Hidden Markov Models and Their Applications in Bioinformatics

• Activity Number: 200
• Type: Contributed
• Date/Time: Tuesday, August 10, 2004 : 9:00 AM to 10:50 AM
• Sponsor: IMS
• Author(s): Sujay Datta*+
• Companies: Northern Michigan University

Abstract

A hidden Markov model (HMM) is a generalization of Markov chain models that allows for more modeling flexibility. The essential difference between a regular Markov chain and a hidden Markov model is that for the latter, there is no longer a deterministic one-to-one correspondence between the true state of the process and its observable manifestation. The true underlying states constitute a Markov chain but at every state, we observe a manifestation "probabilistically chosen" from a fixed and time-independent collection of possible manifestations. Over the years, a number of algorithms have been developed to efficiently compute the likelihood of an observed sequence of manifestations (given the parameters of the model), to estimate the true underlying state-sequence that maximizes the likelihood of a given sequence of manifestations, and so forth. These developments have been motivated by the rapidly expanding horizon of applications for these models in areas such as statistical genomics, proteomics and bioinformatics. A HMM is one of the most statistically sophisticated way of representing protein domain family sequences. It is useful for computing multiple sequence alignments.