Structure-activity study of antiulcerous and antiinflammatory drugs by discriminant analysis

Discriminant analysis was used in the structure-activity study of antiulcerous benzoguanamines, antiinflammatory phenylacetic acids, and aminouracils. The usual discriminant analysis requires the equality of covariance matrix for the multivariate normal distribution between observation groups. When this condition is not fulfilled for some pairs of groups, a modified procedure, the "admissible" discriminant analysis after Anderson and Bahadur, was applied. In this procedure, the model of equal covariance is not the prerequisite for the analysis. As the primary criterion for selecting the best combination of variables in the discriminant functions, we used the number of misclassified compounds which is minimum. The discriminant variables were selected from the physiochemical parameters used to analyze the variation in hydrophobicity due to structural modifications. The potency scores divided into three groups for each of the three series of compounds were predicted with more than 80% accuracy, when the two-group analysis was performed for the most potent and least potent groups omitting the intermediary group.