Editorial Board

JMI2013B-9 A quantitative activity-activity relationship model based on covariance structure analysis, and its use to infer the NOEL values of chemical substances (pp.151-159)

Author(s): Jun-ichi Takeshita, Masashi Gamo, Koji Kanefuji and Hiroe Tsubaki

J. Math-for-Ind. 5B (2013) 151-159.

Abstract
Here we prove the usefulness of the quantitative activity-activity relationship (QAAR) approach by giving an example of its practical application. To assess and manage chemical substances, experiments are often done on animals. However, in terms of time and cost efficiencies and the need for animal protection, there is increasing global demand to use statistical methods to reduce the need for animal testing. Although the QAAR approach has been introduced to estimate unknown toxicity values from the relationships between different toxicity endpoints (item for observation of hazardousness), there are few examples of its practical application. We considered the QAAR approach was useful from the viewpoint of effective utilization of existing data. We therefore adopted covariance structure analysis as a statistical method to develop our QAAR model for inferring the missing no-observable-effect level (NOEL) values for each target (e.g. internal organ) in animal testing data from repeated dose toxicity studies and reproductive toxicity studies. We emphasize here that the data were sparse. One of the major advantages of our model is that it enables us to make estimations by using confidence intervals, which means that we not only infer the missing NOEL values but also quantify their uncertainty. As a specific example, we discuss toluene, for which there were 19 missing NOELs out of 48 endpoints. Finally, we discuss the validation of the model in accordance with the Organisation for Economic Co-operation and Development's Principles for Quantitative Structure-Activity Relationship models, and we conclude that the accuracy of this model is high.

Keyword(s).  animal testing data, covariance structure analysis, no-observable-effect-level (NOEL), prediction of missing values, quantitative activity-activity relationship (QAAR)