Quantitative Structure-Property Relationship Study for Prediction of Flash Point of Some Organic Compounds Based On SW-MLR Method

Abstract

A quantitative structure-property relationship (QSPR) model was developed for prediction of logFP values of some of organic compounds based on their molecular descriptors. In this study, we have attempted to develop a simple and fast multiple linear regression (MLR) model with high accuracy and precision. The molecular descriptors, which cover different information of molecular structures, were calculated by Dragon software. The most feasible descriptors were selected using stepwise selection. The QSPR model was validated by external set compounds without any contribution in model development step. The accuracy of the suggested model is illustrated using cross-validation, validation through an external test set and Y-randomization. The obtained multivariate linear three descriptors model by MLR has correlation coefficient (R2=0.905, Q2LGO = 0.844, Q2LGO=0.833, REP%=-7.807, F=47.626). All molecular descriptors which have entered in this model are calculated from chemical structure of any molecule. This model was used for the prediction of the flash point values of some organic compounds which were not used in the modeling procedure.