The work described in this thesis involves a novel method
of substructural analysis, with potential application
for structure-
property correlation and
information
retrieval within computerised
chemical
information
systems.
A
review
is
given of
the development of
the concept of chemical
structure and
its
representation,
its application
in
computerised
chemical
information
systems, and methods
for correlating structure
with molecular properties.
A method
is presented
for derivation of structural
features,
representing
the whole structure,
from Wiswesser Line Notation
(WLN)
by computer program.
These features are
then used as variables
in
statistical analysis procedures:
in this work multiple regression
analysis and cluster analysis are used. This procedure allows
for a
rapid, convenient and
thorough analysis of
large data-sets. The type
of structural
features used may be easily varied, allowing
for investi-
gation of
factors such as ring substitution patterns, group
interactions,
and
three-dimensional structure.
The method
is applicable
to sets of
diverse or structurally related compounds. Statistical tests of
the
results enable quantitative
testing of hypotheses.
Multiple
regression analysis allows a direct, quantitative
correlation between
structure and molecular property, and subsequent
property prediction.
It is applied
to sets of aliphatic, alicyclic aromatic, and heterocyclic compounds,
including sets of highly diverse
structures. Properties examined
include biological effects,
toxicty,
pK,
thermochemical properties, boiling point, solubility, and
partition coefficient. Some of
these properties are highly dependent
upon electronic and steric effects, and hence upon relative position
of substituents, and on
three-dimensional structure. Highly
significant
correlations are obtained
in all cases, and
the potential
for property
prediction
is demonstrated.
Cluster analysis
is applied
to several sets of structures.
Intuitively sensible classifications are obtained, and
the potential
for both property prediction and
information
retrieval
discussed.
Since these techniques involve the widely used WLN,
relatively simple COBOL programs, and standard statistical packages,
they should
be applicable within operational environments.