|Abstract or Summary
- In a two part thesis, studies were made of two
different types of structure-activity relationships.
The first part consists of the synthesis and biological
evaluation of a number of 4-substituted pyrazoles and
isoxazoles as inhibitors of liver alcohol dehydrogenase.
In an effort to better define the steric and electronic
effects of substituents, the inhibitory potencies of the
compounds were determined in vitro, and these were
correlated with the lipophilicity parameter [superscript pi] and with
several steric and electronic substituent constants.
Lipophilicity showed the most significant relationship
with activity. Electronic terms were less significant,
but a consistently negative slope coefficient for these
terms suggests that a positive charge develops on the
pyrazole ring during interaction with LADH, which is not
consistent with a previously proposed anionic
In the second part of the thesis, retrospective
research was performed to determine the usefulness of
fragment and whole-molecule molecular connectivity (MC)
indices in the multivariate classification of drug
structures. Initially, using a template method of
position assignment, a number of diphenylmethyl-based
compounds from a previous study were examined using
linear and quadratic discriminant analysis. Relative
molar refraction was compared with several fragment MC
values as descriptor indices. First order fragment MC
was found to be the most useful descriptor of those
examined, and this was verified in a second study using a
number of steroids classified into five therapeutic
categories. In a third study, these steroids were
examined using a number of pattern recognition techniques.
It was concluded that there was no advantage to the use
of pattern recognition techniques over discriminant
analysis, especially since stepwise variable selection is
available in the latter.
Finally, using a composite of the structures that
had been studied, a comparison was made between the
recently introduced SIMCA method of modelling and
classification, and conventional discriminant analysis.
Simple and valence whole-molecule MC indices were used as
descriptors to avoid the problems of position assignment.
The original sample classification results were similar
using both techniques. When a holdout-by-four method of
validation was performed, linear discriminant analysis
gave better results than either SIMCA or the quadratic
discriminant function. This reflects the effects of
small sample sizes on the latter two methods. A comparison
of the canonical discriminant function values and the
principal component scores from SIMCA as predictors of the
quantitative potencies of a number of androgens was
made, and higher correlations were found with the
discriminant function values. This evidence again
suggests that no advantages are gained by using techniques
other than discriminant analysis, although
more experience with pattern recognition methods is
necessary to be certain of this conclusion.