TFBS Identification with Genetic Algorithms using Suffix Tree

1
TFBS Identification with Genetic Algorithms using
Suffix Tree

• Cyrus Chan

2
Biological Background
Transcription Factor
Expressed
Coexpressed Genes
TFBS motifs
3
Problem Description
Upstream Sequences
4
Related Methods

• Deterministic Methods
• (l, d)-motif discovery problem
• Weakly conserved motifs in real problemsl?
• Machine learning methods
• Training data or strong prior knowledge
• May bias the training data

5
Related Works

• Genetic Algorithms / Evolutionary Computation
• Consensus-led
• Evaluate the fitness of aligned subsequences
• Rely on additional techniques Alignments
  Clustering
• Positions-led
• Evaluate the fitness with a position matrix
• Can not locate similar motifs fast

6
GAST

• Genetic Algorithm using Suffix Tree (GAST)
• Combines positions and consensus
• Suffix Tree Content Addressable
• Generalized Suffix Tree For multiple sequences
• Locate motifs quickly
• Positions diversity
• Based on the similarities between different
  motifsl need not be specified

7
Representation
Individual
Consensus
Positions
8
Genetic Operators

• Mutation
• Type1 randomly mutate the consensus with the
  positions not completely preserved
• Use suffix tree
• Type2 single mutation of the positions within an
  individual
• Multi-point?

9
Genetic Operators

• Single-point Crossover
• Performed on individuals in representation of
  positions

10
Fitness Function

• Information Content
• G. B. Fogels Fitness score is also tried

Where fb,i is the observed frequency of
nucleotide b on the column i and pb (0.25) is
the background frequency of the same nucleotide.
The summation is taken over the four possible
types of nucleotides (b? A, T, C, G). l is the
motif length.
11
Selection

• K- Tournament
• Elitism is tried but the result is not good
  premature convergence
• K- Tournament adopted from G. B. Fogels work
• Each individual competes with K random ones
• Rank individuals by the wins
• Re-rank by Fitness if there is a tie

12
Rearrangement

• Take Advantage of Suffix Tree
• Rank each motif within an individual according to
  its similarity to the consensus
• Locate the occurrence of motifs in different
  sequences with Suffix Tree from highest to lowest
  similarity

13
Other Issues

• Deal with TATA box and Poly-A
• Penalize the fitness of an individual with T, A
  percentage higher than a user defined threshold
• Phase problem
• The solution is a shifted version of the optimal
  one
• Can not be handled by genetic operators
• Shift based on the gain of Information Content

14
Preliminary Experiments

• The experiment of GACluster Paul et al GECCO06
• CRP motifs
• The method with 3 motifs found outperforms the
  binary GA, which finds no real motifs

True Motifs
15
Preliminary Experiments

• GAST is able to identify 5 motifs in the
  sequences (Population 1000 Generation 500)
• CRP 18 sequences of 105 bp
• Locate some of the motifs
• More generation is needed

16
Discussion

• Mutation Rate /Crossover Rate
• Convergence
• Evaluate Fitness?
• Structured motifs / Modules

Expressed
17
TFBS Identification with Genetic Algorithms using
Suffix Tree

• Thank You!
• Cyrus Chan