Genomewide motif identification using a dictionary model

This paper surveys and extends models and algorithms for identifying binding sites in noncoding regions of DNA. Binding sites control the transcription of genes into messenger RNA in preparation for translation into proteins. The base sequence of most binding sites is not entirely fixed, with the different permitted spellings collectively constituting a "motif." After summarizing the underlying biological issues, we review three different models for binding site identification. Each model was developed with a different type of dataset as reference. We then present a unified model that borrows from the previous ones and integrates their main features. In our unified model, one can identify motifs and their unknown positions along a sequence. One can also fit the model to data using maximum likelihood and maximum a posteriori algorithms. These algorithms rely on recursive formulas and the maximization/minorization principle. Finally, we conclude with a prospectus of future data analyses and theoretical research.