|
(83-19) Performance reliability determination of the QSAR test battery. Jaworska, Joanna*,1, McDowell, Robert2, 1 Procter & Gamble, Strombeek - Bever, Belgium2 USDA, APHIS, Riverdale, USA ABSTRACT- Following current interest in using QSARs in management of chemicals in particular risk assessments and classification and labeling there is an urgent need to optimize QSAR predictivity. Single QSAR predictivity is constrained by, most importantly, experimental error associated with the modeled endpoint. Increasing sample size in a training set will not remove that inherent to the endpoint maximum achievable predictivity and goodness of fit. In order to increase reliability of QSAR based predictions a Bayesian battery selection method is proposed in which results of several QSAR models with varying sensitivity and specificity are combined. In addition to improving reliability, this method allows for science-based consensus building. Application to ready biodegradability prediction will be presented. Key words: QSARs, reliability, biodegradability prediction |
