469
14 Quantitative Structure-Property Relationship Analysis …
by the very low quality of collected dataset that contained plenty of duplicates and
high inconsistencies in bioavailability values. In the same time, it is still clear that
powerful descriptors related to carrier-mediated transport and first-pass metabolism
are needed for building a useful prediction model for human oral bioavailability.
Recently, Hirono et al. published a study of the quantitative physico-chemical
propertybioavailability relationships for 188 noncongeneric diverse organic API
using the “fuzzy adaptive least squares” method [30]. The model was validated by
“leave one out” cross-validation, and not an independent test set. The compounds
were divided into three groups, non-aromatics, aromatics, and heteroaromatics, and
separate equations were formulated for each group which were statistically reliable
and satisfactory. However, in addition to the need for prior classification of the
compounds into one of the three groups, the lipophilicity of the compounds was not
separately identified as a factor, although many studies have reported that this is
one of the most important properties which determines absorption and metabolism.
Finally, Bains et al. proposed evolutionary and adaptive methods for classifying
drug bioavailability into “high” and “low” classes [31], and showed that obtaining
predictive models on the basis of the molecular structure alone is possible. Innovative concepts for correlating molecular structures with biological activities are
represented by fuzzy logic (FL) [32]. In fact, FL methods based on the possibility to
handle the “concept of partial truth”, provide interesting solutions to classification
problems within the context of imprecise categories, in which ADME properties can
be included. Fuzzy classification represents the boundaries between neighbouring
classes as continuous, assigning to the compounds a degree of membership of each
class. FL has been widely used in the field of process control, where the idea is to
convert human expert knowledge into fuzzy rules, and it is able to extract relevant
structure-activity relationships (SAR) from a database, without a priori knowledge.
Wessel et al. have reported a model based on 76 compounds with reported human
intestinal absorption data, using GA with an ANN scoring function [33]. A standard
error of 16 % was obtained for the test set of 10 molecules. Clark reports the use of
polar surface area (PSA) to create a classification model to separate poor (< 10 %)
from well absorbed (≥ 10 %) compounds [5], using the same dataset as reported by
[33]. Egan et al. published a model for intestinal absorbtion HIA based on PSA and
AlogP descriptors alone [34]. This classification model had accuracy of 74–92 %
for different validation sets. Abraham and collaborators have recently reported a
model based on a comprehensive intestinal absorption dataset [35]. The absorption
of 111 drug and drug-like compounds was evaluated from 111 references based on
the ratio of urinary excretion of drugs following oral and intravenous administration to intact rats and biliary excretion of bile duct-cannulated rats. Ninety-eight
drug compounds with known both human and rat absorption data were selected for
correlation analysis between the human and rat absorption. The results showed that
the extent of absorption in these two species is similar. For 94 % of the drugs the absorption difference between humans and rats is less than 20 % and for 98 % of drugs
the difference is less than 30 %. There is only one drug for which human absorption is significantly different from rat absorption. The standard deviation is 11 %
14 Quantitative Structure-Property Relationship Analysis …
by the very low quality of collected dataset that contained plenty of duplicates and
high inconsistencies in bioavailability values. In the same time, it is still clear that
powerful descriptors related to carrier-mediated transport and first-pass metabolism
are needed for building a useful prediction model for human oral bioavailability.
Recently, Hirono et al. published a study of the quantitative physico-chemical
propertybioavailability relationships for 188 noncongeneric diverse organic API
using the “fuzzy adaptive least squares” method [30]. The model was validated by
“leave one out” cross-validation, and not an independent test set. The compounds
were divided into three groups, non-aromatics, aromatics, and heteroaromatics, and
separate equations were formulated for each group which were statistically reliable
and satisfactory. However, in addition to the need for prior classification of the
compounds into one of the three groups, the lipophilicity of the compounds was not
separately identified as a factor, although many studies have reported that this is
one of the most important properties which determines absorption and metabolism.
Finally, Bains et al. proposed evolutionary and adaptive methods for classifying
drug bioavailability into “high” and “low” classes [31], and showed that obtaining
predictive models on the basis of the molecular structure alone is possible. Innovative concepts for correlating molecular structures with biological activities are
represented by fuzzy logic (FL) [32]. In fact, FL methods based on the possibility to
handle the “concept of partial truth”, provide interesting solutions to classification
problems within the context of imprecise categories, in which ADME properties can
be included. Fuzzy classification represents the boundaries between neighbouring
classes as continuous, assigning to the compounds a degree of membership of each
class. FL has been widely used in the field of process control, where the idea is to
convert human expert knowledge into fuzzy rules, and it is able to extract relevant
structure-activity relationships (SAR) from a database, without a priori knowledge.
Wessel et al. have reported a model based on 76 compounds with reported human
intestinal absorption data, using GA with an ANN scoring function [33]. A standard
error of 16 % was obtained for the test set of 10 molecules. Clark reports the use of
polar surface area (PSA) to create a classification model to separate poor (< 10 %)
from well absorbed (≥ 10 %) compounds [5], using the same dataset as reported by
[33]. Egan et al. published a model for intestinal absorbtion HIA based on PSA and
AlogP descriptors alone [34]. This classification model had accuracy of 74–92 %
for different validation sets. Abraham and collaborators have recently reported a
model based on a comprehensive intestinal absorption dataset [35]. The absorption
of 111 drug and drug-like compounds was evaluated from 111 references based on
the ratio of urinary excretion of drugs following oral and intravenous administration to intact rats and biliary excretion of bile duct-cannulated rats. Ninety-eight
drug compounds with known both human and rat absorption data were selected for
correlation analysis between the human and rat absorption. The results showed that
the extent of absorption in these two species is similar. For 94 % of the drugs the absorption difference between humans and rats is less than 20 % and for 98 % of drugs
the difference is less than 30 %. There is only one drug for which human absorption is significantly different from rat absorption. The standard deviation is 11 %