Rapid computation of distance estimators from nucleotide and amino acid alignments

Abstract

Distance estimators are needed as input for popular distance based phylogenetic reconstruction methods such as UPGMA and neighbour-joining. Computation of these takes O(n2l) time for n sequences with length l which is usually fast compared to reconstructing a phylogenetic tree of n taxa. However, with the introduction of fast search heuristics for distance based phylogenetic reconstruction methods, the computation of distance estimators has become a bottleneck especially for long sequences. Elias et al. have shown how distance estimators can be computed efficiently from unaligned nucleotide sequences using vectorisation of code. In this paper we extend their method to allow efficient computation of distance estimators from aligned nucleotide and amino acid sequences using vectorisation of code and parallelisation on both CPUs and GPUs. Experiments are presented which show an increase in performance of up to 36x and 8x relative to the naive approach when computing distance estimators from nucleotides and amino acids alignments respectively.