Segmentation with an Isochore Distribution

Abstract

Abstract. We introduce a novel generative probabilistic model for segmentation problems in molecular sequence analysis. All segmentations that satisfy given minimum segment length requirements are equally likely in the model. We show how segmentation-related problems can be solved with similar efficacy as in hidden Markov models. In particular, we show how the best segmentation, as well as posterior segment class probabilities in individual sequence positions can be computed in O(nC) time in case of C segment classes and a sequence of length n.