Modeling of Gibbs Energy of Formation of Organic Compounds by Linear and Nonlinear Methods

Abstract

Two quantitative models for the prediction of the Gibbs energy of formation (DeltaGf degrees ) of 177 organic compounds were developed. These molecules contain elements such as H, C, N, O, F, S, Cl, and Br, with the molecular weight in the range of 16.04-202.25. The molecules were represented by six selected 2D-structure descriptors. At first, the complex relationship between DeltaGf degrees and the six selected input descriptors was depicted by a two-dimensional Kohonen's self-organizing neural network (KohNN) map; on the basis of the KohNN map, the whole data set was split into a training set consisting of 130 compounds and a test set (or a validation set and a test set) including 47 compounds. Then, DeltaGf degrees was predicted using a multilinear regression (MLR) analysis and a back-propagation (BPG) neural network. For 177 organic compounds, root-mean-square deviations of 17.8 and 15.4 kcal mol-1 were achieved by MLR and the BPG neural network, respectively.