Open Collections

Achar, Jerry Collince

University of British Columbia

Improving regulatory confidence in and acceptance of in silico toxicology methods and their predictions requires assessment and transparent communication of associated uncertainty to facilitate the evaluation of whether they are fit for purpose. This thesis develops frameworks and methods to facilitate systematic and transparent analysis of and accounting for uncertainty in in silico toxicology methods, with a central focus on QSARs. This is done through four studies. Studies 1 and 2 conduct a literature review to identify key components relevant to in silico toxicology problem formulation and modeling processes, which are then systematically categorized and rationalized as potential sources of uncertainty. The outcomes from the studies are four problem formulation components (as in Study 1) and 20 sources of uncertainty (as in Study 2). Study 3 focuses on analyzing implicit and explicit uncertainties expressed within QSAR studies predicting neurotoxicity of chemicals. To this end, implicit and explicit uncertainty indicators are identified, whereafter, the indicators are used to identify uncertainties. A systematic categorization of the uncertainties, according to the 20 uncertainty sources in Study 2, reveals that implicit uncertainty is expressed at a higher rate (64%) in the analyzed studies and within most uncertainty sources (65%; 13/20), indicating that uncertainty is predominantly expressed implicitly in the field. Study 4 proposes a consensus method combining TEST and CATMoS model predictions to produce conservative predictions. The level of conservativeness of the consensus predictions against the individual models is evaluated based upon the agreement of predicted LD50-based GHS categories with the corresponding experimental LD₅₀-based GHS categories. The results show that the consensus method produces a higher over-prediction rate (39%; 2,504/6,410) than TEST (24%) or CATMoS (25%), while its under-prediction rate is lower at 8% than TEST (20%) or CATMoS (10%), which indicates that, by design, it is the most conservative. The outcomes from the four studies benefit the field of in silico toxicology by contributing to efforts aimed towards addressing the issue of uncertainty, promoting regulatory acceptance of models (e.g., QSARs) and their predictions, as well as reducing and (where possible) replacing use of animal in chemical toxicity assessment.

Text

2025-01-30

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/90233

Resources, Environment and Sustainability

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/