Publicly available models to predict normal boiling point of organic compounds

Quantitative structure–property models to predict the normal boiling point (Tb) of organic compounds were developed using non-linear ASNNs (associative neural networks) as well as multiple linear regression – ISIDA-MLR and SQS (stochastic QSAR sampler). Models were built on a diverse set of 2098 organic compounds with Tb varying in the range of 185–491K. In ISIDA-MLR and ASNN calculations, fragment descriptors were used, whereas fragment, FPTs (fuzzy pharmacophore triplets), and ChemAxon descriptors were employed in SQS models. Prediction quality of the models has been assessed in 5-fold cross validation. Obtained models were implemented in the on-line ISIDA predictor at http://infochim.u-strasbg.fr/webserv/VSEngine.html.