## Computational grand challenges in assembling the tree of life: Problems & solutions (2005)

Venue: | THE IEEE AND ACM SUPERCOMPUTING CONFERENCE 2005 (SC2005) TUTORIAL |

Citations: | 7 - 0 self |

### BibTeX

@INPROCEEDINGS{Bader05computationalgrand,

author = {David A. Bader and Usman Roshan and Alexandros Stamatakis},

title = {Computational grand challenges in assembling the tree of life: Problems & solutions},

booktitle = {THE IEEE AND ACM SUPERCOMPUTING CONFERENCE 2005 (SC2005) TUTORIAL},

year = {2005},

publisher = {}

}

### OpenURL

### Abstract

The computation of ever larger as well as more accurate phylogenetic (evolutionary) trees with the ultimate goal to compute the tree of life represents one of the grand challenges in High Performance Computing (HPC) Bioinformatics. Unfortunately, the size of trees which can be computed in reasonable time based on elaborate evolutionary models is limited by the severe computational cost inherent to these methods. There exist two orthogonal research directions to overcome this challenging computational burden: First, the development of novel, faster, and more accurate heuristic algorithms and second, the application of high performance computing techniques. The goal of this chapter is to provide a comprehensive introduction to the field of computational evolutionary biology to an audience with computing background, interested in participating in research and/or commercial applications of this field. Moreover, we will cover leading-edge technical and algorithmic developments in the field and discuss open problems and potential solutions.

### Citations

2284 |
The neighbor-joining method: A new method for reconstructing phylogenetic trees
- SAITOU, NEI
- 1987
(Show Context)
Citation Context ...quence of the NP-hardness of the problems) run quickly, but may offer no quality guarantees and may not even have a well-defined optimization criterion, such as the popular neighbor-joining heuristic =-=[38]-=-. Optimization based on the criterion of maximum parsimony (MP) [21] seeks the phylogeny with the least total amount of change needed to explain modern data. Finally, optimization based on the criteri... |

1861 |
Numerical recipes in C: the art of scientific computing
- Press, Teukolsky, et al.
- 1992
(Show Context)
Citation Context ...Due to this observation Brent has recently been replaced by Newton–Raphson in the new version of IQPNNI [89] (Version 3.0). In addition, the latest version of IQPNNI also incorporates the BFGS method =-=[95]-=- for multi–dimensional optimization of model parameters (Bui Quang Minh, personal communication). BFGS is very efficient for parameter– rich models such as GTR+Γ or complex protein models, in comparis... |

1472 |
Phylogenetic Analysis Using Parsimony (*and Other Methods), vers. 4.0b10. Sinauer Associates
- Swofford
(Show Context)
Citation Context ...ct of Parallelization There are a variety of software packages to reconstruct sequence-based phylogeny. The most popular phylogeny software suites that contain parsimony methods are PAUP* by Swofford =-=[20]-=-, PHYLIP by Felsenstein [21], and TNT and NONA by Goloboff [22, 23]. We have developed a freely-available sharedmemory code for computing MP, that is part of our software suite, GRAPPA (Genome Rearran... |

1238 |
Phylip (phylogeny inference package) version 3.5c. distributed by the author. department of genetics, university of
- Felsenstein
- 1993
(Show Context)
Citation Context ...are a variety of software packages to reconstruct sequence-based phylogeny. The most popular phylogeny software suites that contain parsimony methods are PAUP* by Swofford [20], PHYLIP by Felsenstein =-=[21]-=-, and TNT and NONA by Goloboff [22, 23]. We have developed a freely-available sharedmemory code for computing MP, that is part of our software suite, GRAPPA (Genome Rearrangement Analysis through Pars... |

1130 |
Algorithmic Graph Theory and Perfect Graphs
- Golumbic
- 1984
(Show Context)
Citation Context ...rmed on “short subtrees” around each edge. Since the short subtree graph G is triangulated, we can find maximal clique separators in polynomial time (these are just cliques in the graph, as proven in =-=[41]-=-), and hence we can find (also in polynomial time) a clique separator X that minimizes maxi|X ∪ Ci|, whereG − X is the union of k components C1,C2,...,Ck. This is the same decomposition technique used... |

1095 |
Modeltest: testing the model of DNA substitution
- POSADA, CRANDALL
- 1998
(Show Context)
Citation Context ...ood ratio tests. Thus, if the likelihood obtained for a fixed tree topology with HKY85 is not significantly worse than the GTR-based likelihood value, HKY85 should be used. Programs such as Modeltest =-=[69]-=- can be applied to determine the appropriate model of evolution for a specific dataset. Another very important and rarely discussed modeling issue concerns the way rate heterogeneity among sites (alig... |

955 |
Evolutionary trees from DNA sequences : A maximum likelihood approach
- Felsenstein
- 1981
(Show Context)
Citation Context ...iterion of maximum parsimony (MP) [21] seeks the phylogeny with the least total amount of change needed to explain modern data. Finally, optimization based on the criterion of maximum likelihood (ML) =-=[22]-=- seeks the phylogeny that is the most likely to have given rise to the modern data. Heuristics are fast and often rival the optimization methods in terms of accuracy, at least on datasets of moderate ... |

850 | A simple, fast and accurate algorithm to estimate large phylogenies by maximal likelihood
- Guindon, Gascuel
- 2003
(Show Context)
Citation Context ...the high computational cost—in terms of floating point operations—of each tree evaluation per se. To date, the main focus of researchers has been on improving the search algorithms (RAxML [57], PHYML =-=[58]-=-, GAML [59], IQPNNI [46], MetaPIGA [60], Treefinder [61]) and on accelerating the likelihood function via algorithmic means by detecting and re-using previously computed values [62, 63]. Due to the al... |

681 |
MrBayes: Bayesian inference of phylogenetic trees
- HUELSENBECK, RONQUIST
- 2001
(Show Context)
Citation Context ...r 1,000,000 organisms based on the rather more conservative estimates. The increasing concern about memory consumption is also reflected by the recent changes introduced in the new release of MrBayes =-=[94]-=- (version 3.1.1). Despite the fact that MrBayes performs Bayesian inference of phylogenetic trees the underlying technical problems are the same since the likelihood value of alternative tree topologi... |

466 |
Evolution of protein molecules
- Jukes, Cantor
- 1969
(Show Context)
Citation Context ...versible model of nucleotide substitution (GTR [65] [66]) is reduced to 6 due to required symmetries. The less general time-reversible models of nucleotide substitution such as the Jukes-Cantor (JC69 =-=[67]-=-) or HasegawaKishino-Yano (HKY85 [68]) model can be derived from GTR by further restriction of possible transition types. It is important to note, that there exists a trade-off between speed and quali... |

446 |
Dating of the human-ape splitting by a molecular clock of mitochondria1 DNA sequences
- Hasegawa, Kishino, et al.
- 1985
(Show Context)
Citation Context ...tion (GTR [65] [66]) is reduced to 6 due to required symmetries. The less general time-reversible models of nucleotide substitution such as the Jukes-Cantor (JC69 [67]) or HasegawaKishino-Yano (HKY85 =-=[68]-=-) model can be derived from GTR by further restriction of possible transition types. It is important to note, that there exists a trade-off between speed and quality among substitution models. The sim... |

274 |
Molecular evolution and phylogenetics
- Nei, Kumar
- 2000
(Show Context)
Citation Context ...oted tree. Reconstruction of a phylogenetic tree is a statistical inference of a true phylogenetic tree, which is unknown. There are many methods to reconstruct phylogenetic trees from molecular data =-=[8]-=-. Common methods are classified into two major groups: criteria-based and direct methods. Criteriabased approaches assign a score to each phylogenetic tree according to some criteria (e.g., parsimony,... |

254 | Maximum likelihood phylogenetic estimation from dna sequences with variable rates over sites: approximate methods
- Yang
- 1994
(Show Context)
Citation Context ... vector entries have to be computed per site i. In the continuous case likelihoods must be integrated over the entire probability distribution. The most commonly used distributions are the continuous =-=[74]-=- and discrete [71] Γ distributions. Typically, a discrete Γ distribution with ρ = 4 points/rates is used since this represents an acceptable trade–off between inference time, memory consumption, and a... |

238 |
Stochastic Local Search: Foundations and Applications
- Hoos, Stützle
- 2004
(Show Context)
Citation Context ... reconstruction. It is an iterative procedures which applies an existing base method to both DCM3 subsets and the complete dataset. Rec-I-DCM3 can also be viewed as an iterated local search technique =-=[42]-=- which uses a DCM3 decomposition to escape local minima. Figure 7 provides a full description of the Rec-I-DCM3 algorithm. The Recursive-DCM3 routine performs the work of dividing the dataset into sma... |

182 |
Phylogenetic inference
- Swofford, Olsen, et al.
- 1996
(Show Context)
Citation Context ...ques). A phylogeny (phylogenetic tree) is usually a rooted or unrooted bifurcating tree with leaves labeled with species, or more precisely with taxonomic units (called taxa) that distinguish species =-=[7]-=-. Locating the root of the evolutionary tree is scientifically difficult so a reconstruction method only recovers the topology of the unrooted tree. Reconstruction of a phylogenetic tree is a statisti... |

168 |
fastDNAml : a tool for construction of phylogenetic trees of DNA sequences using maximum likelihood",CABIOS
- Olsen, Matsuda, et al.
- 1994
(Show Context)
Citation Context ...f a one-parameter function lh(t) wherelh is the phylogenetic likelihood function and t the branch length. Some of the most commonly used optimization methods are the NewtonRaphson method in fastDNAml =-=[70]-=- or Brent’s rule in PHYML [58]. Typically, the two basic operations: computation of the likelihood value and optimization of the branch lengths, require ≈ 90% of the complete execution time of every M... |

161 |
The general stochastic model of nucleotide substitution
- Rodríguez, Oliver, et al.
- 1990
(Show Context)
Citation Context ...identic if followed forward or backward in time. Essentially, this means that the maximum number of possible transition types in the General Time Reversible model of nucleotide substitution (GTR [65] =-=[66]-=-) is reduced to 6 due to required symmetries. The less general time-reversible models of nucleotide substitution such as the Jukes-Cantor (JC69 [67]) or HasegawaKishino-Yano (HKY85 [68]) model can be ... |

137 | Yano T: Dating of the human-ape splitting by a molecular clock of mitochondrial DNA - Hasegawa, Kishino - 1985 |

121 | ARB: a software environment for sequence data - Ludwig, Strunk, et al. - 2004 |

118 | MrBayes: Bayesian inference of phylogeny - HUELSENBECK, RONQUIST - 2001 |

115 | H: RAxML-III: A fast program for maximum likelihood-based inference of large phylogenetic trees
- Stamatakis, Ludwig, et al.
(Show Context)
Citation Context ...iency) and JAVA-based distributed systems will be discussed. In addition, the interdependence and benefits of simultaneous algorithmic and technical development will be outlined by example of RAxML-V =-=[41, 44]-=-. Finally, an overview over possible future HPC implementations of the novel algorithms is provided including Grid-based solutions, mixed MPI/OpenMP implementations for hybrid supercomputer architectu... |

112 |
Tree-Puzzle: maximum likelihood phylogenetic analysis using quartets and parallel computing
- STRIMMER, VINGRON, et al.
- 2002
(Show Context)
Citation Context ...ts. Due to this exhaustive search strategy the execution time increases more steeply with the number of taxa than in other programs. The largest tree computed with the parallel version of TREE-PUZZLE =-=[88]-=- contained 257 taxa due to limitations caused by the data structures used (Heiko Schmidt, personal communication). However, TREE-PUZZLE provides mainly advantages concerning quality-assessment for med... |

110 | A linear-time algorithm for computing inversion distance between signed permutations with an experimental study
- Bader, Moret, et al.
(Show Context)
Citation Context ...ore biologically-meaningful inversion phylogeny and provides a highly-optimized code that can make use of distributed- and shared-memory parallel systems (see [12, 9, 27, 29, 25, 51] for details). In =-=[13]-=- we give the first linear-time algorithm and fast implementation for computing inversion distance between two signed permutations. We ran GRAPPA on a 512-processor IBM Linux cluster with Myrinet and o... |

108 | Among-site rate variation and its impact on phylogenetic analyses
- Yang, Z
- 1996
(Show Context)
Citation Context ...e variation (also called rate heterogeneity) among sites in nucleotide substitution methods, since sites (alignment columns) usually do not evolve at the same speed. It has been demonstrated, e.g. in =-=[71]-=-, that ML inference under the assumption of rate homogeneity can lead to erroneous results if rates vary among sites. Rate heterogeneity among sites can easily be accommodated by incorporating an addi... |

103 |
Genetic algorithm approaches for the phylogenetic analysis of large biological sequence data sets under the maximum likelihood criterion
- Zwickl
- 2006
(Show Context)
Citation Context ...ihood values (factor 1.000014 for 50 rate categories, and factor 1.000037 for 25 rate categories). Similar experimental results have been obtained by Derrick Zwickl on different datasets. Citing from =-=[76]-=-, p. 62: “In practice, performing inferences using the GTR+CAT model in RAxML has proven to be an excellent method for obtaining topologies that score well under the GTR+Γ model.” The large speedup of... |

90 |
A new method for calculating evolutionary substitution rates
- Lanave, Preparata, et al.
- 1984
(Show Context)
Citation Context ...o be identic if followed forward or backward in time. Essentially, this means that the maximum number of possible transition types in the General Time Reversible model of nucleotide substitution (GTR =-=[65]-=- [66]) is reduced to 6 due to required symmetries. The less general time-reversible models of nucleotide substitution such as the Jukes-Cantor (JC69 [67]) or HasegawaKishino-Yano (HKY85 [68]) model ca... |

89 |
Phylogeny reconstruction
- Swofford, Olsen
- 1990
(Show Context)
Citation Context ...set. By default RAxML performs a standard hillclimbing search for ML trees but begins with an estimate of the MP tree (constructed using heuristics implemented in Phylip [52]). We use the HKY85 model =-=[53]-=- throughout the study whenever we run RAxML (even as the base and global methods for Rec-I-DCM3). For more details on RAxML we refer the reader to Section 3 of this chapter where RAxML is thoroughly d... |

85 |
Atlas of protein sequence and structure
- Dayhoff
- 1978
(Show Context)
Citation Context ...this value the global upper bound, f(xI), the incumbent’s objective function. In our implementation, the first incumbent is the best returned by any of several heuristic methods. The greedy algorithm =-=[18]-=-, an alternative incumbent heuristic, proceeds as follows. Begin with a three-taxa core tree and iteratively add one taxon at a time. For an iteration with an k-leaf tree, try each of the n − k remain... |

78 | Disk-covering, a fast-converging method for phylogenetic tree reconstruction - Huson, Nettles, et al. - 1999 |

77 | Multiple genome rearrangement and breakpoint phylogeny
- Sankoff, Blanchette
- 1998
(Show Context)
Citation Context ...ce software suite, GRAPPA (Genome Rearrangement Analysis through Parsimony and other Phylogenetic Algorithms) that we developed, GRAPPA extends Sankoff and Blanchette’s breakpoint phylogeny algorithm =-=[39]-=- into the more biologically-meaningful inversion phylogeny and provides a highly-optimized code that can make use of distributed- and shared-memory parallel systems (see [12, 9, 27, 29, 25, 51] for de... |

65 |
The Parsimony Ratchet, a new method for rapid parsimony analysis
- Nixon
- 1999
(Show Context)
Citation Context ...to reconstruct sequence-based phylogeny. The most popular phylogeny software suites that contain parsimony methods are PAUP* by Swofford [20], PHYLIP by Felsenstein [21], and TNT and NONA by Goloboff =-=[22, 23]-=-. We have developed a freely-available sharedmemory code for computing MP, that is part of our software suite, GRAPPA (Genome Rearrangement Analysis through Parsimony and other Phylogenetic Algorithms... |

64 | The median problems for breakpoints are NPcomplete
- Pe’er, Shamir
- 1998
(Show Context)
Citation Context ...derstanding evolution. Problems related to phylogeny reconstruction are widely studied. Most have been proven or are believed to be NP-hard problems that can take years to solve on realistic datasets =-=[5, 6]-=-. Many biologists throughout the world compute phylogenies involving weeks or years of computation without necessarily finding global optima. Certainly more such computational analyses will be needed ... |

58 | von Haeseler A, Strimmer K: TREEFINDER: a powerful graphical analysis environment for molecular phylogenetics - Jobb |

55 | A new implementation and detailed study of breakpoint analysis
- Moret, Wyman, et al.
- 2001
(Show Context)
Citation Context ...point phylogeny algorithm [39] into the more biologically-meaningful inversion phylogeny and provides a highly-optimized code that can make use of distributed- and shared-memory parallel systems (see =-=[12, 9, 27, 29, 25, 51]-=- for details). In [13] we give the first linear-time algorithm and fast implementation for computing inversion distance between two signed permutations. We ran GRAPPA on a 512-processor IBM Linux clus... |

51 | An empirical comparison of phylogenetic methods on chloroplast gene order data in Campanulaceae
- Cosner, Jansen, et al.
- 2000
(Show Context)
Citation Context ...e is that it does not constrain the character states of the input and can use real molecular data and also characters reduced from gene-order data such as Maximum Parsimony on Binary Encodings (MPBE) =-=[25]-=-. The University of New Mexico operates Los Lobos, the NSF / Alliance 512processor Linux supercluster. This platform is a cluster of 256 IBM Netfinity 4500R nodes, each with dual 733 MHz Intel Xeon Pe... |

47 | Formulations and hardness of multiple sorting by reversals
- Caprara
- 1999
(Show Context)
Citation Context ...derstanding evolution. Problems related to phylogeny reconstruction are widely studied. Most have been proven or are believed to be NP-hard problems that can take years to solve on realistic datasets =-=[5, 6]-=-. Many biologists throughout the world compute phylogenies involving weeks or years of computation without necessarily finding global optima. Certainly more such computational analyses will be needed ... |

46 | Maximum likelihood of evolutionary trees is hard
- Chor, Tuller
- 2005
(Show Context)
Citation Context ... within less than 24 hours. Therefore, large-scale phylogenetic analyses with ML are becoming more common recently [55]. The computation of the ML tree has recently been demonstrated to be NPcomplete =-=[56]-=-. The problem of finding the optimal ML tree is particularly difficult due to the immense amount of alternative tree topologies which have to be evaluated and the high computational cost—in terms of f... |

42 |
Analyzing large data sets in reasonable times: Solutions for composite optima
- Goloboff
- 1999
(Show Context)
Citation Context ...to reconstruct sequence-based phylogeny. The most popular phylogeny software suites that contain parsimony methods are PAUP* by Swofford [20], PHYLIP by Felsenstein [21], and TNT and NONA by Goloboff =-=[22, 23]-=-. We have developed a freely-available sharedmemory code for computing MP, that is part of our software suite, GRAPPA (Genome Rearrangement Analysis through Parsimony and other Phylogenetic Algorithms... |

40 |
Branch and Bound Algorithms to Determine Minimal Evolutionary Trees
- PENNY
- 1982
(Show Context)
Citation Context ...e a much smaller set of most promising topologies and choose the best one examined. One advantage of B&B is that it provides instance-specific lower bounds, showing how close a solution is to optimal =-=[15]-=-. The phylogeny reconstruction problem with maximum parsimony (MP) is defined as follows. The input is a set of c characters and a set of taxa represented as length-c sequences of values (one for each... |

33 | Rec-I-DCM3: a fast algorithmic technique for reconstructing large phylogenetic trees
- Roshan, Moret
- 2004
(Show Context)
Citation Context .... This technique will also be available under the CIPRES project software distribution. Usman’s publications can be found at his web page at http://www.cs.njit.edu/usman. Representative Publications: =-=[37, 35, 36, 30, 31, 28, 32, 17, 37]-=-. Alexandros Stamatakis received his Ph.D. in Computer Science from the Technical University of Munich (Germany) in October 2004. He currently works as postdoctoral researcher at the Institute of Comp... |

33 |
ARB: a software environment for sequence data. Nucleic Acids Res
- Ludwig, Strunk, et al.
- 2004
(Show Context)
Citation Context ...ignment along with its sequence length and source. 1. 101 RNA, 1858 bp [44], obtained from Alexandros Stamatakis 2. 150 RNA, 1269 bp [44], obtained from Alexandros Stamatakis 3. 150 ssu rRNA, 3188 bp =-=[45]-=-, obtained from Alexandros Stamatakis 4. 193 ssu rRNA [46], obtained from Alexandros Stamatakis 5. 200 ssu rRNA, 3270 bp [45], obtained from Alexandros Stamatakis 6. 218 ssu rRNA, 4182 bp [47], obtain... |

32 |
Haeseler, IQPNNI: moving fast through tree space and stopping in time
- Vinh, von
(Show Context)
Citation Context ...RNA, 1858 bp [44], obtained from Alexandros Stamatakis 2. 150 RNA, 1269 bp [44], obtained from Alexandros Stamatakis 3. 150 ssu rRNA, 3188 bp [45], obtained from Alexandros Stamatakis 4. 193 ssu rRNA =-=[46]-=-, obtained from Alexandros Stamatakis 5. 200 ssu rRNA, 3270 bp [45], obtained from Alexandros Stamatakis 6. 218 ssu rRNA, 4182 bp [47], obtained from Alexandros Stamatakis 7. 250 ssu rRNA [45], obtain... |

30 | A fast, parallel spanning tree algorithm for symmetric multiprocessors (smps
- Bader, Cong
(Show Context)
Citation Context ... Mexico’s 512-processor NSF Linux SMP supercluster, and achieved a one-million-fold speedup over prior methods—a speed-up that he has since increased to over one billion. Representative Publications: =-=[1,7,8,11,12,14,13,9,15,16,6,2,3,4,5,18,20,19,23,24,26,27,29,25, 33, 34, 40, 48, 49, 50]-=-. Bader has co-presented several tutorials at prior SC meetings: at SC1999 (high-performance computing in the Alliance), at SC2000 (an updated tutorial with the same focus and a second tutorial on par... |

29 | Industrial Applications of High-Performance Computing for Phylogeny Reconstruction
- Bader, Moret, et al.
- 2001
(Show Context)
Citation Context ...medicine (as well as linguistics). Naturally, scientists are interested in phylogenies for the sake of knowledge, but such analyses also have many uses in applied research and in the commercial arena =-=[12]-=-. Existing phylogenetic reconstruction techniques suffer from serious problems of running time (or, when fast, of accuracy). The problem is particularly serious for large data sets: even though data s... |

29 | Parallel implementation and performance of fastDNAml - a program for maximum likelihood phylogenetic inference
- Stewart, Hart, et al.
- 2001
(Show Context)
Citation Context ...achine alignments, i.e., those created solely by a computer program with no human intervention. Below we list the size of each alignment along with its sequence length and source. 1. 101 RNA, 1858 bp =-=[44]-=-, obtained from Alexandros Stamatakis 2. 150 RNA, 1269 bp [44], obtained from Alexandros Stamatakis 3. 150 ssu rRNA, 3188 bp [45], obtained from Alexandros Stamatakis 4. 193 ssu rRNA [46], obtained fr... |

28 |
Stochastic search strategy for estimation of maximum likelihood phylogenetic trees
- Salter, Pearl
- 2001
(Show Context)
Citation Context ...rison to RAxML, IQPNNI is both slower and less accurate [80].sSimulated Annealing Approaches: The first application of simulated annealing techniques to ML tree searches was proposed by Salter et al. =-=[81]-=- (the technique has previously been applied to MP phylogenetic tree searches by D. Barker [82]). However, the respective program SSA has not become very popular due to the limited availability of nucl... |

27 | Initial experiences porting a bioinformatics application to a graphics processor
- Charalambous, Trancoso, et al.
- 2005
(Show Context)
Citation Context ...ouble require more memory space. Thus, the choice of double precision appears to represent a reasonable trade-off. A porting of RAxML from double to float for the purposes of the GPGPU implementation =-=[97]-=- (see Section 3.4) did not yield better results in terms of execution times. Number of Likelihood Scalings 2.5e+06 2e+06 1.5e+06 1e+06 500000 "likelihood_scalings" 0 0 500 1000 1500 2000 2500 Number o... |

26 |
Toward defining the course of evolution: Minimal change for a specific tree topology
- Fitch
- 1971
(Show Context)
Citation Context ...rictions on the changes between adjacent nodes) are often classified into Fitch, Wagner, Dollo, and Generalized (Sankoff) Parsimony [7]. In this example, we use the simplest criteria, Fitch parsimony =-=[16]-=-, which imposes no constraints on permissible character state changes. The optimization techniques we discuss are similar across all of these types of parsimony. Given a topology with leaf labels, we ... |

26 |
Obesity alters gut microbial ecology
- Ley
(Show Context)
Citation Context ... and the increasing use of parallel computers allow for inference of huge phylogenies within less than 24 hours. Therefore, large-scale phylogenetic analyses with ML are becoming more common recently =-=[55]-=-. The computation of the ML tree has recently been demonstrated to be NPcomplete [56]. The problem of finding the optimal ML tree is particularly difficult due to the immense amount of alternative tre... |

26 |
The metapopulation genetic algorithm: An efficient solution for the problem of large phylogeny estimation
- Lemmon, Milinkovich
(Show Context)
Citation Context ... floating point operations—of each tree evaluation per se. To date, the main focus of researchers has been on improving the search algorithms (RAxML [57], PHYML [58], GAML [59], IQPNNI [46], MetaPIGA =-=[60]-=-, Treefinder [61]) and on accelerating the likelihood function via algorithmic means by detecting and re-using previously computed values [62, 63]. Due to the algorithmic progress which has been achie... |