Multiple Sequence Alignment Using Parallel Genetic Algorithms

1
Multiple Sequence Alignment Using Parallel
Genetic Algorithms

• Luke Ulrich Colin Bauer

2
Genetic Algorithms (GA)

• Search method characterized by
• Mechanism of natural evolution
• Ability to converge amidst many local minima
• Locate near-optimal solutions in large search
  space
• Inherently parallel

3
Parts of a GA

• A population of possible solutions
• Functions for manipulating individuals (e.G.
  Crossover and mutation)
• Evaluation function determine the fitness of
  each individual

4
GA Technique 1 (Bauer)

• Create initial population with random positions
  for each individual
• Evaluation of initial population
• Mutation
• Random repositioning
• Gibbs sampling
• Frame shifting
• Reproduction retain best/most-fit 50 of
  population unchanged
• Crossover generate remaining 50
• Evaluation
• Go to step 3 unless solution is acceptable

5
GA Technique 2 (Ulrich)
Initial Population (size 2)
Mutation
Gibbs, frameshift, etc
Cut Here
Cut Here
Crossover
Create 12 children


Total population 16
6
Evaluation


-25
-14
-112
-3
Winnow
Select the best children
-112
-25
Repeat process until satisfactory solution is
obtained.
7
Parallel Framework

• Decide Number of Children/Processor
• Decide Number of Processors
• Distribute sequences
• Each node executes GA for x iterations
• Gather all node evaluations to root process,
  which performs a global sort
• Root process determines which nodes work on
  subsequent alignments based on the respective
  fitness
• Continue until satisfactory alignment is obtained

8
The Plan
Parallel Model
GA Technique 1
GA Technique 2
Benchmark
Benchmark
Publish Results
Questions?