397
12 Consensus Drug Design Using IT Microcosm
Altogether, in all strategies and methods of testing, the maximum value of all three
indices F 0 , F a and F n amounts to 100 %.
The estimates of prediction accuracy for the best decision rules (for the F 0 value
sum for ST, LOOCV and SHCV) are shown in Table 12.4. In this case, the conservative strategy proved to be the best; optimal decision rules were obtained for 26 activities out 34 (76 %), mainly for systemic pharmacological effects. This is not surprising because the presence/absence of activity is mostly determined by “standard”
regularities, and the structural diversity of compounds underlies the multiplicity of
action mechanisms, especially when dealing with therapeutic activities. The normal
and risk strategies yielded 4 optimum decision rules (12 % each). Notably, the risk
strategy produced decision rules for only the receptor activity type, where subtle
interaction mechanisms between the ligand and receptor site play a key role.
A summary of testing adequate decision rules in predicting a high-level activity
in structurally diverse compounds is shown in Table 12.5. In this case, the maximum values F 0 , F a and F n in all strategies only amount to 100 % in the self-prediction model. In the leave-one-out and split-half cross-validations, the maximum
values of F 0 , F a , and F n were 99 %, 95 %, and 100 %, respectively.
The estimates of the best decision rules accuracy in predicting a high activity
are shown in Table 12.6. In this case, the conservative strategy again gave the best
results; optimum decision rules were obtained for 7 activities out of 16 (44 %).
However, the normal (5 optimum regularities, 31 %) and risk strategies (4 optimum
regularities, 25 %) were used more often. This outcome is also understandable because highly active compounds often show a “nonstandard” chemical structure and
as a result tend to have unconventional mechanisms of action.
Table 12.3 General indices of prediction accuracy for the activity of structurally diverse
compounds
Strategy
ST, %
LOOCV,
%
SHCV, %
DLOOCV,
%
Min
Max
Min
Max
Min
Max Min
Max
Accuracy F 0
Conservative 97
100
77
100
70
100
–
–
Normal
81
100
69
100
65
95
69
100
Risk
85
100
67
100
65
97
–
–
Sensitivity F a
Conservative 88
100
67
100
67
100
–
–
Normal
83
100
63
100
62
97
63
100
Risk
83
100
67
100
63
100
–
–
Specificity F n
Conservative 96
100
68
100
60
100
–
–
Normal
79
100
69
100
64
96
69
100
Risk
85
100
67
100
60
100
–
–
DLOOCV is not used in the conservative or risk strategy
12 Consensus Drug Design Using IT Microcosm
Altogether, in all strategies and methods of testing, the maximum value of all three
indices F 0 , F a and F n amounts to 100 %.
The estimates of prediction accuracy for the best decision rules (for the F 0 value
sum for ST, LOOCV and SHCV) are shown in Table 12.4. In this case, the conservative strategy proved to be the best; optimal decision rules were obtained for 26 activities out 34 (76 %), mainly for systemic pharmacological effects. This is not surprising because the presence/absence of activity is mostly determined by “standard”
regularities, and the structural diversity of compounds underlies the multiplicity of
action mechanisms, especially when dealing with therapeutic activities. The normal
and risk strategies yielded 4 optimum decision rules (12 % each). Notably, the risk
strategy produced decision rules for only the receptor activity type, where subtle
interaction mechanisms between the ligand and receptor site play a key role.
A summary of testing adequate decision rules in predicting a high-level activity
in structurally diverse compounds is shown in Table 12.5. In this case, the maximum values F 0 , F a and F n in all strategies only amount to 100 % in the self-prediction model. In the leave-one-out and split-half cross-validations, the maximum
values of F 0 , F a , and F n were 99 %, 95 %, and 100 %, respectively.
The estimates of the best decision rules accuracy in predicting a high activity
are shown in Table 12.6. In this case, the conservative strategy again gave the best
results; optimum decision rules were obtained for 7 activities out of 16 (44 %).
However, the normal (5 optimum regularities, 31 %) and risk strategies (4 optimum
regularities, 25 %) were used more often. This outcome is also understandable because highly active compounds often show a “nonstandard” chemical structure and
as a result tend to have unconventional mechanisms of action.
Table 12.3 General indices of prediction accuracy for the activity of structurally diverse
compounds
Strategy
ST, %
LOOCV,
%
SHCV, %
DLOOCV,
%
Min
Max
Min
Max
Min
Max Min
Max
Accuracy F 0
Conservative 97
100
77
100
70
100
–
–
Normal
81
100
69
100
65
95
69
100
Risk
85
100
67
100
65
97
–
–
Sensitivity F a
Conservative 88
100
67
100
67
100
–
–
Normal
83
100
63
100
62
97
63
100
Risk
83
100
67
100
63
100
–
–
Specificity F n
Conservative 96
100
68
100
60
100
–
–
Normal
79
100
69
100
64
96
69
100
Risk
85
100
67
100
60
100
–
–
DLOOCV is not used in the conservative or risk strategy