No abstract

Features of the principal component analysis, which can be used to decrease the dimension of the space of initial molecular descriptors in the search for classification rules predicting the activity of new compounds, are considered in application to the class of adamantane and indole derivatives possessing hepatoprotector activity. It is shown that (i) when the number of molecular descriptors is greater than the number of compounds in the training set, a preliminary selection of descriptors in the initial set is necessary to determine the principal components ensuring the optimum classification of compounds with respect to the given activity type, and (ii) in order to obtain the most effective classification rules, it is necessary to take into account the principal components with small and very small contributions to the total dispersion of the initial molecular descriptors.

The characteristics of the hepatoprotective activity of some indole derivatives have been determined. The best combinations of substructural molecular descriptors are established, which allows the hepatoprotective activity in the given chemical series to be predicted with a probability of no less than 80% by the methods of linear discriminant analysis and k nearest neighbors. The influence of various structural elements of the molecules under consideration on their hepatoprotective activity is estimated.

The relationships between structure and hepatoprotector activity in the class of adamantane derivatives were studied by means of linear discriminant analysis with the use of various 2D and 3D descriptors. From the standpoint of predictions based on the descriptors of each particular class used separately, 2D descriptors have proved to be more effective than 3D ones. Discriminant functions constructed using molecular descriptors of different classes (2D, 3D, or combined 2D/3D) provide the maximum predicting ability. 135 0091-150X/05/3903-0135

The AM1 valent approximation version of the self-consistent field (SCF) MO LCAO method was used to calculate the electronic structures of 30 adamantane derivatives and 30 indole derivatives with full optimization of molecular geometry. Linear discriminant analysis was used to identify classification rules allowing the different quantum chemical descriptors to predict hepatoprotective detoxifying actions of compounds in these chemical series with high probability. 262 0091-150X/08/4205-0262