Discovering Motif Interactions in Promoter Regions

1
Discovering Motif Interactions in Promoter Regions
We seek to understand the rules governing the
organization of cis-regulatory elements
in promoter regions of saccharomyces cerevisiae
(yeast). Starting with a library of known
motifs, we construct a Hidden Markov Model where
states correspond either to positions
within different motifs or to background states.
By learning the transition probabilities of
this HMM we uncover preferences in ordering and
orientations between cis-regulatory elements.
Ben Szekely1,2, and Nir Friedman1 (1) Harvard
University, Cambridge (2) IBM Internet
Technology, Cambridge
1
Gene Regulation

• Gene expression is regulated by transcription
  factors that bind in the promoter region.
• Each transcription factor binds to specific DNA
  sequences. A motif describes the characteristics
  of binding sites of a specific factor. Ex, ACE1

Motivation Understand the relative positioning
of binding sites
STE12 ? TEC1
8
9
Classifying Sequences
Conclusion

• Sequences may belong to groups, ex. GO
  annotations
• Given a set of groups, train a model for each
  group. These models can decide which group a
  sequence probably belongs to based on likelihood
• Does this work with our model?
• We choose 7 GO annotation groups, train 7
  models, then reclassify and see how well we do.
• Result 95 accuracy

• Conclusions
• Few motifs dominate in yeast genome, some
  interesting isolated pairs
• Motif-motif transitions are statistically
  significant and exist in real sequences
• Model classifies sequences effectively

• Future Work
• Apply methods to specific biological problems
• Further statistical analysis

• Acknowledgements
• Wing Yung and Ed Geraghty for grid computing
  support
• Alister Lewis-Bowen for poster consultation

References Ashburner, M. et al. Gene ontology
tool for the unification of biology. The Gene
Ontology Consortium Nat. Genet. 2525-29
2000. Harbison et al. Transcriptional regulatory
code of a eukaryotic genome. Nature. 43199-104.
2004
Log Likelihood Compared to Background