Original Articles: 2014 Vol: 6 Issue: 6
Prediction of Henry's law constants for organic compounds using multilayer feedforward neural networks based on linear solvation energy relationship
Henry's law constants are crucial to correctly estimate the solubilities of different organic compounds in water. However, the precise values of these constants are difficult to obtain by traditional approaches in laboratory. In previous studies, a Linear Solvation Energy Relationship (LSER) method has been used to express a relationship between the Henry's law constant and the relative descriptors of organic compounds. Some of these studies have developed a linear regression model to calculate the Henry's law constants using a LSER method. In our study, instead of using linear prediction approach, we successfully established an Artificial Neural Network (ANN) model to predict Henry's law constants based on 72 typical organic compounds, using a LSER method to describe the independent variables of the ANN model. This research work indicates that the linear relationship provided by the LSER method can be calculated to be a non-linear relationship with a lower error using ANN models. Within a permissible error range (30% tolerance), results showed that the Multilayer Feedforward Neural Network (MLFN) model with two nodes (MLFN-2) is an effective model for predicting the Henry's law constants of organic compounds, whose average RMS error is 0.14 logH units.