Answers are grounded in this paper and cite their sources.
Accurate prediction of drug terminal elimination half-lives (telim1/2) in edible tissues is essential for establishing withdrawal intervals to protect animal-derived food safety. However, the complexity of drug–species–tissue–route combinations render comprehensive experimental characterization infeasible, while data scarcity for minor food-producing species limits traditional single-species modeling. This study presents a hybrid multi-task transformer framework that simultaneously predicts tissue-specific telim1/2 across six food-producing species (cattle, swine, chickens, turkeys, sheep, and goats) using a ChemBERTa chemical language model integrated with pharmacokinetic covariates. A systematic comparison of four scenarios crossing learning strategy (single-task vs. multi-task) with dataset composition (N = 1625 vs. N = 4044) revealed that multi-task learning achieved the best performance across all species, with a global R² = 0.71. The largest gain observed for data-sparse species was from R² = 0.06 to 0.90 (+0.84) for turkeys. SHAP analysis identified tissue matrix as the dominant predictor (mean |SHAP| = 0.83). These results suggest that multi-task deep learning enables effective cross-species inductive transfer, overcoming data scarcity limitations in chemical residue safety assessments. To bridge the gap between computational modeling and real-world application, the model was deployed as VetHLP: an open-access interactive web portal for instant, cross-species pharmacokinetic predictions for food safety assessment.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: