GanCtrl: A Generative AI Approach to Derive Study-Aligned Synthetic Controls for Reducing Concurrent Control Animal...

Area of Science:

• Toxicology
• Artificial Intelligence
• Animal Welfare

Background:

• Minimizing animal use in research aligns with the 3Rs principles (Replacement, Reduction, Refinement).
• Existing virtual control group (VCG) methods using historical data face limitations due to cross-study variability and data heterogeneity.
• There is a need for VCG approaches that mitigate biases inherent in historical data.

Purpose of the Study:

• To introduce GanCtrl, a generative AI method for creating study-specific synthetic control data from time-matched treatment data.
• To evaluate GanCtrl's ability to generate synthetic controls that mitigate biases associated with historical VCGs.
• To assess the utility of GanCtrl in toxicity studies for reducing concurrent control animal usage.

Main Methods:

• GanCtrl, a generative AI model, was developed to infer control data directly from concurrent treatment data.
• The approach was applied to rat repeat-dose toxicity studies, simulating 38 clinical pathology endpoints.
• Synthetic controls were generated and compared against real concurrent controls and VCGs derived from historical data.

Main Results:

• Synthetic controls generated by GanCtrl closely approximated real concurrent controls.
• The variability of GanCtrl-derived controls was comparable to biological replicate variability and within intra- and inter-laboratory baseline variations.
• GanCtrl-generated controls successfully identified drug-induced toxicity signals and preserved relationships between clinical pathology endpoints (e.g., ALT-AST).
• GanCtrl outperformed historical VCGs when combining data from multiple laboratories.

Conclusions:

• GanCtrl offers a promising approach for generating study-specific virtual control groups.
• This generative AI method has the potential to significantly reduce the number of concurrent control animals used in in vivo studies.
• GanCtrl supports the advancement of the 3Rs principles in toxicological research.