Foodstuff could be a vector for naturally occurring and/or unwanted dangerous substances that can act either as they are or after their bioactivation. The scientific community agrees that the metabolic activity of chemicals should be taken into account for proper risk assessment. Unfortunately, the in vitro evaluation of a metabolic panel and analytical/biochemical detection in food-safety assessment are very expensive and challenging because of the abundance of data to analyze. In this context, properly validated computational protocols could be a useful tool for making metabolic and binding/activity predictions. This strategy has been applied to thioxanthone photoinitiators (TX), identified as food contaminants, especially in infant formulas, as reported by the European Food Safety Authority in 2005. Their lipophilicity suggests rapid hepatic metabolism, but the currently available data only concern 2-ITX. We have predicted phase I metabolites for the TX class of compounds and defined their binding affinity for the AR ligand-binding pocket using a local model based on available information about metabolism and AR activity. Some metabolites should undergo further in vitro or/and in vivo toxicological evaluations because they have proved to be suitable as ligands for AR.