Prediction of solubility parameters using partial least square regression

Area of Science:

• * Computational Chemistry
• * Quantitative Structure-Property Relationships (QSPR)

Background:

• * Solubility parameters are crucial for predicting material compatibility and chemical interactions.
• * Existing models like Hansen and Hoy provide valuable but sometimes limited predictive capabilities.

Purpose of the Study:

• * To develop a robust QSPR model for predicting total solubility parameter (delta) values.
• * To identify key molecular descriptors influencing solubility.
• * To compare the model's predictive performance against established methods.

Main Methods:

• * Computation of molecular descriptors (e.g., heat of formation, dipole moment, SA, SV, Ui, Hy, HD, HB) using HyperChem 7.5 and Dragon Web.
• * Application of multivariate partial least squares (PLS) statistics for model development.
• * Validation using cross-validation (R(2) = 0.853, Q(2) = 0.813) on a diverse dataset of 51 compounds.

Main Results:

• * A predictive model for total solubility parameter (delta) was successfully developed.
• * Key descriptors influencing solubility were identified, including heat of formation, dipole moment, and surface area.
• * The model demonstrated high predictive accuracy with no outliers observed in the dataset.

Conclusions:

• * Computed molecular descriptors and PLS statistics offer an effective approach for predicting solubility parameters.
• * The developed QSPR model shows predictive power comparable to established solubility prediction systems.
• * This method provides a valuable tool for material science and chemical process design.