Predictive Artificial Neural Network Model for Solvation Enthalpy of Organic Compounds in N,N-Dimethylformamide

摘要

A quantitative structure-property relationship (QSPR) strategy was followed for prediction of the solvation enthalpy (ΔH_(solv)) values for 116 organic compounds in N,N-dimethylformamide. At first, a three-parameter multiple linear regression (MLR) model including the hydrophilic factor (Hy), sum of Kier-Hall electrotopological states (Ss) and 3D-MoRSE signal 15 weighted by atomic van der waals volumes (Mor15v) was generated for the enthalpy data. The model showed a standard error of 5.49 and R~2 = 0.9220. The descriptors were then employed to develop an artificial neural network (ANN) model for estimating the solvation enthalpies. The developed ANN with 3-6-1 topology resulted in the R~2 values of 0.9914, 0.9765, and 0.9796 for the training, validation and test sets, respectively. Relative importances of Ss, Mor15v and Hy were found to be 48.86, 32.08, and 19.06%, respectively. The findings proved the significant role of molecular topology, electron density and hydrophilicity as the structural features determining ΔH_(solv) values of the organic compounds in N,N-dimethylformamide.