Neural Network Modeling of Melting Temperatures for Sulfur-Containing Organic Compounds #
BioChem Press, Julian Kozio · 2002
Motivation. Searching for a comprehensive numerical description of the chemical structure and for methods that enable to develop effective and credible QSPR (quantitative structure–property relationships) models represent significant challenges for the contemporary theoretical chemistry. Among these methods artificial neural networks (ANN) appears to be very promising in obtaining models that convert structural features into different properties of chemical compounds. Method. Two different models relating structural descriptors to melting temperatures of sulfur containing organic compounds are developed using ANN. A new set of molecular descriptors is evaluated to determine their suitability for QSPR studies. Using two data sets containing 150 sulfides and 226 sulfones, ANN trained with the back propagation and conjugated gradient algorithms are able to predict the melting temperatures with good accuracy. Results. The results obtained show a good predictive ability for the ANN models, giving R 2 cv equal to 0.880 and 0.794 for the sulfides and sulfones, respectively. Conclusions. The QSPR studies described in this paper provide strong evidence that the tested structural descriptors are useful and effective for the ANN modeling of the melting temperatures of sulfides and sulfones.