Three-dimensional quantitative structure activity relationship computational approaches for prediction of human in vitro intrinsic clearance

Future alternatives to the presently accepted in vitro paradigm of predicti on of intrinsic clearance, which could be used earlier in the drug discover y process, would potentially accelerate efforts to identify better drug can didates with more favorable metabolic profiles and less likelihood of failu re with regard to human pharmacokinetic attributes. In this study we descri be two computational methods for modeling human microsomal and hepatocyte i ntrinsic clearance data derived from our laboratory and the literature, whi ch utilize pharmacophore features or descriptors derived from molecular str ucture. Human microsomal intrinsic clearance data generated for 26 known th erapeutic drugs were used to build computational models using commercially available software (Catalyst and Cerius(2)), after first converting the dat a to hepatocyte intrinsic clearance. The best Catalyst pharmacophore model gave an r of 0.77 for the observed versus predicted clearance. This pharmac ophore was described by one hydrogen bond acceptor, two hydrophobic feature s, and one ring aromatic feature essential to discriminate between high and low intrinsic clearance. The Cerius(2) quantitative structure activity rel ationship (QSAR) model gave an r(2) = 0.68 for the observed versus predicte d clearance and a cross-validated r(2) (q(2)) of 0.42. Similarly, literatur e data for human hepatocyte intrinsic clearance for 18 therapeutic drugs we re also used to generate two separate models using the same computational a pproaches. The best Catalyst pharmacophore model gave an improved r of 0.87 and was described by two hydrogen bond acceptors, one hydrophobe, and 1 po sitive ionizable feature. The Cerius(2) QSAR gave an r(2) of 0.88 and a q(2 ) of 0.79. Each of these models was then used as a test set for prediction of the intrinsic clearance data in the other data set, with variable succes ses. These present models represent a preliminary application of QSAR softw are to modeling and prediction of human in vitro intrinsic clearance.