization of surfactants was performed at the B3LYP/6‐311G⃰⃰ level and quantum chemical and structural descriptors were calculated using relevant computer software programs. The best fitted descriptors were selected using the variable selection of the genetic algorithm (GA‐MLR). The predictive test was performed for an external prediction set of 6 compounds, chosen out of 24 compounds. The resulted GA‐MLR model can reasonably predict the interfacial tension using only three selected descriptors. The deviation between predicted and measured values was found to be <7%.

Quantitative structure-property relationship (QSPR) model for predicting acidities of ketones

Ketones are one of the most common functional groups, and ketone-containing compounds are essential in both the nature and the chemical sciences. As such, the acidities (pKa) of ketones provide valuable information for scientists to screen for biological activities, to determine physical properties or to study reaction mechanisms. Direct measurements of pKa of ketones are not readily available due to their extremely weak acidity. Hence, a quantitative structure-property relationship (QSPR) model that can predict the acidities of ketones and their acidity order is highly desirable. The establishment of an acidity scale in dimethyl sulfoxide (DMSO) solution by Bordwell et al. made such an effort possible. By utilizing the pKa values of forty-eight ketones determined in DMSO as the training set, a QSPR model for predicting acidities of ketones was built by stepwise multiple linear regression analysis. The established model showed statistical significance and predictive power (r2 = 0.91, q2 = 0.86, s = 1.42). Moreover, the QSPR model also gave reasonable acidity predictions for five ketones in an external prediction set that were not included in the model generation phase (r2 = 0.92, s = 1.618). Overall, the reported QSPR model for predicting acidities of ketones provides a useful tool for both biologists and chemists in understanding the biophysical properties and reaction rates of different classes of ketones.

A classification study of human β 3-adrenergic receptor agonists using BCUT descriptors

Experimental EC(50)s for 202 human β(3)-AR agonists are used to develop classification models as a potential screening tool for a large library of target compounds before synthesis. A variable selection approach from random forests (VS-RF) is used to extract the structural information most relevant to the human β(3)-AR activation properties of the collected data set. The obtained results indicate that the VS-RF method can be used for variable selection with smallest sets of non-redundant descriptors with highly predictive accuracy (Q (ex)% = 96% for the external prediction set). Thus, the proposed VS-RF models should be helpful for screening of potential human β(3)-AR agonists before chemical synthesis in drug development.

QSAR models for CXCR2 receptor antagonists based on the genetic algorithm for data preprocessing prior to application of the PLS linear regression method and design of the new compounds using in silico virtual screening.

The CXCR2 receptors play a pivotal role in inflammatory disorders and CXCR2 receptor antagonists can in principle be used in the treatment of inflammatory and related diseases. In this study, quantitative relationships between the structures of 130 antagonists of the CXCR2 receptors and their activities were investigated by the partial least squares (PLS) method. The genetic algorithm (GA) has been proposed for improvement of the performance of the PLS modeling by choosing the most relevant descriptors. The results of the factor analysis show that eight latent variables are able to describe about 86.77% of the variance in the experimental activity of the molecules in the training set. Power prediction of the QSAR models developed with SMLR, PLS and GA-PLS methods were evaluated using cross-validation, and validation through an external prediction set. The results showed satisfactory goodness-of-fit, robustness and perfect external predictive performance. A comparison between the different developed methods indicates that GA-PLS can be chosen as supreme model due to its better prediction ability than the other two methods. The applicability domain was used to define the area of reliable predictions. Furthermore, the in silico screening technique was applied to the proposed QSAR model and the structure and potency of new compounds were predicted. The developed models were found to be useful for the estimation of pIC50 of CXCR2 receptors for which no experimental data is available.