Measures on Hidden Markov Models

Hidden Markov models were introduced in the beginning ofthe 1970's as a tool in speech recognition. During the last decadethey have been found useful in addressing problems in computationalbiology such as characterising sequence families, gene finding,structure prediction and phylogenetic analysis. In this paperwe propose several measures between hidden Markov models. Wegive an efficient algorithm that computes the measures for leftrightmodels, e.g. profile hidden Markov models, and discuss howto extend the algorithm to other types of models. We present anexperiment using the measures to compare hidden Markov modelsfor three classes of signal peptides