next up previous
Next: Results of the maximum Up: Parallel execution Previous: Implementation of parallel computing

Experimentation of the maximum likelihood method in parallel computation

To investigate efficiency of our program in parallel environment, we used 5 actin amino acid sequences. The amino acids sequence were aligned using ClustalW [Thompson et al., 1994], and gap sites were removed (insertion/deletion detected sites). The value of $\epsilon$ for EM algorithm [Felsenstein, 1981] and the threshold value for branch traversing were set to 0.001 and 0.01, respectively. Using Cray CS-6400, the number of CPUs was altered in use gradually. The computational times were investigated by the provided predicate time/3 from KLIC 2.002.

Next, we aligned 10 MRF (Muscle Regulatory Factor) family genes using ClustalW [Thompson et al., 1994], and reconstructed the phylogenetic tree by the neighbor joining method [Saitou and Nei, 1987]. For the topology obtained by the NJ method, branch lengths were estimated by the maximum likelihood method, and compared them with those estimated in the NJ method. One and 20 CPUs were used to estimate the branch lengths, and investigated the speed-up effect.

Multiple alignments used here can be seen in our web site;

http://thinker.lab.nig.ac.jp/alignment/alignment.html



Satoshi OOta
1999-03-06