A distance measure between models: a tool for similarity/diversity analysis of model populations

In many research fields, there is, nowadays, a lot of readily available information, however, it needs processing. This is the case of the field of Quantitative Structure-Activity Relationships (QSAR), which exploits several thousand molecular descriptors, and quality control and multivariate calibration where hundreds of spectroscopic signals are easily obtained from spectroscopic methods. Genetic Algorithms, Simulated Annealing, and Tabu Search are some of the methods that are widely used to process available information to find sets of optimal models. In this case, the problem that arises is how to compare the selected models. This work proposes a new measure of the distance between two models, and we will demonstrate that this model distance allows clusters of similar models to be found and the most diverse models to be caught in such a way as to preserve maximum information and diversity. (C) 2003 Elsevier B.V. All rights reserved.