A Molecular Approach for the Prediction of Sulfur Compound Solubility Parameters

Abstract

A quantitative structure–property relationship (QSPR) study was performed to construct a multivariate linear model and a three-layer feed-forward neural network model. This model relates the solubility parameters of 82 sulfur compounds to their structures. Molecular descriptors, which are extracted from the molecular structure of compounds, have been used as model parameters. The multivariate linear model was gained by a genetic algorithm–based multivariate linear regression; the results showed that the squared correlation coefficient (R²) between predicted and experimental values was 0.964. Next, a three-layer feed-forward neural network model with optimized structure was employed; the results showed that the squared correlation coefficient (R²) is 0.9874, and with this model we can predict the solubility parameter more accurately than the linear model.

Supplemental materials are available for this article. Go to the publisher's online edition of Phosphorus, Sulfur, and Silicon and the Related Elements to view the free supplemental file.

Keywords:

• Genetic algorithm
• mercaptans
• neural networks
• QSPR
• solubility parameter