(94) Multiple Sequence Alignment by Genetic Algorithm
Machine: PIM/UNIX Machine
Environment: PIMOS/UNIX, KLIC
Language: KL1
Source Code: 0.3 MB
Documents: Manual (English / Japanese)
Overview
This parallel system, the combination of a genetic algorithm and an
iterative alignment algorithm, produces better results than iterative
aligners which employ hill-climbing search strategies.
Features
This parallel alignment system featuring a genetic algorithm can
improve an alignment score as rapidly as that with the hill-climbing
algorithm, and the system gradually increased the score to a higher
level which might be close to the optimal alignment score. The reason
why the genetic algorithm shows such high performance seems to be the
modularity of multiple alignment problems. If we can replace a part of
a sequence alignment with a better part from another, we obtain a
better multiple sequence alignment. Crossover operators achieve this
replacement in a statistical manner.
Function
In this system, each individual alignment is assigned to a PE, so the
size of population is 255 when using PIM/m with 256 PEs. A PE is used
as a manager process to select discarded individuals and crossover
candidates. The period of each generation is controlled with the
system timer, whose default value is eighty seconds. When each PE
finds that the system timer has passed eighty seconds, it does not
apply another mutation operator but sends the current alignment to the
manager PE. On average, four mutations occur in a generation with the
crossover operation, or six mutations occur in a generation without
it. The maximum alignment score among the population is decided as
having converged when it is invariable through more than one hundred
generations. It took about two hours to obtain the converged score
when solving a practical-scale alignment problem, which consists of
twenty-two sequences with eighty amino-acid letters each.
This system can also execute on computers with basic functions of UNIX
operating system using KLIC.
FTP
www-admin@icot.or.jp