Prediction of the melting points of fatty acids from computed molecular descriptors: a quantitative structure-property relationship study.

The aim of these works present in this paper consisted in the development and evaluation of quantitative structure property models (QSPR) for the prediction of the melting points of a series constituted by 62 fatty acids. The best multilinear regressions method (MLR) is used to develop models for the prediction of the melting points. The descriptors of the model are selected among an extended set of more than 500 descriptors (constitutional, topological, geometric, quantum chemical and thermodynamic descriptors). Applicability domains were defined and the predictive power was determined using a set of validations The QSPR models are established using the BMLR method implemented in CODESSA software, It turns out that the best QSPR model (R(2)=0.948, R(adj)(2)=0.936, SD=0.940 and F-test=190.90) is obtained with five molecular descriptors. Published by Elsevier Ireland Ltd.