A matter of trust: Learning lessons about causality will make qAOPs credible

Area of Science:

• Toxicology
• Computational Biology
• Risk Assessment

Background:

• The field of toxicology is transitioning from traditional animal testing to Next Generation Risk Assessment (NGRA).
• NGRA leverages large in vitro datasets and computational modeling, often employing machine learning.
• A key challenge for NGRA models is gaining end-user credibility despite their predictive power.

Purpose of the Study:

• To propose causal inference and reasoning as a method to enhance the development, use, and acceptance of quantitative Adverse Outcome Pathway (AOP) models.
• To argue for the integration of Judea Pearl's causality concepts into toxicological modeling.
• To accelerate a "Toxicological Revolution" by adopting principles from the "Causal Revolution".

Main Methods:

• Review and commentary on the application of causal inference principles to toxicological risk assessment.
• Exploration of how established causality concepts can be integrated into quantitative AOP modeling.
• Discussion of the potential for "constructive disruption" within the current toxicological paradigm.

Main Results:

• Causal inference provides a framework to improve the robustness and interpretability of computational toxicology models.
• Adopting causality science can bridge the gap between predictive model performance and end-user trust.
• The integration of causal reasoning is expected to facilitate the acceptance of non-animal testing approaches.

Conclusions:

• The science of causal inference offers a pathway to increase the credibility and adoption of Next Generation Risk Assessment models.
• Importing causality concepts can significantly advance the "Toxicological Revolution" towards more reliable, non-animal-based risk assessments.
• This approach promises to make computational toxicology models more trustworthy and actionable for end-users.