Hidden Markov models

Yuan et al. (53) presented a hidden Markov model approach for selecting differentially expressed genes from replicated time course experiments with multiple biological conditions. They considered all possible equality and inequality relations among means across biological conditions as states, and the expression pattern process was modeled as a Markov chain, with either time-homogeneous or non-homogeneous transition matrices. The observations were conditionally independent given the state of the chain. In this approach, dependence between gene expression values at different times was completely described by the pattern process (i.e., the hidden Markov chain), and genes were selected based on the posterior probabilities of states of interest. This is an example of using an HMM to model time-dependence in microarray time course data, while many others have used HMMs in this context for clustering.

0 0

Post a comment