Journal of Chemical Information and Modeling 2007-01-01

Evaluation of a published in silico model and construction of a novel Bayesian model for predicting phospholipidosis inducing potential.

Dennis J Pelletier, Daniel Gehlhaar, Anne Tilloy-Ellul, Theodore O Johnson, Nigel Greene

Index: J. Chem. Inf. Model. 47 , 1196-205, (2007)

Full Text: HTML

Abstract

The identification of phospholipidosis (PPL) during preclinical testing in animals is a recognized problem in the pharmaceutical industry. Depending on the intended indication and dosing regimen, PPL can delay or stop development of a compound in the drug discovery process. Therefore, for programs and projects where a PPL finding would have adverse impact on the success of the project, it would be desirable to be able to rapidly identify and screen out those compounds with the potential to induce PPL as early as possible. Currently, electron microscopy is the gold standard method for identifying phospholipidosis, but it is low-throughput and resource-demanding. Therefore, a low-cost, high-throughput screening strategy is required to overcome these limitations and be applicable in the drug discovery cycle. A recent publication by Ploemen et al. (Exp. Toxicol. Pathol. 2004, 55, 347-55) describes a method using the computed physicochemical properties pKa and ClogP as part of a simple calculation to determine a compound's potential to induce PPL. We have evaluated this method using a set of 201 compounds, both public and proprietary, with known in vivo PPL-inducing ability and have found the overall concordance to be 75%. We have proposed simple modifications to the model rules, which improve the model's concordance to 80%. Finally, we describe the development of a Bayesian model using the same compound set and found its overall concordance to be 83%.