High Performance Phylogenetic Analysis on CUDA-compatible GPUs

摘要

The operation of phylogenetic analysis aims to investigate the evolution and relationships among species. It is widely used in the fields of system biology and comparative genomics. However, phylogenetic analysis is also a computationally intensive operation as the number of tree topology grows in a factorial way with the number of species involved. Therefore, due to the large number of species in the real world, the computational burden has largely thwarted phylogenetic reconstruction. In this paper, we describe the detailed GPU-based multi-threaded design and implementation of a Markov Chain Monte Carlo (MCMC) maximum likelihood algorithm for phylogenetic analysis on a set of aligned nucleotide sequences. The implementation is based on the framework of the most widely used phylogenetic analysis tool, namely MrBayes. The proposed approach resulted in 6x-8x speed-up on an NVidia Geforce 460 GTX GPU compared to an optimized GPP-based software implementation running on a desktop computer with a single Intel Xeon 2.53 GHz CPU and 6.0 GB RAM.