Global and Local PLS Regression Models to Predict Vapor Pressure

Abstract

The vapor pressure is a key property in determining the distribution and fate of environmentally relevant compounds, but experimental determinations are only available for a limited number of the chemicals in current commercial use. Despite experimental efforts there is a need for estimation methods. The liquid or subcooled liquid vapor pressures at 298.15 K were collected from the literature for a diverse set of 1340 organic compounds. Theoretical molecular descriptors were derived after optimization to low-energy conformations and used to investigate the performance of global and local Quantitative Structure -Property Relationships (QSPR). A global PLSR model with ten latent variables was found to be optimal. The predictive performance of this model, within the domain of applicability, was estimated at n ¼ 420, Q 2 Ext ¼ 0.980, and RMSEP ¼ 0.410 (log Pa). This model can be used in conjunction with other estimation models to assess the potential for a long range atmospheric transport.

Automated QSAR with a Hierarchy of Global and Local Models

“…Feher and Ewing created fusion global models consisting of many local linear models and demonstrated that the approach improved on the accuracy of predictions when compared to a global linear model. [6] While similar effects have been seen in other studies with linear modelling algorithms, [3][4][5] Helgee et al [7] have shown that applications of this approach to non-linear methods did not result in the same improvements to prediction accuracy. Over the past decade, attempts have been made to automate the QSAR modelling process, thus freeing up computational chemistry resources and allowing more exhaustive searches of QSAR model space.…”

“…[3][4][5][6] Local QSAR models are applicable to a specific chemical domain, for example, a single chemical series or single class of compounds, whereas global models are generated from diverse sets of compounds and are usually intended to be applicable to any drug-like compounds. Local models can be more sensitive to small changes in chemical structure than global models, and attempts have been made to exploit this difference by creating global models by combining many local models.…”

Comparison of theoretical and experimental models for characterizing solvent properties using reversed phase liquid chromatography

“…The result was in accordance with that generated from modified global LSER. Furthermore, a PLSR model with theoretical molecular descriptors cannot only predict the chromatographic retention behavior, but also the chemical properties of both solutes and solvents [43,44].…”

CADASTER QSPR Models for Predictions of Melting and Boiling Points of Perfluorinated Chemicals

“…Individual models based on 0D‐2D Dragon descriptors,18 E‐state descriptors19 and fragment based descriptors20 as well as consensus models (by a simple average of the results predicted by all the models developed by project partners) were developed. Different approaches namely Multiple Linear Regression (MLR), Partial Least Squares Regression (PLSR)21 and Neural Network (NN)22 based models, were developed as a collaborative study between project partners. The predictive QSAR studies on rodents’ LC 50 inhalation and oral toxicity of these compounds23 as well as QSPR modeling of Vapor Pressure and Aqueous solubility data21, 24 for PFCs were recently reported.…”