Baldi

Hidden Markov Models.

Hidden Markov models (HMMs) are flexible probabilistic models that generalize the notion of profile and are applicable to a wide range of sequence analysis problems. With efficient learning algorithms, HMMs can be applied to both protein and DNA/RNA sequences in:

	   1. local and global alignments, multiple alignments;
	   2. classification and data mining;
	   3. structural analysis and pattern discovery;
	      HMM librairies can also be constructed and combined in hierarchical
	      and modular fashion.
This tutorial briefly covers the main statistical theory behind HMMs and concentrates on "hands-on" applications to specific problems demonstrated using, among other things, a computer graphical interface to an HMM simulator program. Examples of applications include modeling and aligning protein families, using HMM protein libraires such as Pfam, predicting splice sites, promoters, and genes, discovering periodic patterns in DNA, and DNA structural analysis (bendability, nucleosome positioning,...). Finally, extensions of HMMs such as Input-Output HMMs, interpolated Markov models, and stochastic grammars are discussed.

Reference: P. Baldi and S. Brunak (1998) "Bioinformatics: the Machine Learning Approach" MIT Press.