Abstract

The assessment of iDEA pkEXPRESS™ 1.1 for the in silico prediction of Caco-2 permeabilities was investigated. We found that when the software was used in a high/low classification scheme with a cutoff Caco-2 permeability value 5 nm/s, the prediction ability of the tool to evaluate large data sets was acceptable. Using a Caco-2 library of 666 compounds, approximately 78% of the predictions in a high/low classification scheme were correct. In addition, the average fold error of these correct predictions was approximately 3. After removing compounds contained in the training set (275 compounds), the prediction ability of the tool was still acceptable where overall predictions in a high/low classification scheme was 76% correct with an average fold error of approximately 3. The software could be useful for analyzing hits after high-throughput screening of large structurally diverse compound libraries. Unfortunately, the software was unable to provide a measure of confidence to the predictions. Without being able to identify poor predictions, it is difficult to evaluate small data sets using the software.