Nonlinear Quantitative Structure-property Relationship Research Articles

Chemical structure-based models for prediction of density of ammonium and phosphonium-based deep eutectic solvents

In this work, linear and nonlinear quantitative structure–property relationship (QSPR) models are proposed to predict the density of deep eutectic solvents (DESs). A large experimental database of 2005 density data belonging to chloride and bromide-based DESs with ammonium and phosphonium-based cations and various hydrogen bond donors (HBDs) was collected from literatures. Using a modified particle swarm optimization (MPSO) variable-selection approach, subsets of relevant descriptors were selected to relate the density of DESs to their corresponding cation and HBD structures. To develop the nonlinear models, an adaptive neuro-fuzzy inference system (ANFIS) method was used with the same inputs and outputs of linear models. The squared correlation coefficient (R2) of linear models for chloride and bromide-based DESs are 0.897 and 0.960, while the density of DESs with anion part of chloride and bromide is predicted by ANFIS with the R2 of 0.956 and 0.980, respectively. The internal and external validations of the proposed models indicate that these models can be used for prediction of DESs densities with high accuracy and reliability.

QSPR modelling for intrinsic viscosity in polymer–solvent combinations based on density functional theory

ABSTRACT Linear and nonlinear quantitative structure–property relationship (QSPR) models were developed based on a dataset with 65 polymer-solvent combinations. Seven quantum chemical descriptors, dipole moment, hardness, chemical potential, electrophilicity index, total energy, HOMO and LUMO orbital energies, were calculated with density functional theory at the B3LYP/6-31 G(d) level for polymers and solvents. Considering the strong correlation between intrinsic viscosity and weight, size, shape as well as topological structure of polymers and solvents, topological descriptors were also applied in this work. Meanwhile, the most appropriate polymer structure representation was investigated by considering 1–5 monomeric repeating units. The molecular descriptors were first screened by using the genetic algorithms-multiple linear regression (GA-MLR), with coefficient of determinations () of 0.78 and 0.83 for the training set and the prediction set, respectively. The support vector machine model (SVM) model based on the selected descriptors subset showed a value of 0.95 for the training set and 0.93 for the prediction set. All statistical results suggest that the established QSPR models have good predictability. Furthermore, a new test set obtained from the literature was used for further validation. The values were 0.81 for the MLR model and 0.90 for the SVM model.

Prediction of corrosion inhibition efficiency of pyridines and quinolines on an iron surface using machine learning-powered quantitative structure-property relationships

Linear and non-linear quantitative structure–property relationship (QSPR) models were developed to predict corrosion inhibition efficiency for a series of 41 pyridine and quinoline N-heterocycles. Out of 20 physicochemical and quantum chemical variables related to the surface adsorption behaviour of the inhibitors, consensus models were constructed using the genetic algorithm-partial least squares (GA-PLS) and genetic algorithm-artificial neural network (GA-ANN) methods. The consensus GA-PLS model comprised of eight variables (exponential of the calculated adsorption energy, LUMO, van der Waals surface area and volume, polarizability, electron affinity, electrophilicity, electron donor capacity) exhibited an %RMSECV of 16.5% and mean %RMSE of 14.9%. Such a model moderately captured the complex relationships between inhibition efficiency and the quantum chemical variables. The consensus GA-ANN model comprised of nine input variables (exponential of the calculated adsorption energy, HOMO, LUMO, HOMO-LUMO Gap, electronegativity, softness, electrophilicity, electron donor capacity and N atomic charge) exhibited an %RMSECV of 16.7 ± 2.3% and mean %RMSE (training/testing/validation) of 8.8%, performing better than its linear counterpart in terms of predictive ability. Both models revealed the importance of adsorption to the metal surface, and electronic parameters quantifying electron acceptance/donation to/from the iron surface, suggesting key corrosion inhibition design principles.

Prediction of the half-lives of polychlorinated biphenyls based on the IEF-PCM calculations

Three-dimensional structures of 62 polychlorinated biphenyl (PCB) congeners were optimized with the integral equation formalism polarizable continuum model (IEF-PCM) in combination with the density functional theory (DFT) method at 6-31G(d) level. By applying support vector machine (SVM) algorithm, a nonlinear quantitative structure–property relationship (QSPR) model was built to predict half-lives (log [Formula: see text]) of 62 PCBs in juvenile rainbow trout. The optimal SVM model based on the parameters [Formula: see text] of 854.721 and [Formula: see text] of 0.0565 produces the root-mean-square (rms) errors of 0.0352 for the training set and 0.0446 for the test set, which are less than that of the previous models reported. The results suggest that it is feasible to build SVM models for the half-lives of PCBs with IEF-PCM and B3LYP/6-31G(d) for deriving structural descriptors.

Prediction of 13C NMR Chemical Shifts of Quinolone Derivatives Based on DFT Calculations

At present, there exists subjectivity in selecting descriptor sets for the quantitative structure property relationship (QSPR) models. A complete set is perfect, in which there is no any element redundant or needed to be added. This paper reports the complete sets of descriptors used to develop QSPR models for 13C NMR chemical shifts (δC parameters) of carbon atoms in quinolone derivatives. These descriptors in the complete sets used are calculated by applying the PBE1PBE functional of density functional theory (DFT) and the 6-311G(2d,2p) basis set. The multiple linear regression (MLR) technique and the support vector machine (SVM) algorithm are, respectively, used to develop linear and nonlinear QSPR models for δC parameters. The four QSPR models have the root mean square (RMS) errors less than 2.0 ppm, which approximately equal one fourth of the errors from the previous model. Further, our models have more samples in the test sets and less descriptors in the models. These results suggest that our four models of δC parameters have better statistical qualities. The feasibility of applying complete sets of descriptors to develop QSPR models for 13C NMR chemical shifts is demonstrated.

Prediction of 13C NMR chemical shifts of quinolone derivatives based on DFT calculations

At present, there exists subjectivity in selecting descriptor sets for the quantitative structure property relationship (QSPR) models. A complete set is perfect, in which there is no any element redundant or needed to be added. This paper reports the complete sets of descriptors used to develop QSPR models for 13C NMR chemical shifts (δC parameters) of carbon atoms in quinolone derivatives. These descriptors in the complete sets used are calculated by applying the PBE1PBE functional of density functional theory (DFT) and the 6-311G(2d,2p) basis set. The multiple linear regression (MLR) technique and the support vector machine (SVM) algorithm are, respectively, used to develop linear and nonlinear QSPR models for δC parameters. The four QSPR models have the root mean square (RMS) errors less than 2.0 ppm, which approximately equal one fourth of the errors from the previous model. Further, our models have more samples in the test sets and less descriptors in the models. These results suggest that our four models of δC parameters have better statistical qualities. The feasibility of applying complete sets of descriptors to develop QSPR models for 13C NMR chemical shifts is demonstrated.

Pencil and Paper Estimation of Hansen Solubility Parameters.

Simple procedures to estimate Hansen solubility parameter (HSP) components from structural formulas are investigated. The best results are obtained using a simple relationship with molar volume and refractivity for the dispersion component, and using additivity models based on tailored fragments specifically designed for the polar and hydrogen bonding components. Despite large errors for some classes of chemicals, including small inorganic molecules, ionic liquids, and high halogen compounds, these models yield average absolute deviations from reference on par with state-of-the-art models and lower than reported using molecular dynamics simulations or nonlinear quantitative structure–property relationship models based on a limited set of quantum chemical descriptors. In contrast to group contribution methods that are either more restricted in scope or heavily parameterized, they are thoroughly validated and very easy to apply. Furthermore, the errors observed are easy to rationalize and may usually be anticipated. This work sheds light on some limitations inherent to pure additivity approaches for HSP prediction and provides a first step toward better models. A Python script implementing the procedure and the fully detailed results are provided as the Supporting Information.

Understanding and predicting the diffusivity of organic compounds in polydimethylsiloxane material for passive sampler applications using a simple quantitative structure-property relationship model.

The diffusivity of 145 compounds in polydimethylsiloxane (PDMS) material was determined in the laboratory using a film stacking technique. The results were pooled with available literature data, providing a final data set of 198 compounds with diffusivity (DPDMS ) spanning over approximately 5 log units. The principal variables controlling the diffusivity of penetrants were investigated by comparing DPDMS within and between different homologous series. The dipole moment, molecular size, and flexibility of penetrants appear to be the most prevalent factors controlling a compound's diffusivity. A nonlinear quantitative structure-property relationship is proposed using as predicting variables the molecular volume, the number of rotatable bonds, the topological polar surface area, and the number of O and N atoms. The final relationship has a correlation coefficient of R2 = 0.81 and a mean absolute error of 0.26 m2 s-1 (log unit), approaching the average error for the experimentally determined values (0.12 m2 s-1 ). The model, based on a heuristic approach, is ready for use by analytical chemists with no specific background in theoretical chemistry (notably for passive sampler development). Environ Toxicol Chem 2018;37:1291-1300. © 2018 SETAC.

The QSPR models to predict the solubility of CO2 in ionic liquids based on least-squares support vector machines and genetic algorithm-multi linear regression

Global warming is an issue of high concern which is mostly caused by growing concentrations of carbon dioxide in the atmosphere. Many novel technologies offer different solutions to decline carbon dioxide emissions to the environment. Ionic liquids (ILs) are counted to be highly promising media for CO2 capture in the near future. Due to costly nature of ionic liquids and time consuming laboratory research procedures, modeling and prediction of solubility of CO2 based on the structure of ILs are highly required. Some studies on this field demonstrate the relationship between the structure and the CO2 absorption capacity of ILs. One of the modeling approaches for stating this relationship is quantitative structure-property relationship (QSPR). In this work, an efficient approach based on the combination of genetic algorithm-multi linear regression (GA-MLR) and least-squares support vector machines (LS-SVM) was utilized to build a nonlinear QSPR model. The nonlinear model can give very satisfactory prediction results: the square of correlation coefficient (R2) and the root mean square error (RMSE) were 0.962 and 0.015, respectively for the whole dataset. In addition, another QSPR model, multi-linear regression (MLR), was also implemented and R2and RMSE were 0.876 and 0.027, respectively. The results demonstrate that the LS-SVM model drastically enhances the ability of prediction in QSPR studies and is superior to MLR one.

A genetic programming-based QSPR model for predicting solubility parameters of polymers

In this study, linear and nonlinear quantitative structure-property relationship (QSPR) models, respectively called the multiple linear regression based QSPR (MLR-QSPR) model and the genetic programming based QSPR (GP-QSPR) model, were built to predict the solubility parameters of polymers with structure –(C1H2–C2R3R4)–, as function of some constitutional, topological and quantum chemical descriptors. The results from the internal validation analysis indicated that the GP-QSPR model has better goodness of fit statistics. The external and overall validation measures also confirmed that the GP-QSPR model significantly outperforms the MLR-QSPR model in terms of some performance metrics over the same testing data set, and that genetic programming has good potential to obtain more accurate models in QSPR studies.

In Silico Predictions of Human Skin Permeability using Nonlinear Quantitative Structure-Property Relationship Models.

Predicting human skin permeability of chemical compounds accurately and efficiently is useful for developing dermatological medicines and cosmetics. However, previous work have two problems; 1) quality of databases used, and 2) methods for prediction models. In this paper, we attempt to solve these two problems. We first compile, by carefully screening from the literature, a novel dataset of chemical compounds with permeability coefficients, measured under consistent experimental conditions. We then apply machine learning techniques such as support vector regression (SVR) and random forest (RF) to our database to develop prediction models. Molecular descriptors are fully computationally obtained, and greedy stepwise selection is employed for descriptor selection. Prediction models are internally and externally validated. We generated an original, new database on human skin permeability of 211 different compounds from aqueous donors. Nonlinear SVR achieved the best performance among linear SVR, nonlinear SVR, and RF. The determination coefficient, root mean square error, and mean absolute error of nonlinear SVR in external validation were 0.910, 0.342, and 0.282, respectively. We provided one of the largest datasets with purely experimental log kp and developed reliable and accurate prediction models for screening active ingredients and seeking unsynthesized compounds of dermatological medicines and cosmetics.

Linear and nonlinear quantitative structure-property relationship modelling of skin permeability

In this work, quantitative structure-property relationship (QSPR) models were developed to estimate skin permeability based on theoretically derived molecular descriptors and a diverse set of experimental data. The newly developed method combining modified particle swarm optimization (MPSO) and multiple linear regression (MLR) was used to select important descriptors and develop the linear model using a training set of 225 compounds. The adaptive neuro-fuzzy inference system (ANFIS) was used as an efficient nonlinear method to correlate the selected descriptors with experimental skin permeability data (log Kp). The linear and nonlinear models were assessed by internal and external validation. The obtained models with three descriptors show good predictive ability for the test set, with coefficients of determination for the MPSO-MLR and ANFIS models equal to 0.874 and 0.890, respectively. The QSPR study suggests that hydrophobicity (encoded as log P) is the most important factor in transdermal penetration.

Prediction of Bovine Serum Albumin-Water Partition Coefficients of a Wide Variety of Neutral Organic Compounds by Means of Support Vector Machine.

Support vector machine (SVM) was used to develop a quantitative structure property relationship (QSPR) model that correlates molecular structures to their bovine serum albumin water partition coefficients (KBSA/W ). The performance and predictive aptitude of SVM are considered and compared with other methods such as multiple linear regression (MLR) and artificial neural network (ANN) methods. A set of 83 natural organic compounds and drugs were selected and suitable sets of molecular descriptors were calculated. Genetic algorithm (GA) was used to select important molecular descriptors, and linear and nonlinear models were applied to correlate the selected descriptors with the experimental values of log KBSA/W . The correlation coefficients, R, between experimental and predicted log KBSA/W for the validation set by MLR, ANN and SVM are 0.951, 0.986 and 0.991, respectively. Results obtained document the reliability and good predictability of the nonlinear QSPR model to predict partition coefficients of organic compounds. Comparison between the values of statistical parameters demonstrates that the predictive ability of the SVM model is comparable or superior to those obtained by MLR and ANN.

Improved QSPR generalized interaction parameters for the nonrandom two-liquid activity coefficient model

Phase equilibrium properties are essential in developing and optimizing numerous processes. The objective of this work is the application of a theory-framed quantitative structure-property relationship (QSPR) modeling approach to provide a priori vapor–liquid equilibrium (VLE) predictions. For this purpose, we apply the nonrandom two-liquid (NRTL) activity coefficient model to describe the phase behavior, and then use the QSPR methodology to generalize the substance-specific parameters of the model. Generalizing the parameters of a proven phase behavior model, such as the NRTL, will minimize the need for acquiring costly VLE experimental data for the systems of interest.The newly developed generalized NRTL-QSPR activity coefficient model constitutes a significant improvement over our previous generalization of the NRTL model. Specifically, an internally consistent generalization is provided for the NRTL interaction parameters using a more extensive database involving 578 binary systems. A non-linear QSPR model was developed for the NRTL parameters, where evolutionary algorithms combined with artificial neural networks were used to perform molecular descriptor reduction. The model predicts pressure and temperature of a binary VLE system within 6% and 0.6% average absolute deviation (AAD), respectively. Further, the generalized NRTL phase behavior predictions show a significant improvement over to the group contribution method, Universal Functional Activity Coefficient model (UNIFAC), which resulted in 9% AAD for pressure predictions.

Molecular modeling of the standard state heat of formation

The standard heat of formation is a basic thermophysical property required in determining enthalpies of reaction and in thermodynamic stability analyses. Further, the enthalpies of formation are important in investigating bond energies, resonance energies and the nature of chemical bonds. Therefore, the development of accurate structure-based estimation methods for large varieties of chemical species is greatly beneficial in enhancing capability in process and product development.In this work, quantitative structure–property relationship (QSPR) models were developed for a structurally diverse DIPPR dataset of standard heats of formation comprising 1765 pure compounds involving 82 chemical classes. We have employed both linear and nonlinear QSPR modeling techniques. The linear approach involves the use of constricted binary particle swarm optimization (BPSO) for feature selection and multiple-linear regression. In the nonlinear approach, the optimum network architecture and its associated inputs are identified using a wrapper-based feature selection algorithm combining differential evolution and artificial neural networks. Model predictions for the root-mean-square error of the BPSO and nonlinear approaches were 138 and 97kJ/mol, respectively.

A non-linear structure–property model for octanol–water partition coefficient

Octanol–water partition coefficient (Kow) is an important thermodynamic property used to characterize the partitioning of solutes between an aqueous and organic phase and has importance in such areas as pharmacology, pharmacokinetics, pharmacodynamics, chemical production and environmental toxicology. We present a non-linear quantitative structure–property relationship model for determining Kow values of new molecules in silico. A total of 823 descriptors were generated for 11,308 molecules whose Kow values are reported in the PhysProp dataset by Syracuse Research. Optimum network architecture and its associated inputs were identified using a wrapper-based feature selection algorithm that combines differential evolution and artificial neural networks. A network architecture of 50-33-35-1 resulted in the least root-mean squared error (RMSE) in the training set. Further, to improve on single-network predictions, a neural network ensemble was developed by combining five networks that have the same architecture and inputs but differ in layer weights. The ensemble predicted the Kow values with RMSE of 0.28 and 0.38 for the training set and internal validation set, respectively. The ensemble performed reasonably well on an external dataset when compared with other popular Kow models in the literature.

A New Variable Selection Method Based on Mutual Information Maximization by Replacing Collinear Variables for Nonlinear Quantitative Structure-Property Relationship Models

Selection of the most informative molecular descriptors from the original data set is a key step for development of quantitative structure activity/property relationship models. Recently, mutual information (MI) has gained increasing attention in feature selection problems. This paper presents an effective mutual information-based feature selection approach, named mutual information maximization by replacing collinear variables (MIMRCV), for nonlinear quantitative structure-property relationship models. The proposed variable selection method was applied to three different QSPR datasets, soil degradation half-life of 47 organophosphorus pesticides, GC-MS retention times of 85 volatile organic compounds, and water-to-micellar cetyltrimethylammonium bromide partition coefficients of 62 organic compounds.The obtained results revealed that using MIMRCV as feature selection method improves the predictive quality of the developed models compared to conventional MI based variable selection algorithms.

Support vector regression based QSPR for the prediction of retention time of pesticide residues in gas chromatography–mass spectroscopy

In this study, a quantitative structure–property relationship (QSPR) method was employed to predict the retention time (tR) of 368 pesticide residues in animal tissues separated by gas chromatography–mass spectroscopy (GC–MS). The variable selection method of genetic algorithm-partial least squares (GA-PLS) was employed to select most favorable subset of descriptors. The six descriptors selected using GA-PLS were used as inputs of PLS, ANN and SVM to predict the retention times. These descriptors are: number of nitrogen atoms, solvation connectivity index‐Chi 1, Balaban Y index, Moran autocorrelation‐lag 2/weighted by atomic Sanderson electronegativity, total absolute charge and radial distribution function-6.0/unweighted. The correlation coefficients, R, between experimental and predicted tR for the prediction set by PLS, ANN and SVM are 0.907, 0.963 and 0.985 respectively. Results obtained reveal the reliability and good predictability of nonlinear QSPR model to predict the retention time of pesticides. Comparison between the values of statistical parameters reveals the superiority of the SVM model over PLS and ANN ones.

IN SILICO PREDICTION OF MELTING POINTS OF IONIC LIQUIDS BY USING MULTILAYER PERCEPTRON NEURAL NETWORKS

Quantitative structure–property relationship (QSPR) was used to predict melting points of 62 ionic liquids (ILs), which include ammonium, pyrrolidiniu, imidazolium, pyridiniu, piperidiniu, phosphonium ionic liquid salts. The structures of ionic liquids were optimized by Hyperchem software and MOPAC program, and stepwise multiple linear regression method was applied to select the relevant structural descriptors. The predicting models correlating selected descriptors and melting points were set up using multiple linear regressions (MLR) and multilayer perceptron neural network (MLP NN), separately. The obtained linear and nonlinear QSPR models were validated by internal and external test sets. According to the obtained results, the correlation coefficients between predicted and experimental melting points for training, test and validation sets were; 0.91, 0.86 and 0.79 for MLR model. These values for MLP NN model were; 0.97, 0.96 and 0.85, respectively. The results of this study revealed the high applicability of QSPR approach to melting point prediction of ILs.

Virtual Design of Chemical Penetration Enhancers for Transdermal Drug Delivery

Traditional drug design is a laborious and expensive process that often challenges the pharmaceutical industries. As a result, researchers have turned to computational methods for computer-assisted molecular design. Recently, genetic and evolutionary algorithms have emerged as efficient methods in solving combinatorial problems associated with computer-aided molecular design. Further, combining genetic algorithms with quantitative structure-property relationship analyses has proved effective in drug design. In this work, we have integrated a new genetic algorithm and nonlinear quantitative structure-property relationship models to develop a reliable virtual screening algorithm for the generation of potential chemical penetration enhancers. The genetic algorithms-quantitative structure-property relationship algorithm has been implemented successfully to identify potential chemical penetration enhancers for transdermal drug delivery of insulin. Validation of the newly identified chemical penetration enhancer molecular structures was conducted through carefully designed experiments, which elucidated the cytotoxicity and permeability of the chemical penetration enhancers.