Lm. Egolf et Pc. Jurs, PREDICTION OF BOILING POINTS OF ORGANIC HETEROCYCLIC-COMPOUNDS USING REGRESSION AND NEURAL-NETWORK TECHNIQUES, Journal of chemical information and computer sciences, 33(4), 1993, pp. 616-625
High quality models which relate structural descriptors to normal boil
ing points have been developed for large, diverse groups of heterocycl
ic compounds using both linear regression and neural network technique
s. Parallel experiments were designed to compare the performance of th
ese complementary modeling techniques on two different data sets. A fo
rmerly studied data set comprised of 299 tetrahydrofuran (THF), thioph
ene, furan, and pyran compounds was reexamined using neural networks.
In addition, a new data set of 572 pyridine compounds was investigated
to increase our understanding of the nitrogen-containing heterocycles
. First, several new descriptors were developed to explore chemical pr
inciples which govern the boiling point process. In particular, descri
ptors that reflect hydrogen bonding and dipole-dipole interactions pro
ved especially useful for improving the predictive models in the pyrid
ine regression work. With each data set, neural networks were trained
to predict boiling points with close to experimental accuracy using th
e back-propagation learning algorithm. Results from these boiling poin
t investigations show that once the key structural features are indent
ified through traditional regression techniques, neural networks gener
ally provide access to superior predictive equations. On the basis of
this information, further studies were initiated to explore using neur
al networks as a tool to upgrade, structural feature selection. Result
s from this phase of the study demonstrate that this methodology can b
e used to identify the most informationally rich descriptors.