Quantitative Structure−Property Relationship (QSPR) Prediction of Solvation Gibbs Energy of Bifunctional Compounds by Recursive Neural Networks

Abstract

In this paper we apply a recursive neural network (RNN) model to the prediction of the standard Gibbs energy of solvation in water of mono- and polyfunctional organic compounds. The proposed model is able to directly take as input a structured representation of the molecule and to model a direct and adaptive relationship between the molecular structure and the target property. A data set of 339 mono- and polyfunctional acyclic compounds including alkanes, alkenes, alkynes, alcohols, ethers, thiols, thioethers, aldehydes, ketones, carboxylic acids, esters, amines, amides, haloalkanes, nitriles, and nitroalkanes was considered. As a result of the statistical analysis, we obtained for the predictive capability estimated on a test set of molecules a mean absolute residual of about 1 kJ·mol−1 and a standard deviation of 1.8 kJ·mol−1. This results is quite satisfactory by considering the intrinsic difficulty of predicting solvation properties in water of compounds containing more than one functional group.