Motif Discovery

1
Motif Discovery

• Presented by
• Chase A. Jahn
• CS329E
• April 29, 2008

2
Outline

• Motif discovery problem
• Search methods
• Exhaustive
• Gibbs Sampling

3
What is a motif?
Problem Description

• Short sequences of DNA or RNA (or amino acids)
• Often consist of 5-16 residues
• May contain gaps
• Examples include
• Transcription factor binding sites
• Aide in identifying gene networks
• Splice sites
• Start/stop codons
• Phosphorylation sites
• Coiled-coil domains

4
Motif Problem
Problem Description

• Given sequences
• Find motif
• The number of motifs
• The width of each motif
• The locations of motif occurrences

5
Inherent Difficulties
Problem Description

• Very long input sequences
• Thousands to Millions of residues
• Length variation
• Motif may be short, or only partially similar in
  all input sequences

6
Exhaustive Searching
Exhaustive

• BruteForceMotifSearch(DNA,t,n,l)
• bestScore 0
• For each (s1.st) from (1,.1) to (n-l1,,n-l1)
• If Score(s,DNA) gt bestScore
• bestScore Score(s,DNA)
• bestMotif (s1, s2,,st)
• Return bestMotif
• Running time
• Number of positions
• Score(s,DNA) O(l)
• Overall

7
Gibbs Sampling

• Stochastically select starting positions for N
  l-mers (N5)

Gibbs Sampling
8
Gibbs Sampling

• Stochastically select one sequence
• Generate profile P from the remaining (N-1)
  l-mers
• For each position in the chosen sequence,
  calculate the probability that the position is
  generated by P

Gibbs Sampling
9
Gibbs Sampling

• Scoring Profile P is a position weight matrix
  (PWM) which contains log likelihood weights for
  computing a match score

Gibbs Sampling
MacIsaac, Practical Strategies for Discovering
Regulatory DNA Sequence Motifs, 2006.
10
Gibbs Sampling

• Choose a new starting position in the chosen
  sequence randomly
• Compute its score, accept the replacement
  depending on the score.

Gibbs Sampling
11
Review
Best Location
Gibbs Sampling
New Location

• Start with random motif locations and calculate
  a motif model
• Randomly select a sequence, remove its motif and
  recalculate tempory model
• With temporary model, calculate probability of
  motif at each position on sequence
• Select new position based on this distribution
• Update model and Iterate

12
Summary

• Gibbs sampling works well and is more efficient
  than the exhaustive technique
• May converge to local optima if motifs are subtle
  or residue distribution is skewed

Summary
13
Questions?
Questions