Identification of functional cryptic recombination signals in the mouse 
genome by statistical modeling.

Lindsay Grey Cowell, Dept. of Biostatistics and Bioinformatics,
Duke University Laboratory of Computational Immunology,
Center for Bioinformatics and Computational Biology


Ninety-seven percent of the human genome is non-protein-coding, and 20-30%
of the non-coding sequence in eukaryotic genomes is expected to play a
role in gene regulation.  This suggests a significant role for the
regulation of gene expression in organismal complexity.  At the crux of
gene regulation is the cognate binding of DNA by regulatory proteins.
Although binding is sequence-specific, alignments of the known binding
sites for a protein show that binding-site DNA sequences are often highly
variable.  Consequently, one of the most challenging problems in
molecular/computational biology is the identification of binding sites and
the prediction of their functional level.  I will briefly discuss the
difficulties of modeling an alignment of known binding sites and describe
a novel method of model selection developed to model the binding sites
(recombination signals, RS) for the V(D)J recombinase.  V(D)J
recombination is a DNA recombination mechanism that generates 
antigen-receptor (eg. antibody) encoding genes.

A healthy immune response depends on a highly diverse antigen-receptor 
repertoire.  Genetic mechanisms have arisen that specifically diversify 
the genes encoding antigen receptors, primarily V(D)J recombination and 
somatic hypermutation.  Our statistical model of the V(D)J recombinase 
binding sites has identified novel sites of recombination and 
predicts their role in a previously unidentified mechanism of 
antigen-receptor gene diversification.