Exploring New Search Algorithms and Hardware for Phylogenetics: RAxML Meets the IBM Cell

Abstract

Phylogenetic inference is considered to be one of the grand challenges in Bioinformatics due to the immense computational requirements. RAxML is currently among the fastest and most accurate programs for phylogenetic tree inference under the Maximum Likelihood (ML) criterion. First, we introduce new tree search heuristics that accelerate RAxML by a factor of 2.43 while returning equally good trees. The performance of the new search algorithm has been assessed on 18 real-world datasets comprising 148 up to 4,843 DNA sequences. We then present the implementation, optimization, and evaluation of RAxML on the IBM Cell Broadband Engine. We address the problems and provide solutions pertaining to the optimization of floating point code, control flow, communication, and scheduling of multi-level parallelism on the Cell.

Deciphering microbial landscapes of fish eggs to mitigate emerging diseases

“…The evolutionary model used for the molecular data in MrBayes was HKY þ G and was obtained by running the data sets in jModelTest 2 (Darriba et al, 2012). Maximum Likelihood inference analysis was carried out with RaxmlGUI v7.4.2 (Silvestro and Michalak, 2012) with a random starting tree, included the GTRGAMMA option and employed the rapid hill-climbing algorithm (Stamatakis et al, 2007). Clade support was assessed with 1000 bootstrap replicates, with the rapid-hill climbing algorithm (Stamatakis et al, 2008).…”

Evolution of the hymenopteran megaradiation

“…When one group was not represented in the MCMC sample, we obtained a conservative estimate of the PMO by simply adding one sample to the missing group, based on the notion that the next MCMC sample could go against the signal seen in all the previous ones. RAxML analyses -A different likelihood approach was taken for the combined (EA and SS, and with or without nt3) and single gene data sets using RAxML v.7.0.0 (Stamatakis et al, 2007(Stamatakis et al, , 2008. Gene regions were partitioned for separate optimization of per-site substitution rates.…”

“…Ten randomized starting trees were generated to determine the initial rearrangement setting (-i) and number of distinct rate categories (-c). Independent searches of 1000 repetitions were used to find the best-known likelihood (BKL) tree and bootstrap searches using the ''rapid hill climbing algorithm'' (Stamatakis et al, 2007). Additional analyses, including single gene searches, were conducted using the CIPRES portal (http://www.phylo.org/sub_sections/portal/) and the rapid bootstrap search algorithm (RBS) (Stamatakis et al, 2008), in which bootstrap analyses are conducted first with 500 repetitions, followed by fast and then slow searches on the sampled trees to find the BKL tree.…”

Relações filogenéticas entre espécies de "Flectonotus" (Anura: Hemiphractidae) isoladas geograficamente reveladas por dados moleculares, de comportamento e morfológicos

“…ML analyses were conducted on the aligned sequence data in RAxML ver. 7.0.4 (Stamatakis 2006) using a random starting tree, the faster rapid hill-climbing algorithm proposed by Stamatakis et al (2007), and the GTR + Γ model of sequence evolution for each partition. (Nylander et al 2008), we discarded the first six million generations; estimated sample sizes (ESS) from the four combined runs were all above 300.…”