Finding DNA Regulatory Motifs with Position-dependent Models
Huihai Wu1, Prudence W.H. Wong2, Mark X. Caddick2, and Chris Sibthorp2
1. University of Surrey, Guildford, United Kingdom
2. University of Liverpool, Liverpool, United Kingdom
2. University of Liverpool, Liverpool, United Kingdom
Abstract—We consider the problem of de novo DNA motifdiscovery. The position weight matrix (PWM) model hasbeen extensively used, yet this model makes the assumptionthat nucleotides at different positions are independent ofeach other. Recent results have shown that nucleotidesbound by transcription factors often exhibit adjacent ornonadjacent dependencies. We address this problem bydevising positional dependency models capable of capturingadjacent dependencies and non-adjacent dependencies(SPWDM). Our algorithms are based on Gibbs sampling toupdate the model parameter and dependencies structure.We compare two scoring functions: c2-score and aconditional probability based score. We also improveseveral Gibbs sampling stages. Experiments are carried outon simulated and real data, showing that the SPWDMmodel makes improvement over pure PWM. Themodifications to the Gibbs sampling algorithm are alsoshown to be effective.
Index Terms—DNA motif discovery, position weight matrix,position-dependent model, Gibbs sampling
Cite: Huihai Wu, Prudence W.H. Wong, Mark X. Caddick, and Chris Sibthorp, "Finding DNA Regulatory Motifs with Position-dependent Models", Journal of Medical and Bioengineering, vol. 2, no. 2, pp.103-109, 2013. doi: 10.12720/jomb.2.2.103-109
Index Terms—DNA motif discovery, position weight matrix,position-dependent model, Gibbs sampling
Cite: Huihai Wu, Prudence W.H. Wong, Mark X. Caddick, and Chris Sibthorp, "Finding DNA Regulatory Motifs with Position-dependent Models", Journal of Medical and Bioengineering, vol. 2, no. 2, pp.103-109, 2013. doi: 10.12720/jomb.2.2.103-109
Array