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

Area of Science:

• Toxicology
• Computational Chemistry
• Risk Assessment

Background:

• Lack of toxicological data for many substances, including naturally occurring ones and those under new European chemicals legislation, necessitates reliable prediction tools.
• Quantitative toxicological data is vital for the risk assessment of substances with repeated human exposure.

Purpose of the Study:

• To evaluate the predictive accuracy of the TOPKAT 6.2 software for (sub)chronic oral lowest observed adverse level (LOAEL) in industrial chemicals.
• To compare TOPKAT's predicted LOAELs with experimentally derived values from repeated dose animal studies.

Main Methods:

• Utilized the TOPKAT 6.2 software, which contains a (sub)chronic oral LOAEL prediction model trained on 393 chemicals.
• Tested the model on 807 industrial chemicals (>95% purity), comparing predicted LOAELs against experimental data.
• Analyzed prediction performance based on exclusion criteria and prediction accuracy within defined fold-differences.

Main Results:

• Predictions could not be performed for 460 out of 807 chemicals due to exclusion criteria (e.g., low LD50, outside domain of applicability, training set inclusion).
• For the remaining 347 chemicals, 34-62% of predicted LOAELs were within a 5-fold difference of experimental values.
• A broader range (1/100 to 100-fold) encompassed 84-99% of predictions, indicating significant uncertainty.

Conclusions:

• The TOPKAT software demonstrates limited accuracy for predicting (sub)chronic oral LOAELs, with substantial uncertainty.
• A refined prediction tool is needed for reliable risk assessment.
• Applying a large uncertainty factor (e.g., 10,000) may be necessary to utilize predicted NOAELs for risk assessment.