Synergistic interactions among QSAR descriptors

Peterangelo, Stephen C.; Seybold, Paul G.

doi:10.1002/qua.10591

Cited by 35 publications

(26 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…PCA has produced model results that are statistically similar to the variable ranking approach as considered by random forests, yet, PCA still requires the computation of all 1485 descriptors for its application which is a relevant shortcoming. The fact that the results produced by PCA and variable ranking approach as considered by random forests are similar is an evidence, as also argued by some authors [51], that the effects of correlation between descriptors mostly affects the interpretation of the model, with only slight effect on its predictive power. Thus the random forest based variable ranking approach is the natural choice for a final model, which, for the present problem, is able to reach robust models using only 89 molecular descriptors.…”

Section: Resultssupporting

confidence: 67%

Random forests for feature selection in QSPR Models - an application for predicting standard enthalpy of formation of hydrocarbons

2013

View full text Add to dashboard Cite

BackgroundOne of the main topics in the development of quantitative structure-property relationship (QSPR) predictive models is the identification of the subset of variables that represent the structure of a molecule and which are predictors for a given property. There are several automated feature selection methods, ranging from backward, forward or stepwise procedures, to further elaborated methodologies such as evolutionary programming. The problem lies in selecting the minimum subset of descriptors that can predict a certain property with a good performance, computationally efficient and in a more robust way, since the presence of irrelevant or redundant features can cause poor generalization capacity. In this paper an alternative selection method, based on Random Forests to determine the variable importance is proposed in the context of QSPR regression problems, with an application to a manually curated dataset for predicting standard enthalpy of formation. The subsequent predictive models are trained with support vector machines introducing the variables sequentially from a ranked list based on the variable importance.ResultsThe model generalizes well even with a high dimensional dataset and in the presence of highly correlated variables. The feature selection step was shown to yield lower prediction errors with RMSE values 23% lower than without feature selection, albeit using only 6% of the total number of variables (89 from the original 1485). The proposed approach further compared favourably with other feature selection methods and dimension reduction of the feature space. The predictive model was selected using a 10-fold cross validation procedure and, after selection, it was validated with an independent set to assess its performance when applied to new data and the results were similar to the ones obtained for the training set, supporting the robustness of the proposed approach.ConclusionsThe proposed methodology seemingly improves the prediction performance of standard enthalpy of formation of hydrocarbons using a limited set of molecular descriptors, providing faster and more cost-effective calculation of descriptors by reducing their numbers, and providing a better understanding of the underlying relationship between the molecular structure represented by descriptors and the property of interest.

show abstract

Section: Resultssupporting

confidence: 67%

Random forests for feature selection in QSPR Models - an application for predicting standard enthalpy of formation of hydrocarbons

2013

View full text Add to dashboard Cite

show abstract

“…The model of the experimental lattice and binding enthalpies has in the K p − ( pp − odd) − 0 ψ I index its best single descriptor, and in the K p − ( pp − odd) couple of indices, { 0 ψ I , 1 ψ I } the best two‐index descriptors. These two indices have a correlation of r = 0.976, which, at the light of research on the subject should not be considered dramatic 33–36. The good X term, which describes these properties, includes the 0 χ v index together with the D v index.…”

Section: Discussionmentioning

confidence: 99%

Model of the physical properties of halides with complete graph‐based indices

Pogliani¹

2004

Int J of Quantum Chemistry

View full text Add to dashboard Cite

ABSTRACT:Complete graph-based molecular connectivity indices are used to model different theoretical and experimental properties of a class of 20 alkali halides, as well as a theoretical quantum property of a class of fluorides and chlorides. The experimental properties are the lattice enthalpy, the binding enthalpy, and the polarizability in three different media of the alkali halides. The experimental lattice and binding enthalpy are used as training set to evaluate the other sets of theoretical energies, with which the two enthalpies have, finally, been "reconstructed." The theoretical energies are Coulomb energy, polarization energy, van der Waals energy, repulsion energy, and zero-point energy. The lattice enthalpy descriptors are rather good descriptors of its constituent energies: Coulomb, van der Waals, repulsion, and zero-point. The descriptors of the binding enthalpies, instead, fail to model some of its constituent energies: Coulomb, polarization, van der Waals, and repulsion energies. All types of energies are described by basis molecular connectivity indices based on odd complete graph algorithms, which allow an optimal description of the "reconstructed" calculated enthalpies. A common description of the experimental lattice and binding enthalpies with a descriptor trained on the binding enthalpies alone has also been possible. The model of three sets of polarization values of the metal halides with a descriptor trained with one set of polarization values compares positively with a previous study on the polarization of organic compounds. The model of this property requires molecular connectivity basis indices based on a sequential complete graph algorithm. The model of the quantum theoretical property, the electron density at the bond critical point of a class of fluorides and chlorides, allows an interesting comparison with density functional theory calculations.

show abstract

“…Moreover, since the computation of all the quantum‐mechanical descriptors depends on the orbital energies, a correlation between them is already expected. Therefore, any other criterion, for example, widely used variance inflation factor also becomes redundant here but only for the models with high predictivity as has also been highlighted by Peterangelo and Seybold …”

Section: Methodsmentioning

confidence: 99%

Exploring the role of quantum‐mechanical descriptors in the concentration‐dependent adsorption of aromatic organic compounds by multiwalled carbon nanotubes

Lata

Vikas

2018

Int J of Quantum Chemistry

View full text Add to dashboard Cite

An accurate quantitative prediction for concentration‐dependent adsorption of organic compounds by carbon‐nanotubes (CNTs) is increasingly in demand to evaluate not only their applications but also risk associated using CNTs. This is often carried out using poly‐parameter linear‐solvation‐energy‐relationships (pp‐LSERs) based on adsorbate descriptors. This work examines the predictivity of existing pp‐LSERs in the prediction of adsorption of aromatic organic compounds by multi‐walled CNTs (MWCNTs) at five different adsorbate concentrations, and compares that with the predictivity of quantum‐mechanical models while revealing an essential role of electron‐correlation. Notably, the most influencing descriptor for the adsorption was found to be the mean polarizability but that arising from the quantum‐mechanical exchange interactions between electrons of the same spin. This work also proposes reliable predictive models based on a combination of the quantum‐mechanical descriptors and pp‐LSER's descriptors, which were applied to predict the adsorption of nucleobases and steroid hormones, with Progesterone found to be maximally adsorbed.

show abstract

Synergistic interactions among QSAR descriptors

Cited by 35 publications

References 17 publications

Random forests for feature selection in QSPR Models - an application for predicting standard enthalpy of formation of hydrocarbons

Random forests for feature selection in QSPR Models - an application for predicting standard enthalpy of formation of hydrocarbons

Model of the physical properties of halides with complete graph‐based indices

Exploring the role of quantum‐mechanical descriptors in the concentration‐dependent adsorption of aromatic organic compounds by multiwalled carbon nanotubes

Contact Info

Product

Resources

About