Progress in toxinformatics: the challenge of predicting acute toxicity

Historically, acute toxicity based on LC(50) and LD(50) tests has been analyzed using various quantitative structure-activity relationships (QSARs). The obtained QSAR equations cannot be related to the multiple health effects reflected in the experimental data of analyzed compounds. Therefore little predictive power for novel compounds can be achieved. New methods based on classification SAR (C-SAR) analysis offer new mechanistic knowledge that can be related to new health effects, resulting in better predictive power. To this end, a very careful interpretation of the obtained results is required, implying the use of the existing mechanistic information to the maximum possible extent. The current mini-review aims at determining ways of automated extraction of new mechanistic knowledge from existing data, as well as ways of relating this knowledge to various health effects. A comparison of "statistical induction" and "knowledge-based" approaches is provided. The existing and future developments in predictive acute toxicity are discussed.