Centipede

CENTIPEDE is the only one with an actual software package released. The benefit of their training on specific motifs is that there appears to be a unique DNase cutting pattern for each TF. This potentially allows for a more accurate prediction of the TF binding. In reality, I don't think the sequencing depth is such that you actually see the average pattern at each site. There is also the HMM method from Duke (in the same GR issue as CENTIPEDE) but that has no released software because the training was essentially impossible to automate.