Boosting the limits in early ADME prediction

COMP 189

Marco Pintore1, Nadège Piclin1, Han van de Waterbeemd2, and Jacques R. Chretien1. (1) BioChemics Consulting, Centre d'Innovation, 16, rue Leonard de Vinci, Orleans cedex 2, 45074, France, (2) PDM, Department of Drug Metabolism, Pfizer Global Research and Development, IPC351, Sandwich, CT13 9NJ, United Kingdom
Early ADMET remains high amongst current challenges. The slow progress observed despite huge efforts, is due to the poor or imprecise quality of the information content of the commonly available data. To circumvent this crucial drawback and to boost the limits of ADME early prediction, we will address here the quality assessment of the database mining strategy. It is supported by an innovative Adaptive Fuzzy Partitioning (AFP) algorithm, which was applied to different bioactivities and to ADME data. Progresses up to 10 to 25% have been observed in the predictions with particular cases [1]. Different key points will be considered and presented with examples: (i) open selection and permutation of molecular descriptors issued from large sets; (ii) impact of surface molecular descriptors using Volsurf; (iii) robustness of derived models; (iv) degree of validity of model inside the drug space; (v) estimation of the raw data noise.

1. M. Pintore, H. van de Waterbeemd, N. Piclin, J.R. Chrétien, Prediction of oral bioavailability by adaptive fuzzy partitioning, Eur J Med Chem., 2003 (in press).