393
12 Consensus Drug Design Using IT Microcosm
activity [94, 101, 105]. Nevertheless, the concept of pharmacophore analysis is
preserved in IT Microcosm because it is the conventional approach to structureactivity tasks and it is notable for its demonstrativeness and simplicity. Within the
framework of this approach, IT Microcosm defines a new class of QSAR objects,
pharmacophorepatterns, each including multiple conjunctions of “conventional”
pharmacophores, i.e., fixed fragments of the chemical structure. Compared to conventional pharmacophores, pharmacophorepatterns show a much greater predictive
ability; however, this predictive ability is greatly inferior to that of the generalized
pattern. The procedure of constructing pharmacophorepatterns is described in detail
in reference [79]; it consists of two stages.
Stage one entails calculating the lists of the primary statistically significant QL
features of the desired type (level) of activity.
The Bayesian frequency of occurrence of descriptor ij in class k = a, n
(12.22)
where
S ijk is the number of descriptors of ij-type in class k; and
S ik is the total number of descriptors of i-type in class k.
For each descriptor ij, we evaluate the statistical significance of the Bayesian frequency of the occurrence of this descriptor in classes a and n according to the hypergeometric test
(12.23)
where
!
!(
)!
R
R
r
r R r
=
-
is the combination.
The QL descriptor of ij-type is a statistically significant feature of activity if
(12.24)
where Pr 0 is the statistical significance threshold (0.05, 0.01 or 0.001) selected in
relation to the reliability extent desired by the researcher.
Stage two entails the construction of pharmacophorepatterns according to the
lists of statistically significant QL features. By the pair-wise comparison of all of the
features on the list, the pairs of QL descriptors that have coinciding or structurally
1 ,
1, ..., ,
1, ..., 11
ijk
ijk
ik
i
j
d i
i
S
P
S
d
=
=
+
=
+
(
2)
(
2 ) (
2)
(
1)
P
(
) (
1)
r
,
(
2 )
(
)
i j
S
S
S
S
d
S
S
ija
ijn
ia
in
i
ija
ijn
S
S
d
S
ija
ia
i
ija
S
S
d
ia
in
i
S
d
ia
i
+
+
+
+
-
+
+
×
+
+
-
+
+
+
+
=
P
ija
P
ijn
ij
≥
≤
and Pr
Pr ,
0
12 Consensus Drug Design Using IT Microcosm
activity [94, 101, 105]. Nevertheless, the concept of pharmacophore analysis is
preserved in IT Microcosm because it is the conventional approach to structureactivity tasks and it is notable for its demonstrativeness and simplicity. Within the
framework of this approach, IT Microcosm defines a new class of QSAR objects,
pharmacophorepatterns, each including multiple conjunctions of “conventional”
pharmacophores, i.e., fixed fragments of the chemical structure. Compared to conventional pharmacophores, pharmacophorepatterns show a much greater predictive
ability; however, this predictive ability is greatly inferior to that of the generalized
pattern. The procedure of constructing pharmacophorepatterns is described in detail
in reference [79]; it consists of two stages.
Stage one entails calculating the lists of the primary statistically significant QL
features of the desired type (level) of activity.
The Bayesian frequency of occurrence of descriptor ij in class k = a, n
(12.22)
where
S ijk is the number of descriptors of ij-type in class k; and
S ik is the total number of descriptors of i-type in class k.
For each descriptor ij, we evaluate the statistical significance of the Bayesian frequency of the occurrence of this descriptor in classes a and n according to the hypergeometric test
(12.23)
where
!
!(
)!
R
R
r
r R r
=
-
is the combination.
The QL descriptor of ij-type is a statistically significant feature of activity if
(12.24)
where Pr 0 is the statistical significance threshold (0.05, 0.01 or 0.001) selected in
relation to the reliability extent desired by the researcher.
Stage two entails the construction of pharmacophorepatterns according to the
lists of statistically significant QL features. By the pair-wise comparison of all of the
features on the list, the pairs of QL descriptors that have coinciding or structurally
1 ,
1, ..., ,
1, ..., 11
ijk
ijk
ik
i
j
d i
i
S
P
S
d
=
=
+
=
+
(
2)
(
2 ) (
2)
(
1)
P
(
) (
1)
r
,
(
2 )
(
)
i j
S
S
S
S
d
S
S
ija
ijn
ia
in
i
ija
ijn
S
S
d
S
ija
ia
i
ija
S
S
d
ia
in
i
S
d
ia
i
+
+
+
+
-
+
+
×
+
+
-
+
+
+
+
=
P
ija
P
ijn
ij
≥
≤
and Pr
Pr ,
0