Application of QSPR for the prediction of gas to 1-octanol solvation enthalpy using support vector regression

Abstract

A quantitative structure property relationship model was developed to predict gas to 1-octanol solvation enthalpy (ΔH Solv) of 127 different organic compounds using support vector machine (SVM). The variable selection method of genetic algorithm (GA) was employed to select optimal subset of descriptors. The five descriptors selected by GA were used as inputs for construction of the multiple linear regression (MLR), artificial neural network (ANN) and SVM models. The standard errors of for the prediction data set given by MLR, ANN and SVM were 6.438, 4.004 and 2.711 kJ mol−1, respectively, which indicated that the prediction result agrees well with the experimental values of these compounds and also revealed the superiority of SVM over MLR and ANN models. Furthermore, the derived models also provided some insight into what structural features are related to the gas to ΔH Solv of various organic compounds.