371
12 Consensus Drug Design Using IT Microcosm
studied compounds or where no structural data are available for a given target, such
as when the desired effect is of a systemic nature. For this reason, QSAR appears
to be the most common method of in silico drug discovery [27]. MFTA [73], HIT
QSAR [51], PASS [70], ISIDA [92], NASAWIN [8], and CORAL [89] are QSAR
software packages that have been successfully utilized for the prediction of a wide
variety of biological activities.
The fragmental approach is one the most productive QSAR methods [43, 111,
139]; this approach is based on the idea that a chemical structure is a set of substructural fragments that can be isolated from the structural formulas of compounds
according to certain rules, together with various parameters that characterize these
fragments. In a certain sense, the classical structural formula is superior to any 3D
model with respect to the reliability of information about a compound; the reliability of 3D models depends on the methods and conditions of their development. An
adequate methodology of extracting information contained in the structural formula
yields up to 90 % of for information about the properties of a given substance, even
without resorting to 3D modeling [104, 110].
Any parametric description of a chemical structure, including a fragment-based
description, destroys the pattern of a compound as a whole [27, 43], which can
lead to the loss of information about the substance as a unique object at an overcybernetic level of organization. Therefore, an adequate representation of a chemical structure, without the loss of information specific to any compound, is of the utmost importance in QSAR. This is especially relevant for highly active compounds,
which commonly show chemical novelty and can be termed “upstarts” according to
their characteristics. Such compounds do not follow the conventional regularities
describing compounds with medium activity [135]. One rewarding way to overcome the loss of information about the integrity of a compound is to resort to a
multilevel hierarchical description of the chemical structure by a set of substructural
descriptors with increasing complexity [51, 107].
A wide range of methods for restoring empirical regularity are used within the
framework of the fragmental approach for the calculation of structure-activity relationships: regression [51, 137], pattern recognition [23, 57], artificial neural networks [9], and machine learning [23, 31]. Meanwhile, the relationship between the
pharmacological activity and structure of a chemical compound is not originally
continuous in nature because it includes a multiplicity of discrete components such
as pleiotropic effects [71] (i.e., multiple physiological mechanisms of action [25]
or interactions with several biological targets [62]), selective complementarity to
certain pockets of binding sites (the privileged molecule phenomenon) [21], selective transport by specific proteins [7], synergy with other compounds [86], and
so on. In addition, the variability of pharmacological data is very high, due to the
extreme complexity of higher animals; for example, the parameter dispersion of
techniques for studying behavioral performance can be as high as dozens of percent
[14]. Therefore, the use of methods for restoring smooth continuous dependencies
in QSAR analysis of pharmacological activity is considerably limited.
The vast majority of QSAR studies depend on a rather controversial working
concept: the choice of a “better equation” for the prediction of the studied activity.
12 Consensus Drug Design Using IT Microcosm
studied compounds or where no structural data are available for a given target, such
as when the desired effect is of a systemic nature. For this reason, QSAR appears
to be the most common method of in silico drug discovery [27]. MFTA [73], HIT
QSAR [51], PASS [70], ISIDA [92], NASAWIN [8], and CORAL [89] are QSAR
software packages that have been successfully utilized for the prediction of a wide
variety of biological activities.
The fragmental approach is one the most productive QSAR methods [43, 111,
139]; this approach is based on the idea that a chemical structure is a set of substructural fragments that can be isolated from the structural formulas of compounds
according to certain rules, together with various parameters that characterize these
fragments. In a certain sense, the classical structural formula is superior to any 3D
model with respect to the reliability of information about a compound; the reliability of 3D models depends on the methods and conditions of their development. An
adequate methodology of extracting information contained in the structural formula
yields up to 90 % of for information about the properties of a given substance, even
without resorting to 3D modeling [104, 110].
Any parametric description of a chemical structure, including a fragment-based
description, destroys the pattern of a compound as a whole [27, 43], which can
lead to the loss of information about the substance as a unique object at an overcybernetic level of organization. Therefore, an adequate representation of a chemical structure, without the loss of information specific to any compound, is of the utmost importance in QSAR. This is especially relevant for highly active compounds,
which commonly show chemical novelty and can be termed “upstarts” according to
their characteristics. Such compounds do not follow the conventional regularities
describing compounds with medium activity [135]. One rewarding way to overcome the loss of information about the integrity of a compound is to resort to a
multilevel hierarchical description of the chemical structure by a set of substructural
descriptors with increasing complexity [51, 107].
A wide range of methods for restoring empirical regularity are used within the
framework of the fragmental approach for the calculation of structure-activity relationships: regression [51, 137], pattern recognition [23, 57], artificial neural networks [9], and machine learning [23, 31]. Meanwhile, the relationship between the
pharmacological activity and structure of a chemical compound is not originally
continuous in nature because it includes a multiplicity of discrete components such
as pleiotropic effects [71] (i.e., multiple physiological mechanisms of action [25]
or interactions with several biological targets [62]), selective complementarity to
certain pockets of binding sites (the privileged molecule phenomenon) [21], selective transport by specific proteins [7], synergy with other compounds [86], and
so on. In addition, the variability of pharmacological data is very high, due to the
extreme complexity of higher animals; for example, the parameter dispersion of
techniques for studying behavioral performance can be as high as dozens of percent
[14]. Therefore, the use of methods for restoring smooth continuous dependencies
in QSAR analysis of pharmacological activity is considerably limited.
The vast majority of QSAR studies depend on a rather controversial working
concept: the choice of a “better equation” for the prediction of the studied activity.