Generalized Cluster Significance Analysis and Stepwise Cluster Significance Analysis with Conditional Probabilities
Valerie S. Rose, John Wood · Quantitative Structure-Activity Relationships · 1998
Cluster Significance Analysis (CSA) was developed by McFarland and Gans in 1986 as a method of determining which physicochemical properties of a set of compounds are associated with biological activity. The active compounds are expected to be similar to each other with respect to these ‘important’ properties, and so will cluster together in the space defined by the corresponding axes, compared with the compound set as a whole. The method was suitable for activity data classified as ‘active’ or ‘inactive’. Two extensions to the method have been developed to increase its range of applicability and its interpretability: (1) The method has been generalized to enable the analysis of a continuous measure of activity (2) A forward stepwise procedure has been developed which provides significance probability values which are conditional on the previous model having been accepted. A data set previously analysed by McFarland and Gans has been re-analysed using the new methods.