Modelling bioaccumulation processes at the food web level is the main step to analyse the effects of pollutants at the global ecosystem level. A crucial question is understanding which species play a key role in the trophic transfer of contaminants to disclose the contribution of feeding linkages and the importance of trophic dependencies in bioaccumulation dynamics. In this work we present a computational framework to model the bioaccumulation of organic chemicals in aquatic food webs, and to discover key species in polluted ecosystems. As a result, we reconstruct the first PCBs bioaccumulation model of the Adriatic food web, estimated after an extensive review of published concentration data. We define a novel index aimed to identify the key species in contaminated networks, Sensitivity Centrality, and based on sensitivity analysis. The index is computed from a dynamic ODE model parametrised from the estimated PCBs bioaccumulation model and compared with a set of established trophic indices of centrality. Results evidence the occurrence of PCBs biomagnification in the Adriatic food web, and highlight the dependence of bioaccumulation on trophic dynamics and external factors like fishing activity. We demonstrate the effectiveness of the introduced Sensitivity Centrality in identifying the set of species with the highest impact on the total contaminant flows and on the efficiency of contaminant transport within the food web.