392
P. M. Vassiliev et al.
ω Cts is the conformity index {0, 1} for compound C based on the results of activity gradation t prediction in strategy s, and the corresponding value in the
template of gradation activity q.
To determine the order of experimental testing for a high level of activity, the compounds are first ranked in decreasing order by the number of positive final prediction estimates and then ranked in descending order of conformity coefficient (21).
When the prediction is refined for a certain compound, a full spectrum of prediction
estimates and the values of membership functions are used in all three strategies.
12.2.7 Levels and Types of Consensus
Let us define the level of consensus as the step number in a consecutive hierarchical
generalization of classification estimates.
In IT Microcosm, the chemical structure is represented by 11 types of QL descriptors. For each i level of QL description, separate decision rules are calculated
including all of the i-type descriptors. Thus, each primary prediction estimate of an
activity obtained by one of the four prediction methods for the given level of QL
description is the result of a first-level consensus. All of the prediction methods
implement a procedure of general weighted consensus, and the contribution of each
descriptor of type ij to the final estimate is symbatic with its number in the structure
of the predicted compound.
On the basis of a generalized spectrum of 44 primary prediction estimates, the
integral decision rules are constructed with the help of three prediction strategies.
Each of the three strategies models decision-making by an independent expert
group with a unique method. This procedure may be referred to as second-level
integral consensus. As stated above, the conservative strategy implements a model
of general nonweighted consensus, the normal strategy implements a model of selective weighted consensus, and the risk strategy implements a model of supremum
consensus.
The generalization of prediction results by three strategies corresponds to decision-making by three expert groups employing different methods to determine the
“quality” of the predicted object. This type of a QSAR model is something akin to
a “consensus of consensuses;” it is a third-level integral consensus.
To enhance the reliability of predictions and determine the order of experimental
testing of compounds, IT Microcosm uses a fourth-level integral consensus in the
generalization of prediction estimates derived for several activity levels.
12.2.8 Pharmacophore Analysis
The theoretical concepts of IT Microcosm do not include the basic concept of a
pharmacophore; pharmacophores are merely considered to be extreme particular
manifestations of a generalized pattern of compound showing a desired biological
P. M. Vassiliev et al.
ω Cts is the conformity index {0, 1} for compound C based on the results of activity gradation t prediction in strategy s, and the corresponding value in the
template of gradation activity q.
To determine the order of experimental testing for a high level of activity, the compounds are first ranked in decreasing order by the number of positive final prediction estimates and then ranked in descending order of conformity coefficient (21).
When the prediction is refined for a certain compound, a full spectrum of prediction
estimates and the values of membership functions are used in all three strategies.
12.2.7 Levels and Types of Consensus
Let us define the level of consensus as the step number in a consecutive hierarchical
generalization of classification estimates.
In IT Microcosm, the chemical structure is represented by 11 types of QL descriptors. For each i level of QL description, separate decision rules are calculated
including all of the i-type descriptors. Thus, each primary prediction estimate of an
activity obtained by one of the four prediction methods for the given level of QL
description is the result of a first-level consensus. All of the prediction methods
implement a procedure of general weighted consensus, and the contribution of each
descriptor of type ij to the final estimate is symbatic with its number in the structure
of the predicted compound.
On the basis of a generalized spectrum of 44 primary prediction estimates, the
integral decision rules are constructed with the help of three prediction strategies.
Each of the three strategies models decision-making by an independent expert
group with a unique method. This procedure may be referred to as second-level
integral consensus. As stated above, the conservative strategy implements a model
of general nonweighted consensus, the normal strategy implements a model of selective weighted consensus, and the risk strategy implements a model of supremum
consensus.
The generalization of prediction results by three strategies corresponds to decision-making by three expert groups employing different methods to determine the
“quality” of the predicted object. This type of a QSAR model is something akin to
a “consensus of consensuses;” it is a third-level integral consensus.
To enhance the reliability of predictions and determine the order of experimental
testing of compounds, IT Microcosm uses a fourth-level integral consensus in the
generalization of prediction estimates derived for several activity levels.
12.2.8 Pharmacophore Analysis
The theoretical concepts of IT Microcosm do not include the basic concept of a
pharmacophore; pharmacophores are merely considered to be extreme particular
manifestations of a generalized pattern of compound showing a desired biological