During the early ADME profiling the development of simple, interpretable and reliable in silico tools is very important. In this study, rule-based and QSPR approaches were investigated using a large Caco-2 permeability database. Three permeability classes were determined: high (H), moderate (M) and low (L). The main physicochemical properties related with permeability were ranked as follows: Polar Surface Area (PSA)>Lipophilicity (logP/logD)>Molecular Weight (MW)>number of Hydrogen Bond donors and acceptors>Ionization State>number of Rotatable Bonds>number of Rings. The best rule, based on the combination of PSA-MW-logD (3PRule), was able to identify the H, M and L classes with accuracy of 72.2, 72.9 and 70.6 %, respectively. Subsequently, a consensus system based on three voting binary classification trees was constructed. It accurately predicted 78.4/76.1/79.1 % of H/M/L compounds on training and 78.6/71.1/77.6 % on test set. Finally, the 3PRule and multiclassifier were validated with 23 drugs in a Caco-2 assay. The rule is very useful to improve assay design and prioritize the high absorption candidates. Meanwhile the QSPR model exhibits appropriate classification performance. Due to the simplicity, easy interpretation and accuracy, the 3PRule and consensus model developed here can be used in early ADME profiling.
Keywords: ADME profiling; Caco-2 cell permeability; Classification tree; Physicochemical property; Virtual screening.
Copyright © 2013 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.