Topixantrone hydrochloride

A conceptually simple, fully in silico model to predict total clearance of new compounds in humans is described. Based on the premise that similar molecules will exhibit similar pharmacokinetic properties, we used a k-nearest-neighbors (kNN) technique to predict total clearance by comparison with known reference agents. Molecular similarity was defined using readily calculated one- and two-dimensional molecular descriptors, and the reference set was obtained by combining the Obach and Berellini sets of human pharmacokinetic data. Neutral molecules and drugs whose biological activity is associated with a metal center were removed from the combined set. The remaining 462 compounds were partitioned into a training and external test set of 370 and 92 compounds, respectively. For acids, bases, zwitterions, and quaternary ammonium/pyridinium ions, average prediction accuracy was within two-fold of observed for the external test set (n = 92). Using a collection of 20 drugs from the literature with > or =3 preclinical animal species allometric scaling data, accuracy of the in silico kNN model was not as good as the rule of exponents, but better than simple allometry (SA), and approached that of combination multiexponential allometry (ME) as defined by the number of predictions with < or =50% error. For a collection of 18 drugs with two species (rat-dog) data, the kNN model outperformed both SA and combination ME using the same performance standard. Since the model is fully in silico and, therefore, capable of generating total clearance predictions in the absence of any experimental data, it can be used to help guide early drug discovery research efforts, such as virtual compound library screening, and analogue prioritization prior to chemical synthesis and biological evaluation. Model validation was accomplished using the external test set, three- and five-fold cross-validation and two different y-randomization techniques (y-shuffling and random number pseudodescriptors).