ave intrinsic limitations. Bioinformatic prediction tools are trained on sets of previously described epitopes, but since the ava