EVE, a Distance Based Approach for Discriminating Nonlinearly Separable Groups
Romualdo Benigni · Quantitative Structure-Activity Relationships · 1994
Abstract We previously demonstrated that the Euclidian distances between objects provide a description of the data field, which is more flexible and versatile than the description provided by the original variables, from which the distances were calculated. In fact, the distance matrix also contains, together with the information carried by the original variables, information about the relationships between objects. These properties of the Euclidian distances permit the separation of both linearly and nonlinearly separable groups. In this paper, the findings of our previous work were implemented in a computer program (EVE), which permits: a) the identification of the variables that separate two classes; b) the construction of discriminant equations. The potential of EVE was demonstrated by applying it to a number of simulated cases, as well as to a real QSAR problem (antiparasitic activity of Praziquantel and its analogues), where the active chemicals are embedded in the group of the inactives.