Predicting Penetration Across the Blood-Brain Barrier A Rough Set Approach

This paper reports on the results of experiments regarding a biomedical data set describing blood-brain barrier penetration ability of molecules. In this data set 415 cases represent organic compounds with known steady-state concentrations of a drug in the brain and blood. In our experiments we used two different discretization algorithms, based on agglomerative and divisive approaches of cluster analysis, respectively, and two different approaches to missing attribute values: deletion of cases with missing attribute values and deletion of attributes with missing values. Using ten-fold cross validation we concluded that the best strategy is based on a divisive approach of cluster analysis and deleting cases affected by missing attribute values. Moreover, prediction accuracy of this strategy is comparable with the other successful approaches reported in this area.