Large Number Of Molecular Descriptors Research Articles

Comparative QSAR Modeling for Predicting Anticancer Potency of Imidazo[4,5-b]Pyridine Derivatives Using GA-MLR and BP-ANN Techniques

Background: Prediction of toxicity of imidazo[4,5-b]pyridine derivatives is carried out using GA-MLR and BPANN methods. Objective: A quantitative structure-property relationship (QSPR) was determined based on methods, including genetic algorithm-multiple linear regression (GA-MLR) and backpropagation artificial neural network (BP-ANN). These methods were employed for modeling and predicting the anticancer potency of imidazo[4,5-b]pyridine derivatives. Materials and Methods: A dataset of imidazo[4,5-b]pyridine derivatives was randomly divided into two groups, training and test sets consisting of 75% and 25% of data points, respectively. The optimized conformation of compounds was obtained using the DFT-B3LYP method and 6-31G* basis sets level with Gaussian 09 software. A large number of molecular descriptors were calculated using Dragon software. The QSAR models were optimized using multiple linear regressions (MLR). Results: The most relevant molecular descriptors were obtained using the genetic algorithm (GA) and backward stepwise regression. The predictive powers of the GA-MLR models were studied using leaveone- out (LOO) cross-validation and an external test set. Conclusion: The obtained results of statistical parameters showed the BP-ANN model to have better performance compared to the GA-MLR model. To assess the predictive ability of QSAR models, many statistical terms, such as correlation coefficient (R2), leave-one-out cross-validation (LOOCV), root mean squared error (RMSE), and external and internal validation were used. The results of validation methods demonstrate the QSAR model to be robust and with high predictivity.

Prediction of kinase inhibitors binding modes with machine learning and reduced descriptor sets

Protein kinases are receiving wide research interest, from drug perspective, due to their important roles in human body. Available kinase-inhibitor data, including crystallized structures, revealed many details about the mechanism of inhibition and binding modes. The understanding and analysis of these binding modes are expected to support the discovery of kinase-targeting drugs. The huge amounts of data made it possible to utilize computational techniques, including machine learning, to help in the discovery of kinase-targeting drugs. Machine learning gave reasonable predictions when applied to differentiate between the binding modes of kinase inhibitors, promoting a wider application in that domain. In this study, we applied machine learning supported by feature selection techniques to classify kinase inhibitors according to their binding modes. We represented inhibitors as a large number of molecular descriptors, as features, and systematically reduced these features in a multi-step manner while trying to attain high classification accuracy. Our predictive models could satisfy both goals by achieving high accuracy while utilizing at most 5% of the modeling features. The models could differentiate between binding mode types with MCC values between 0.67 and 0.92, and balanced accuracy values between 0.78 and 0.97 for independent test sets.

Wrapper-based feature selection using regression trees to predict intrinsic viscosity of polymer

This paper introduces different types of regression trees for viscosity property forecasting in polymer solutions. Although regression trees have been extensively used in other fields, they do not have been explored to predict the viscosity. One key issue in the context of materials science is to determine a priori which characteristics must be included to describe the prediction model due to a large number of molecular descriptors is obtained. To deal with this, we propose a wrapper method to select the features based on regression trees. Thus, we use regression trees to evaluate different subsets of attributes and build a model from the subset of features that achieved the minimum error. In particular, the performance of eight regression tree algorithms, including both linear and non-linear models, is evaluated and compared to other forecasting approaches using a dataset composed of 64 polymers and 2962 molecular descriptors. The results show that regression trees with nearest neighbors based local models in leaves predict with high accuracy. Moreover, results have been compared to other forecasting approaches such as multivariate linear regression, neural networks and support vector machines showing remarkable improvements in terms of accuracy.

Structural Relationship Study of Octanol-Water Partition Coefficient of Some Sulfa Drugs Using GA-MLR and GA-ANN Methods.

Sulfonamides (sulfa drugs) are compounds with a wide range of biological activities and they are the basis of several groups of drugs. Quantitative Structure-Property Relationship (QSPR) models are derived to predict the logarithm of water/ 1-octanol partition coefficients (logP) of sulfa drugs. A data set of 43 sulfa drugs was randomly divided into 3 groups: training, test and validation sets consisting of 70%, 15% and 15% of data point, respectively. A large number of molecular descriptors were calculated with Dragon software. The Genetic Algorithm - Multiple Linear Regressions (GA-MLR) and genetic algorithm -artificial neural network (GAANN) were employed to design the QSPR models. The possible molecular geometries of sulfa drugs were optimized at B3LYP/6-31G* level with Gaussian 98 software. The molecular descriptors derived from the Dragon software were used to build a predictive model for prediction logP of mentioned compounds. The Genetic Algorithm (GA) method was applied to select the most relevant molecular descriptors. The R2 and MSE values of the MLR model were calculated to be 0.312 and 5.074 respectively. R2 coefficients were 0.9869, 0.9944 and 0.9601for the training, test and validation sets of the ANN model, respectively. Comparison of the results revealed that the application the GA-ANN method gave better results than GA-MLR method.

QSPR Models to Predict Thermodynamic Properties of Cycloalkanes Using Molecular Descriptors and GA-MLR Method.

Aims and Objectives:QSPR models establish relationships between different types of structural information to their observed properties. In the present study the relationship between the molecular de-scriptors and quantum properties of cycloalkanes is represented.Materials and Methods:Genetic Algorithm (GA) and Multiple Linear Regressions (MLR) were successful-ly developed to predict quantum properties of cycloalkanes. A large number of molecular descriptors were calculated with Dragon software and a subset of calculated descriptors was selected with a genetic algorithm as a feature selection technique. The quantum properties consist of the heat capacity (Cv)/ Jmol-1K-1 entropy(S)/ Jmol-1K-1 and thermal energy(Eth)/ kJmol-1 were obtained from quantum-chemistry technique at the Hartree-Fock (HF) level using the ab initio 6-31G* basis sets.Results:The Genetic Algorithm (GA) method was used to select important molecular descriptors and then they were used as inputs for SPSS software package. The predictive powers of the MLR models were dis-cussed using Leave-One-Out (LOO) cross-validation, leave-group (5-fold)-out (LGO) and external predic-tion series. The statistical parameters of the training and test sets for GA–MLR models were calculated.Conclusion:The resulting quantitative GA-MLR models of Cv, S, and Eth were obtained:[r2=0.950, Q2=0.989, r2ext=0.969, MAE(overall,5-flod)=0.6825 Jmol-1K-1], [r2=0.980, Q2=0.947, r2ext=0.943, MAE(overall,5-flod)=0.5891Jmol-1K-1], and [r2=0.980, Q2=0.809, r2ext=0.985, MAE(overall,5-flod)=2.0284 kJmol-1]. The results showed that the predictive ability of the models was satisfactory, and the constitutional, topological indices and ring descriptor could be used to predict the mentioned properties of 103 cycloalkanes.

Quantitative modeling for prediction of boiling points of phenolic compounds

This work aims to reveal the correlation of the boiling point values of phenolic compounds with their molecular structures using a quantitative structure-property relationship (QSPR) approach. A large number of molecular descriptors have been calculated from molecular structures by the DRAGON software. In this study, all 56 phenolic compounds were divided into two subsets: one for the model formation and the other for external validation, by using the Kennard and Stone algorithm. A four-descriptor model was constructed by applying a multiple linear regression based on the ordinary least squares regression method and genetic algorithm/variables subsets selection. The good of fit and predictive power of the proposed model were evaluated by different approaches, including single or multiple output cross-validations, the Y-scrambling test, and external validation through prediction set. Also, the applicability domain of the developed model was examined using Williams plot. The model shows R² = 0.876, Q²LOO = 0.841, Q²LMO = 0.831 and Q²EXT = 0.848. The results obtained demonstrate that the model is reliable with good predictive accuracy.

QSPR Models for the Prediction of Some Thermodynamic Properties of Cycloalkanes Using GA-MLR Method.

Cycloalkanes have been largely used in the field of medicine, components of food, pharmaceutical drugs, and they are mainly used to produce fuel. In present study the relationship between molecular descriptors and thermodynamic properties such as the standard enthalpies of formation (∆H°f), the standard enthalpies of fusion (∆H°fus), and the standard Gibbs free energy of formation (∆G°f)of the cycloalkanes is represented. The Genetic Algorithm (GA) and multiple linear regressions (MLR) were successfully used to predict the thermodynamic properties of cycloalkanes. A large number of molecular descriptors were obtained with the Dragon program. The Genetic algorithm and backward method were used to reduce and select suitable descriptors. QSPR models were used to delineate the important descriptors responsible for the properties of the studied cycloalkanes. The multicollinearity and autocorrelation properties of the descriptors contributed in the models were tested by calculating the Variance Inflation Factor (VIF), Pearson Correlation Coefficient (PCC) and the Durbin-Watson (DW) statistics. The predictive powers of the MLR models were discussed using Leave-One-Out Cross-Validation (LOOCV) and test set validation methods. The statistical parameters of the training, and test sets for GA-MLR models were calculated. The results of the present study indicate that the predictive ability of the models was satisfactory and molecular descriptors such as: the Functional group counts, Topological indices, GETAWAY descriptors, Constitutional indices, and molecular properties provide a promising route for developing highly correlated QSPR models for prediction the studied properties.

QSAR Models for Predicting Aquatic Toxicity of Esters Using Genetic Algorithm-Multiple Linear Regression Methods.

Esters are of great importance in industry, medicine, and space studies. Therefore, studying the toxicity of esters is very important. In this research, a Quantitative Structure-Activity Relationship (QSAR) model was proposed for the prediction of aquatic toxicity (log 1/IGC50) of aliphatic esters towards Tetrahymena pyriformis using molecular descriptors. A data set of 48 aliphatic esters was separated into a training set of 34 compounds and a test set of 14 compounds. A large number of molecular descriptors were calculated with Dragon software. The Genetic Algorithm (GA) and Multiple Linear Regression (MLR) methods were used to select the suitable descriptors and to generate the correlation models that relate the chemical structural features to the biological activities. The predictive powers of the MLR models are discussed by using Leave-One-Out (LOO) cross-validation and external test set. The best QSAR model is obtained with R2 value of 0.899, Q2 LOO =0.928, F=137.73, RMSE=0.263. The predictive ability of the GA-MLR model with two selected molecular descriptors is satisfactory and it can be used for designing similar group and predicting of toxicity (log 1/IGC50) of ester derivatives.

Application of Ionic Liquids for the Determination of Lipophilicity Parameters Using TLC Method, and QSRR Analysis for the Antipsychotic Drugs.

Reversed-phase liquid chromatography may cause difficulties, especially in the case of basic drugs due to the strong silanophilic interactions in the partition mechanism. Recently, imidazolium-based ionic liquids additives appeared interesting and a convenient solution for suppressing the harmful effect of free residuals of silanol groups, allowing remodeling of the stationary/mobile-phase system, and thus improving the lipophilicity assessment process. The aim of the study was to evaluate the retention behavior of basic antipsychotics using various RP-LC systems, and compare them with data obtained from the modified ionic-liquids RP-TLC systems, and perform the QSRR analysis. Retention and lipophilicity parameters of diverse antipsychotics have been examined in various RP-LC systems. Lipophilicity indices were compared with miscellaneous computed logP values. Furthermore, a large number of molecular descriptors have been computed and compared using various medicinal chemistry software, in order to contribute to the analysis of QSRR. Designated correlation coefficients showed that lipophilicity parameters from TLC systems without [EMIM][BF4] additive correlates very poor with the calculated logPs indices, whereas the indices from the traditional HPLC and TLC systems (with [EMIM][BF4]) were clearly better. Furthermore, QSRR analysis performed for these experimentally obtained lipophilicity parameters showed significant relationships between the retention constants (ROM, logkw) and the in silico calculated physicochemical molecular descriptors. ILs additive may be a significant factor affecting the lipophilicity of basic compounds, thus their use may be favorable in lipophilicity assessment studies. QSRR models with ILs showed that they may be useful in searching/or predicting HPLC/TLC retention parameters for the new/other antipsychotic drugs.

Developing a support vector machine based QSPR model for prediction of half-life of some herbicides

The half-life (t1/2) of 58 herbicides were modeled by quantitative structure–property relationship (QSPR) based molecular structure descriptors. After calculation and the screening of a large number of molecular descriptors, the most relevant those ones selected by stepwise multiple linear regression were used for developing linear and nonlinear models which developed by using multiple linear regression and support vector machine, respectively. Comparison between statistical parameters of linear and nonlinear models indicates the suitability of SVM over MLR model for predicting the half-life of herbicides. The statistical parameters of R2 and standard error for training set of SVM model were; 0.96 and 0.087, respectively, and were 0.93 and 0.092 for the test set. The SVM model was evaluated by leave one out cross validation test, which its result indicates the robustness and predictability of the model. The established SVM model was used for predicting the half-life of other herbicides that are located in the applicability domain of model that were determined via leverage approach.The results of this study indicate that the relationship among selected molecular descriptors and herbicide's half-life is non-linear. These results emphases that the process of degradation of herbicides in the environment is very complex and can be affected by various environmental and structural features, therefore simple linear model cannot be able to successfully predict it.

Development of in Silico Models for Predicting P-Glycoprotein Inhibitors Based on a Two-Step Approach for Feature Selection and Its Application to Chinese Herbal Medicine Screening.

P-glycoprotein (P-gp) is regarded as an important factor in determining the ADMET (absorption, distribution, metabolism, elimination, and toxicity) characteristics of drugs and drug candidates. Successful prediction of P-gp inhibitors can thus lead to an improved understanding of the underlying mechanisms of both changes in the pharmacokinetics of drugs and drug-drug interactions. Therefore, there has been considerable interest in the development of in silico modeling of P-gp inhibitors in recent years. Considering that a large number of molecular descriptors are used to characterize diverse structural moleculars, efficient feature selection methods are required to extract the most informative predictors. In this work, we constructed an extensive available data set of 2428 molecules that includes 1518 P-gp inhibitors and 910 P-gp noninhibitors from multiple resources. Importantly, a two-step feature selection approach based on a genetic algorithm and a greedy forward-searching algorithm was employed to select the minimum set of the most informative descriptors that contribute to the prediction of P-gp inhibitors. To determine the best machine learning algorithm, 18 classifiers coupled with the feature selection method were compared. The top three best-performing models (flexible discriminant analysis, support vector machine, and random forest) and their ensemble model using respectively only 3, 9, 7, and 14 descriptors achieve an overall accuracy of 83.2%-86.7% for the training set containing 1040 compounds, an overall accuracy of 82.3%-85.5% for the test set containing 1039 compounds, and a prediction accuracy of 77.4%-79.9% for the external validation set containing 349 compounds. The models were further extensively validated by DrugBank database (1890 compounds). The proposed models are competitive with and in some cases better than other published models in terms of prediction accuracy and minimum number of descriptors. Applicability domain then was addressed by developing an ensemble classification model to obtain more reliable predictions. Finally, we employed these models as a virtual screening tool for identifying potential P-gp inhibitors in Traditional Chinese Medicine Systems Pharmacology (TCMSP) database containing a total of 13 051 unique compounds from 498 herbs, resulting in 875 potential P-gp inhibitors and 15 inhibitor-rich herbs. These predictions were partly supported by a literature search and are valuable not only to develop novel P-gp inhibitors from TCM in the early stages of drug development, but also to optimize the use of herbal remedies.

QSAR study of chalcone derivatives as anti-Leishmania agents

Quantitative structure activity relationship analyses were used to identify the ideal physicochemical characteristics of potential chalcone derivatives as anti-Leishmania agents. The HyperChem program was used to build chalcone structures and to perform conformational analyses through the semiempirical method followed by the PM3 force field. Dragon calculated a large number of molecular descriptors. Multilinear regression was used for quantitative structure activity relationship modeling. Based on our computational studies, 4 descriptors, SEigv, RDF125v, RDF055u, and O-058, can affect the activity of chalcone derivatives.

Synthesis and QSRR Study for a Series of Phosphoramidic Acid Derivatives

ABSTRACTNew phosphoramidic acid derivatives were synthesized by three different methods in the liquid–liquid and liquid–solid systems. The highest yields were obtained when the liquid–solid system was used. Moreover, for these compounds, the retention factor was predicted, using a robust model obtained for a set of phosphoramidic acid derivatives, which were previously synthesized in our laboratory. For these compounds, the retention factor (k″) was determined by means of the high performance liquid chromatography technique, using a Nucleosil C18 column as a stationary phase and methanol as a mobile phase. The stepwise multiple linear regression (MLR) and partial least square (PLS) methodology were used to investigate the correlation between the retention factor and a number of molecular descriptors for the mentioned compounds. In both procedures, a considerably large number of molecular descriptors have been used. The statistical qualities of MLR and PLS equations have been assessed on the basis of several parameters such as squares of the correlation coefficient (R2 = 0.985), standard error of estimate (SEE = 0.149), Fischer test (F = 257.528), and the cumulative sum of squares of the correlation coefficient R2Y(CUM) = 0.984, the cumulative fraction of the total variation of the Y values that can be predicted for all the extracted principal components by cross‐validation Q2(CUM) = 0.971, respectively, and by the Y‐randomization test. The MLR and PLS equations can be useful for the estimation and comparison of the retention factor, for new synthesized phosphoramidic acid derivatives, using the selected molecular descriptors. © 2013 Wiley Periodicals, Inc. Heteroatom Chem 24:138–145, 2013; View this article online at wileyonlinelibrary.com. DOI 10.1002/hc.21076

A QSAR study for modeling of thyroid receptors β1 selective ligands by application of adaptive neuro-fuzzy inference system and radial basis function

A quantitative structure–activity relationship study of thyroid hormone receptors β1 is described in this paper. We used adaptive neuro-fuzzy inference system (ANFIS) and radial basis function (RBF) methods coupling to genetic algorithm (GA) to predict binding affinity of some ligands with β1 thyroid receptors. A set of 83 selective ligands with known affinity of thyroid receptors β1 (pIC50) were selected, and a large number of molecular descriptors were calculated for each molecule by Dragon. Seven most relevant descriptors were selected by GA-stepwise partial least squares as variable selection tool. The best descriptors (SCBO and EEig08x) and (SCBO, EEig08x, and BEHe1) were applied to train the ANFIS and RBF models, respectively. Then the number and shape of related functions were optimized. The ability and robustness of the GA-ANFIS, GA-RBF, and GA-multiple linear regression (MLR) models in predicting the pIC50 of thyroid receptors β1 are illustrated by internal validation technique of leave one out and also heuristic and randomized techniques as external validation methods. The results have indicated that the proposed models of ANFIS and RBF in this work are superior to MLR method because of generation of simpler models with only two and three descriptors, respectively. Copyright © 2012 John Wiley & Sons, Ltd.

Prediction of skin sensitization with a particle swarm optimized support vector machine.

Skin sensitization is the most commonly reported occupational illness, causing much suffering to a wide range of people. Identification and labeling of environmental allergens is urgently required to protect people from skin sensitization. The guinea pig maximization test (GPMT) and murine local lymph node assay (LLNA) are the two most important in vivo models for identification of skin sensitizers. In order to reduce the number of animal tests, quantitative structure-activity relationships (QSARs) are strongly encouraged in the assessment of skin sensitization of chemicals. This paper has investigated the skin sensitization potential of 162 compounds with LLNA results and 92 compounds with GPMT results using a support vector machine. A particle swarm optimization algorithm was implemented for feature selection from a large number of molecular descriptors calculated by Dragon. For the LLNA data set, the classification accuracies are 95.37% and 88.89% for the training and the test sets, respectively. For the GPMT data set, the classification accuracies are 91.80% and 90.32% for the training and the test sets, respectively. The classification performances were greatly improved compared to those reported in the literature, indicating that the support vector machine optimized by particle swarm in this paper is competent for the identification of skin sensitizers.

Predictive models, based on classification algorithms, for compounds potentially active as mitochondrial ATP-sensitive potassium channel openers

Heart mitochondrial ATP-sensitive potassium channel (mito-K ATP channels) are deeply implicated in the self-defense mechanism of ischemic preconditioning. Therefore, exogenous molecules activating these channels are considered as a promising pharmacological tool to reduce the myocardial injury deriving from ischemia/reperfusion events. In our laboratory, a series of 4-spiro-substituted benzopyran derivatives were earlier synthesized and some of them exhibited anti-ischemic properties. In this study, the above compounds are exploited in order to develop QSAR models, based on classification approaches, capable of discriminating between the ones acting as cardioprotective agents and those that are unable to elicit such a property. Molecules belonging to the whole dataset were subjected to CODESSA and E-Dragon calculations in order to compute a large number of molecular descriptors enabling the construction of classification models. Based on the two program packages used, two different experiments were carried out, with the aim of identify batteries of models to be exploited for designing new cardioprotective agents from libraries of new chemical entities. Both model batteries satisfy the rigorous criteria adopted for the validation, either when tested on the training and test set, according to the most straightforward protocol, and when tested on an additional prediction set. They were proven to ensure successful applications in the field of cardioprotective agent design.

A novel quantitative structure–activity relationship model for prediction of biomagnification factor of some organochlorine pollutants

The biomagnification factor (BMF) is an important property for toxicology and environmental chemistry. In this work, quantitative structure-activity relationship (QSAR) models were used for the prediction of BMF for a data set including 30 polychlorinated biphenyls and 12 organochlorine pollutants. This set was divided into training and prediction sets. The result of diversity test reveals that the structure of the training and test sets can represent those of the whole ones. After calculation and screening of a large number of molecular descriptors, the methods of stepwise multiple linear regression and genetic algorithm (GA) were used for the selection of most important and significant descriptors which were related to BMF. Then multiple linear regression and artificial neural network (ANN) techniques were applied as linear and non-linear feature mapping techniques, respectively. By comparison between statistical parameters of these methods it was concluded that an ANN model, which used GA selected descriptors, was superior over constructed models. Descriptors which were used by this model are: topographic electronic index, complementary information content, XY shadow/XY rectangle and difference between partial positively and negatively charge surface area. The standard errors for training and test sets of this model are 0.03 and 0.20, respectively. The degree of importance of each descriptor was evaluated by sensitivity analysis approach for the nonlinear model. A good results (Q (2) = 0.97 and SPRESS = 0.084) is obtained by applying cross-validation test that indicating the validation of descriptors in the obtained model in prediction of BMF for these compounds.

Affinity prediction on A 1 adenosine receptor agonists: The chemometric approach

In this paper, we are presenting a quantitative-structure–activity relationship (QSAR) study performed on 21 selective A 1 adenosine receptor agonists plus the endogenous substrate, adenosine, so as to identify those predictors which play a key role in describing the binding of the ligand with the A 1 receptor. A large number of molecular descriptors plus a calculated receptor–agonist binding energy and atomic charges were taken into account to derive different QSAR models, using different regression techniques. The results obtained both with linear and nonlinear approaches converge to the selection of the same informative parameters, highlighting the correlation of these descriptors with the biological Response. The evaluation ‘a priori’ of these predictors could therefore represent a useful tool in the screening of large libraries of compounds and in the rational design of new selective agonists.

Quantitative Structure−Activity Relationship of Flavonoid Analogues. 3. Inhibition of p56lck Protein Tyrosine Kinase

Quantitative structure-activity relationship (QSAR) studies on 104 flavonoid derivatives as p56lck protein tyrosine kinase (PTK) inhibitors were performed, using a large number of molecular descriptors calculated by CODESSA software. Multiple linear regression and orthogonalization of descriptors were applied to generate models for the prediction of biological activities for binding flavonoids to PTK. The obtained results demonstrate in detail the importance of electrostatic and quantum chemical descriptors for the interaction of flavonoids with the specific p56lck enzymatic active site environment. In particular, the maximal total interaction for a C-O bond is the most important factor in regression. Use of orthogonalization in regression models provides a valuable improvement for the interpretative and predictive capacity of structure-activity relationships found.

Multivariate Techniques for Parameter Selection and Data Analysis Exemplified by a Study of Pyrethroid Neurotoxicity

AbstractA large number of molecular descriptors (70) were calculated for a series of pyrethroid insecticides using the techniques of computational chemistry. Because the set of descriptors contains considerable redundancy, it is necessary to reduce the size of the data matrix while retaining useful information. Various techniques are available for investigation and reduction of large data sets. Several of these methods have been employed and the results of the different strategies compared. It is concluded that there is no preferred method but that each technique has advantages to offer and is complementary.