Adaptive least-squares method applied to structure--activity correlation of hypotensive N-alkyl-N''-cyano-N'-pyridylguanidines.

A method using an adaptive least-squares (ALS) technique has been developed for the discrimination of ordered categorical data. The method (ALS method) has the advantages of simultaneously considering any number of classes and of producing a single discriminant function which can place patterns in several classes. The ALS method was compared with linear discriminant analysis (LDA) in application to the problem of discriminating three-class hypotensive therapeutic indices of 76 N-alkyl-N''-cyano-N'-pyridylguanidines using nine descriptor variables. With the full data set and in the five leave-out runs, it was shown that the ALS method was superior and more stable in recognition and prediction. The structure--activity relationship is discussed on the basis of discriminant functions formulated.