Chronic oral LOAEL prediction by using a commercially available computational QSAR tool

In the absence of toxicological data, as it is the case for, e.g. naturally occurring substances and chemicals underlying the new European chemicals legislation, distinct tools to derive quantitative toxicological data are of particular interest with regard to risk assessment of substances humans are repeatedly exposed. The software package TOPKAT 6.2 version 3.1 (Accelrys Inc., San Diego, USA) is a commercially available tool containing a (sub)chronic oral low observed adverse level (LOAEL) prediction model constructed by using structures and LOAELs of 393 chemicals contained in publicly accessible data banks. Applying this tool, we tested the prediction of (sub)chronic LOAELS for 807 industrial chemicals (purity >or= 95%) by comparing the predicted values with their experimental LOAELs derived from repeated dose animal experiments performed according to standard guidelines. For 460 chemicals, a prediction could not be performed because of exclusion criteria defined in the system. They had either a lower LD50 as the predicted LOAEL (n = 214) were outside the optimum prediction space which defines the domain of applicability (n = 175), were used in the training data set (n = 155), were not known to the system (n = 50) or fulfilled other criteria for data exclusion (n = 21). Of the remaining 347 substances, 34 to 62% LOAELs were predicted within a range of 1/5 and fivefold of the experimental LOAEL (factor 5), whereas 84 and 99% of the predicted LOAELs were within a range of 1/100 and 100-fold indicating high uncertainty of the prediction. Hence, a refined prediction tool is highly warranted. However, the uncertainty of the prediction could be accounted for if an additional factor of 100 is applied in addition to standard default adjustment factor of 100 which would result in an adjustment factor of 10,000 to be able to use a predicted NOAEL for risk assessment.