Promoter regions of a genome may involve structural signals, as well as primary sequence patterns, for protein recognition of DNA.  Well-characterized systems suggest that the process involves a combination of DNA axis bending, flexibility, and bendability.  With the advent of the human genome project, constructing models of genomic DNA based on flexibility and other related properties of the dynamical structure of a DNA sequence accurately reflecting the current level of knowledge is a matter of potential significance at the junction of the fields of structural biology, bioinformatics, and genomics.
Current models heavily endowed in the idea of one conformation for each base pair step do not take into account the dynamic properties of DNA, and a separate property such as bendability needs to be defined in terms if external perturbation.  The distribution of conformations of DNA base pair steps determined by Molecular Dynamics (MD) simulations has been previously thought of in terms of Gaussian statistics with averages and standard deviations despite the fact they present highly irregular distributions, calling into question the exact consequence of these analyses.  
In this project, independent of any requirements for regular distributions, a new statistical means of treating thermal dispersions of base pair conformations determined by MD simulations has been developed and integrated into Hidden Markov Models (HMMs). HMMs recognize and score patterns arising in multiple variations each with a statistical weight.  As such, these flexibility HMMs have been employed as a means to determine to what extent the flexibility of base pair steps of DNA in solution can be used to characterize proteins? binding sites for model cases and to extend this to more complex systems of biological and medical significance.  We have first sought to validate our methods by examining the well-characterized system of Catabolite Activator Protein (CAP) and its DNA binding sites.  

Based on a statistical treatment of MD derived flexibility parameters of a subset of known Catabolite Activator Protein (CAP) binding sites, HMMs demonstrate the ability to distinguish between binding and non-binding sites.  Thus the contribution of flexibility to the binding event suffices to discriminate between binding and non-binding sites.  Improved discrimination was obtained when sequence information was provided exclusively where a sequence-specific contact is made between the protein and DNA.  This equaled the ability of the sequence-based method with the added advantage that it can find matches which are unlikely candidates based on sequence search.

Ultimately we envision routinely scanning entire genomes for novel binding sites based on flexibility.  Here we present some preliminary figures on the first flexibility scan of a genome, that of e.coli K12 for CAP binding sites.