Finding sequence motifs with Bayesian models incorporating positional information: an application to transcription factor bindin