Optimal Partition of Datasets of QSPR Studies: A Sampling Problem

Abstract

Starting from different partitions of a 160 compounds dataset into training and test sets, we developed discriminant funtions to classify drugs into different categories of human intestinal absorptions rate. For each partition of the dataser, models that included up to ten Dragon descriptors were built, and the performance of each discriminante funtion in teh classification of the training and test sets was assessec and explores graphically through divergence diagrams. Results suggest that external validation tends to underestimate the predictive capability of QSAR models and that the more raliable results from external validation are obtained with even partitions of small and medium size datasets.