Quantitative Structure-Activity Relationshipsin Predicting Chemical Toxicity:When is the Prediction Reliable?

摘要

The possibility of predicting molecular properties before the molecule is even syn-thesized is perhaps the most beneficial application of quantitative structure-activity relationship(QSAR) studies. However, before a QSAR-based prediction for a query chemical can be acceptedwith confidence, it needs to be ascertained that 1) the predictive QSAR model is statisticallyrobust, 2) the QSAR model is applicable to the query chemical, and 3) the chemicals "similar" tothe query are correctly predicted by the QSAR models. This chapter lists steps and methods nec-essary to ascertain the stability and robustness of a QSAR model. This chapter also describes de-termination of the optimum prediction space (OPS) associated with a QSAR model; OPS is thespace inside and near the vicinity of which a QSAR model is applicable. The paper further intro-duces the concept of "QSAR similarity" as opposed to "structure similarity" to evaluate predic-tion accuracy on compounds with QSAR similarity to the query.