You are here:
IN SILICO MODELLING OF HAZARDOUS ENDPOINTS: CURRENT PROBLEMS AND PROSPECTIVES
Citation:
Mekenyan, O. G., S. Dimitrov, P. K. Schmieder, AND G D. Veith. IN SILICO MODELLING OF HAZARDOUS ENDPOINTS: CURRENT PROBLEMS AND PROSPECTIVES. Presented at 2nd Int'l. Symposium on Computational Methods in Toxicology and Pharmacology Integrating Internet Resources, Thessalomiki, Greece, September 17-19, 2003.
Description:
The primary hurdles for Quantitative Structure-Activity Relationships (QSARs) to overcome their acceptance for regulatory purposes will be discussed. They include (a) the development of more mechanistic representations of chemical structure, (b) the classification of toxicity pathways through which chemical reactivity leads to adverse effects, and (c) the design of systematic databases for each pathway that ensures that the domain of any QSAR is adequate for the regulatory decisions. The first issue encompasses the 3D character of the interactions of chemicals with biomolecules. Using 3D models of chemicals, however, requires investigation of chemical flexibility given the fact that the electronic properties (hence reactivity) of conformations of a chemical can vary substantially. Capabilities will be described representing chemicals as a distribution of plausible conformations, quantifying the molecular descriptors as a function of conformation, and examining whether a chemical is flexible enough to conform to an "induced fit" by the receptor itself. Similarly, the predictive methods using QSAR will fail if the predictions are based on the parent compound but the effects are caused by an "activated" metabolite. Consequently, one needs to generate a large list of conformations and a large list of metabolites and their respective conformations as a much more accurate representation of possible active states resulting from chemical interactions. From this large list of chemical structures derived from the target chemical, there is a much greater chance of accurately identifying multiple possible effects of chemicals. Models for ER binding affinity, skin sensitisation and mutagenicity, accounting for molecular flexibility and metabolic activation, will be demonstrated by using the TIMES computer system. The usefulness of QSARs in selecting toxicologically relevant endpoints as well as the selection of chemicals to be tested will be discussed. The use of an applicability domain for assessing the reliability of QSAR predictions, in the context of the endpoints under study, will be presented. This is facilitated through recent advances in techniques to assess the predictive probability within model interpolation areas separate from the model extrapolation areas. Predictions outside an interpolation domain have greater uncertainty due to the absence of structural coverage within the training set used for deriving QSAR models. Once the training set domain is defined within the structural space of the modelling parameters, one can select chemicals for strategic testing that will broaden the interpolation domain and reduce uncertainties in predictions. Different criteria can be imposed for selection of chemicals for either model refinement or for model validation. The latest progress in selecting chemicals for strategic testing will be illustrated by the ChemPick program.
Disclaimer: This abstract does not necessarily reflect USEPA policy.