Parallel implementation of maximum likelihood methods for phylogenetic analysis

Summary form only given. Here we describe a new parallel program package for phylogenetic analysis of DNA sequences. This program is based on a more advanced algorithm, named advanced stepwise addition (ASA), for phylogenetic analysis using maximum likelihood approaches. There are two main advantages of our parallel program package over many existing ones. Firstly the size of the tree search space can be freely chosen and so we are able to fill up a large-scale supercomputer (as long as it is computationally feasible) to alleviate the problems of thoroughness and phylogenetic uncertainty. Secondly we adopt SPMD programming technique in our implementation. Obvious advantages of SPMD technique over the simple master/workers technique are as follows: Firstly, communication procedures are deterministic. This effectively eliminates the requirement for detecting and requesting for communication between processors. Secondly, there is no need to communicate between processors after completion of each single tree likelihood evaluation. The communication takes place only at the end of each iteration by sending/receiving among the processors small messages for identifying the current k best trees and certain number of trees for balancing the workload. This can effectively reduce the overall communication costs and alleviate the problem of communication bottleneck caused by using the simple master/workers technique. Therefore, our approach is more suitable for execution on large-scale high-performance parallel computers.