A structured outputs method for predicting protein binding sites

Protein-protein interactions have essential roles in nearly all biochemical processes. While high-throughput methods exist for ex- perimentally identifying interaction partners, the task of determining binding site locations remains arduous. We consider the prediction of protein binding site location as an instance of the label sequence problem and outline a representation in the framework of structured outputs. Moreover, we compare kernel struc- tured output methods with sliding window classiers in the identication of calmodulin- binding sites.