Effective Molecular Descriptors Research Articles

Interpretable Machine Learning Model for the Highly Accurate Prediction of Efficiency of Ternary Organic Solar Cells Based on Nonfullerene Acceptor using Effective Molecular Descriptors

The challenge of accurately predicting the power conversion efficiency (PCE) of ternary organic solar cells (OSCs) based on a nonfullerene acceptor holds the key to the rational design of a ternary blend. Developing an effective descriptor with experimentally measurable and theoretically computable signatures for accurately predicting the PCE of OSCs based on nonfullerene acceptors is an important step toward achieving this goal. Herein, the electronegativity is first proposed as an effective molecular descriptor for predicting the PCE of OSCs based on nonfullerene acceptors and further analyzing the underlying relationships between material property and device performance. Remarkably, the high accuracy (Coefficient of Determination) > 0.9) can be achieved by constructing the machine learning model with a fewer number of descriptors. In addition, the SHapley Additive exPlanations approach is introduced to provide both local and global interpretations for extracting a deep understanding of complex molecular descriptor–PCE relationships. These results in this study validate the effectiveness of the molecular descriptor, providing an efficient modality for rapid and precise screening of high‐performance ternary materials.

A NEW MOLECULAR DESCRIPTOR FOR EVALUATING COMPONENT COMPATIBILITY OF PHARMACEUTICAL FORMS

Abstract: The chemical diversity of the components of modern dosage forms, the intensity of their pharmacological study and the development of methods for predicting the optimal composition bring to the fore the task of developing new effective molecular descriptors that most fully reflect the physicochemical properties of molecules. At the same time, their creation should not be accompanied by laborious experiments or expensive complex calculations. Based on the analysis of the electronic structure of the model group of excipients, we made an attempt to theoretically substantiate the new molecular descriptor. For this, we set the task of calculating known and introducing new descriptors that characterize the structural features of excipients and allow combining chemically different molecules within a single predictive model. Only this approach can provide a reliable explanation of the phenomenon of technological isosteria (different structures - the same properties) and will allow identifying molecular descriptors, the change of which affects the change in the properties of the dosage form in the required direction. An accurate assessment of the hydrophilicity of molecules is the key to describing their behavior in combined heterogeneous media - the intestinal and gastric walls. Traditional methods of assessment (solubility in the "water-octanol" system) are laborious and, in the case of several tens of compounds, involve a separate expensive practical task, often giving results that are difficult to reproduce. Evaluation of the hydrophilicity of materials of polymer nature by this method is impossible. At the same time, the main components of the DF are these polymer structures of varying degrees of complexity. Therefore, the tasks of evaluative forecasting of the mutual influence of polymers - the components of the DF - and selecting of a cost-effective method for determining their optimal composition have a high degree of relevance.

Accelerating Big Data Quantitative Structure-Activity Prediction through LASSO-Random Forest Algorithm.

The aim of quantitative structure-activity prediction (QSAR) studies is to identify novel drug-like molecules that can be suggested as lead compounds by means of two approaches, which are discussed in this article. First, to identify appropriate molecular descriptors by focusing on one feature-selection algorithms; and second to predict the biological activities of designed compounds.Recent studies have shown increased interest in the prediction of a huge number of molecules, known as Big Data, using deep learning models. However, despite all these efforts to solve critical challenges in QSAR models, such as over-fitting, massive processing procedures, is major shortcomings of deep learning models. Hence, finding the most effective molecular descriptors in the shortest possible time is an ongoing task. One of the successful methods to speed up the extraction of the best features from big datasets is the use of least absolute shrinkage and selection operator (LASSO). This algorithm is a regression model that selects a subset of molecular descriptors with the aim of enhancing prediction accuracy and interpretability because of removing inappropriate and irrelevant features. To implement and test our proposed model, a random forest was built to predict the molecular activities of Kaggle competition compounds. Finally, the prediction results and computation time of the suggested model were compared with the other well-known algorithms, i.e. Boruta-random forest, deep random forest, and deep belief network model. The results revealed that improving output correlation through LASSO-random forest leads to appreciably reduced implementation time and model complexity, while maintaining accuracy of the predictions. Supplementary data are available at Bioinformatics online.

Accurate prediction of standard enthalpy of formation based on semiempirical quantum chemistry methods with artificial neural network and molecular descriptors

AbstractThis work investigates possible improvements in the accuracy of semiempirical quantum chemistry (SQC) methods for the prediction of standard enthalpy of formation (ΔfHo) through the use of an artificial neural network (ANN) with molecular descriptors. A total of 142 organic compounds with enough structural diversity has been considered in the training set. Standard enthalpy of formation for the selected compounds at the semiempirical PM3 and PM6 quantum chemistry methods is collected from literature and is calculated by using the semiempirical PM7 method in this work. The multiple stepwise regression is first used to screen effective molecular descriptors, which are highly correlated with the error terms of the standard enthalpy of formation compared with experimental values. The obtained seven effective molecular descriptors are then used as input set to establish three 7‐11‐1 neural network‐based correction models to improve the accuracy of SQC methods. By using the developed correction models, the mean absolute errors for ΔfHo of PM3, PM6, and PM7 methods are reduced from 22.36, 18.60, and 17.27 to 9.86, 9.83, and 8.95, respectively, in kJ/mol. Meanwhile, the results of the test set show that the neural network does not have the problem of overfitting. Detailed analysis of the seven effective molecular descriptors indicates that the major source of the correction models is the electron‐withdrawing effect. The developed ANN models for the three selected SQC methods provide an efficient method for the quick and accurate prediction of thermodynamic properties.

Effective Molecular Descriptors for Chemical Accuracy at DFT Cost: Fragmentation, Error-Cancellation, and Machine Learning.

Recent advances in theoretical thermochemistry have allowed the study of small organic and bio-organic molecules with high accuracy. However, applications to larger molecules are still impeded by the steep scaling problem of highly accurate quantum mechanical (QM) methods, forcing the use of approximate, more cost-effective methods at a greatly reduced accuracy. One of the most successful strategies to mitigate this error is the use of systematic error-cancellation schemes, in which highly accurate QM calculations can be performed on small portions of the molecule to construct corrections to an approximate method. Herein, we build on ideas from fragmentation and error-cancellation to introduce a new family of molecular descriptors for machine learning modeled after the Connectivity-Based Hierarchy (CBH) of generalized isodesmic reaction schemes. The best performing descriptor ML(CBH-2) is constructed from fragments preserving only the immediate connectivity of all heavy (non-H) atoms of a molecule along with overlapping regions of fragments in accordance with the inclusion-exclusion principle. Our proposed approach offers a simple, chemically intuitive grouping of atoms, tuned with an optimal amount of error-cancellation, and outperforms previous structure-based descriptors using a much smaller input vector length. For a wide variety of density functionals, DFT+ΔML(CBH-2) models, trained on a set of small- to medium-sized organic HCNOSCl-containing molecules, achieved an out-of-sample MAE within 0.5 kcal/mol and 2σ (95%) confidence interval of <1.5 kcal/mol compared to accurate G4 reference values at DFT cost.

Investigation of the most effective molecular descriptors on the thermal behaviour of energetic azido-ester plasticizers through QSPR approach

In this work, the most effective molecular descriptors on the thermal behaviour of energetic azido-ester plasticizers are investigated through quantitative structure–property relationship approach. At first, a new simple and reliable correlation is developed for predicting thermal decomposition temperature (T d) of energetic azido-ester plasticizers through multiple linear regression method. The determination coefficient of the derived correlation is 0.950, and it has root mean square deviation (RMSD) and average absolute deviation (AAD) of 2.74 and 2.09 °C, respectively. The internal and external validation method is shown that the developed correlation has good predictive ability. Then, the relationship between glass transition temperature (T G) of energetic azido-ester plasticizers and their T d is studied. This study shows that the optimum elemental composition, number of ester groups, number of azido groups and the presence of aromatic fragments are the important molecular descriptors as well as several non-additive structural parameters which have significant effect on the thermal behaviour of energetic azido-ester plasticizers. The determination coefficient of this correlation is 0.971, and it has RMSD and AAD of 3.70 and 2.67 °C, respectively. These results could be used to design ideal energetic azido-ester plasticizers.

SambVca 2. A Web Tool for Analyzing Catalytic Pockets with Topographic Steric Maps

Developing more efficient catalysts remains one of the primary targets of organometallic chemists. To accelerate reaching this goal, effective molecular descriptors and visualization tools can represent a remarkable aid. Here, we present a Web application for analyzing the catalytic pocket of metal complexes using topographic steric maps as a general and unbiased descriptor that is suitable for every class of catalysts. To show the broad applicability of our approach, we first compared the steric map of a series of transition metal complexes presenting popular mono-, di-, and tetracoordinated ligands and three classic zirconocenes. This comparative analysis highlighted similarities and differences between totally unrelated ligands. Then, we focused on a recently developed Fe(II) catalyst that is active in the asymmetric transfer hydrogenation of ketones and imines. Finally, we expand the scope of these tools to rationalize the inversion of enantioselectivity in enzymatic catalysis, achieved by point mutat...

Novel PSO-MLR Algorithm to Predict the Chromatographic Retention Behaviors of Natural Compounds

The present report deals with a novel quantitative structure retention relationship (QSRR) study to predict the retention indices (RIs) of a large set of natural compounds involving 95 saturated or unsaturated linear, non-linear, cyclic and heterocyclic terpenoids. A principal component analysis revealed that there was no outlier in the main cluster. After splitting the data set into training and test sets involving 76 and 19 compounds, particle swarm optimization (PSO) algorithm was used to find the most effective molecular descriptors, followed by multiple linear regression (MLR). Four molecular descriptors were adopted for construction of the model belonging to constitutional and topological groups. The method was validated by cross validation and external validation and its robustness was confirmed by Y-randomization test over ten iterations. The promising statistical figures of merit associated with the proposed model (R2train=0.956, R2test=0.968, Q2LGO=0.950, Q2LGO=0.947) represent its high capability to predict RIs with low relative errors of predictions.

A quantitative structure–property relationship for determination of enthalpy of fusion of pure compounds

In this study, the quantitative structure–property relationship method is applied to predict the enthalpy of fusion of pure chemical compounds at their normal melting point. A genetic algorithm-based multivariate linear regression is used to select the most statistically effective molecular descriptors for evaluating this property. To propose a comprehensive and predictive model, 3,846 pure chemical compounds are investigated. The root mean square of error and the average absolute deviation of the model are equal to 2.57 kJ/mol and 9.7%.

QSPR approach for determination of parachor of non-electrolyte organic compounds

In this work, the Quantitative Structure–Property Relationship (QSPR) method is applied to represent/predict the parachor of pure non-electrolyte organic compounds. A Genetic-Algorithm-based Multivariate Linear Regression (GA-MLR) is used to select the most statistically effective molecular descriptors for evaluating this property. To propose a predictive model, 227 pure non-electrolyte organic compounds are investigated. 2.8% absolute average deviation for the represented/predicted parachors of investigated compounds is found from the corresponding experimental values.

Rough Sets for Selection of Molecular Descriptors to Predict Biological Activity of Molecules

Quantitative structure activity relationship (QSAR) is one of the important disciplines of computer-aided drug design that deals with the predictive modeling of properties of a molecule. In general, each QSAR dataset is small in size with large number of features or descriptors. Among the large amount of descriptors presented in the QSAR dataset, only a small fraction of them is effective for performing the predictive modeling task. In this paper, a new feature selection algorithm is presented, based on rough set theory, to select a set of effective molecular descriptors from a given QSAR dataset. The proposed algorithm selects the set of molecular descriptors by maximizing both relevance and significance of the descriptors. An important finding is that the proposed feature selection algorithm is shown to be effective in selecting relevant and significant molecular descriptors from the QSAR dataset for predictive modeling. The performance of the proposed algorithm is studied using R2 statistic of support vector regression method. The effectiveness of the proposed algorithm, along with a comparison with existing algorithms, is demonstrated on three QSAR datasets.

Estimation of molecular diffusivity of pure chemicals in water: a quantitative structure–property relationship study

A quantitative structure–property relationship (QSPR) study was performed to predict the molecular diffusivity of pure chemicals in water. A genetic-algorithm-based multivariate linear regression (GA-MLR) was applied to select the most statistically effective molecular descriptors for modelling the molecular diffusivity of pure chemicals in water. Based on the selected molecular descriptors, a three-layer feed forward neural network (FFNN) was constructed to predict the property. The obtained results showed that the FFNN was able to predict the molecular diffusivity of pure chemicals in water.

Estimation of Aniline Point Temperature of Pure Hydrocarbons: A Quantitative Structure−Property Relationship Approach

In the present work, a quantitative structure−property relationship (QSPR) study is performed to predict the aniline point temperature of pure hydrocarbon components. As a powerful tool, genetic algorithm-based multivariate linear regression (GA-MLR) is applied to select most statistically effective molecular descriptors on the aniline point temperature of pure hydrocarbon components. Also, a three-layer feed forward neural network (FFNN) is constructed to consider the nonlinear behavior of appearing molecular descriptors in GA-MLR result. The obtained results show that the constructed FFNN can accurately predict the aniline point temperature of pure hydrocarbon components.

Prediction of some important physical properties of sulfur compounds using quantitative structure–properties relationships

In this work, physical properties of sulfur compounds (critical temperature (Tc), critical pressure (Pc), and Pitzer's acentric factor (omega)) are predicted using quantitative structure-property relationship technique. Sulfur compounds present in petroleum cuts are considered environmental hazards. Genetic algorithm based multivariate linear regression (GA-MLR) is used to select most statistically effective molecular descriptors on the properties. Using the selected molecular descriptors, feed forward neural networks (FFNNs) are applied to develop some molecular-based models to predict the properties. The presented models are quite accurate and can be used to predict the properties of sulfur compounds.

Prediction of the Watson Characterization Factor of Hydrocarbon Components from Molecular Properties

AbstractIn the present work, a Quantitative Structure–Property Relationship (QSPR) study was performed to predict the Watson characterization factor of hydrocarbon components. A Genetic Algorithm‐based Multivariate Linear Regression (GA‐MLR) was applied to select the most statistically effective molecular descriptors of the Watson characterization factor. Then, based on the selected molecular descriptors by GA‐MLR, a three‐layer Feed Forward Neural Network (FFNN) was constructed to predict the Watson characterization factor. The obtained results showed that the constructed FFNN can accurately predict the Watson characterization factor of hydrocarbon components.

A Molecular‐Based Model for Prediction of Solubility of C60 Fullerene in Various Solvents

In this presented work, a quantitative structure‐property relationship study (QSPR) was done for prediction of solubility of C60 fullerene in various solvents. In this study, genetic algorithm‐based multivariate linear regression (GA‐MLR) was applied to obtain most statistically effective molecular descriptors on solubility of C60 in various solvents. All of these molecular descriptors are only calculated from the chemical structure of solvents. For considering nonlinear behavior of appearing molecular descriptors in GA‐MLR section, a feed forward neural network (FFNN) was constructed and optimized for prediction of solubility of C60 fullerene in solvents. Obtained models considerably showed better accuracy in comparison with the previous models.