---

Home

---

Organization

---

Citations & Statistics

---

Partners

---

Online programs

---

PhyML

---

Benchmarks

---

Datasets

---

Downloads

---

FAQ

---

News

---

Online execution

---

Papers & contacts

---

PhyML versions

---

User guide

---

Binaries

---

Databases

---

Datasets

PhyML 3.0: new algorithms, methods and utilities

---

Overview

PhyML is a phylogeny software based on the maximum-likelihood principle. Early PhyML versions used a fast algorithm to perform Nearest Neighbor Interchanges (NNIs), in order to improve a reasonable starting tree topology. Since the original publication (Guindon and Gascuel 2003), PhyML has been widely used (>1,250 citations in ISI Web of Science), due to its simplicity and a fair accuracy/speed compromise. In the mean time research around PhyML has continued. We designed an efficient algorithm to search the tree space using Subtree Pruning and Regrafting (SPR) topological moves (Hordijk and Gascuel 2005), and proposed a fast branch test based on an approximate likelihood ratio test (Anisimova and Gascuel 2006). However, these novelties were not included in the official version of PhyML, and we found that improvements were still needed in order to make them effective in some practical cases. PhyML 3.0 achieves this task. It implements new algorithms to search the space of tree topologies with user-defined intensity. A non-parametric, Shimodaira-Hasegawa-like branch test is also available. The program provides a number of new evolutionary models and its interface was entirely re-designed. We tested PhyML 3.0 on a large collection of real data sets to ensure that the new version is stable, ready-to-use and still reasonably fast and accurate.

---

User guide

More explanations are given in the PhyML manual .

Online interface

PHYLIP-like interface

Command-line interface

The online interface

It is a web form that you fill with your values for the program options, your name and your email; then click on the "Execute & email results" button to start the program with your settings; after a while you will receive your results at the email address that you provided. Note that you can also ask to be notified by email of the program developments by subscribing to the mailing list.

Screenshot


For a quick start, simply select the "Example" radio button, type your name, country where you are and email, and click the "Execute & email results" button. This will run the program with default values; the example sequence file contains 5 DNA data sets in PHYLIP sequential format with 60 taxa and 500 bp sequences.

---

Results

PhyML produces several results files:

• <sequence file name>_phyml_lk.txt : site likelihood value(s)
• <sequence file name>_phyml_tree.txt : inferred tree(s)
• <sequence file name>_phyml_stat.txt : detailed execution statistics
• <sequence file name>_phyml_boot_trees.txt : bootstrap trees (special case)
• <sequence file name>_phyml_boot_stats.txt : bootstrap statistics (special case)

Here are the possible uses of PhyML:

• One data set, one starting tree
  Standard analysis under a given substitution model, PhyML then returns the inferred tree. Moreover, a special option allows to perform non-parametric bootstrap analysis on the original data set. PhyML then returns the bootstrap tree with branch lengths and bootstrap values, using standard NEWICK format .
• Several data sets, one starting tree
  Several standard analysis start from the same intial tree with different data sets, without the bootstrap option. The results are given in the order of the data sets. This can be used to process multiple genes in a supertree approach.
• One data set, several starting trees
  Several standard analysis of the same data set using different starting tree situations, without the bootstrap option. All results are given in the order of the trees. Moreover, the most likely tree is provided in the *_best_stat.txt and *_best_tree.txt files. This can be used to test alternative topologies, or to avoid being trapped into local optima. With the 'SPR' and 'SPR&NNI' options, it's also possible to add random starting trees to standard BioNJ starting tree.
• Several data sets, several starting trees
  Several standard runs, where each data set is analysed with the corresponding starting tree, without the bootstrap option. The results are given in the order of the data sets. This can be used when comparing the likelihood of various trees regarding different data sets.

---

Options

• Sequences
  The input sequence file is a standard PHYLIP file of aligned DNA or amino-acids sequences. It should look like this in interleaved format :

  5 60
  Tax1        CCATCTCACGGTCGGTACGATACACCTGCTTTTGGCAG
  Tax2        CCATCTCACGGTCAGTAAGATACACCTGCTTTTGGCGG
  Tax3        CCATCTCCCGCTCAGTAAGATACCCCTGCTGTTGGCGG
  Tax4        TCATCTCATGGTCAATAAGATACTCCTGCTTTTGGCGG
  Tax5        CCATCTCACGGTCGGTAAGATACACCTGCTTTTGGCGG
  
  GAAATGGTCAATATTACAAGGT
  GAAATGGTCAACATTAAAAGAT
  GAAATCGTCAATATTAAAAGGT
  GAAATGGTCAATCTTAAAAGGT
  GAAATGGTCAATATTAAAAGGT
  

  The same data set in sequential format:

  5 60
  Tax1        CCATCTCACGGTCGGTACGATACACCTGCTTTTGGCAGGAAATGGTCAATATTACAAGGT
  Tax2        CCATCTCACGGTCAGTAAGATACACCTGCTTTTGGCGGGAAATGGTCAACATTAAAAGAT
  Tax3        CCATCTCCCGCTCAGTAAGATACCCCTGCTGTTGGCGGGAAATCGTCAATATTAAAAGGT
  Tax4        TCATCTCATGGTCAATAAGATACTCCTGCTTTTGGCGGGAAATGGTCAATCTTAAAAGGT
  Tax5        CCATCTCACGGTCGGTAAGATACACCTGCTTTTGGCGGGAAATGGTCAATATTAAAAGGT
  

  On the first line is the number of taxa, a space, then the number of sites in the alignment.
  
  The maximum number of characters in species name MUST not exceed 100. Blanks and the symbols “(),:” are not allowed within sequence names because the Newick tree format makes special use of these symbols. However, blanks (one or more) MUST appear at the end of each species name.
  
  In a sequence, three special characters '.', '-', and '?' may be used: a dot '.' means the same character as in the first sequence, a dash '-' means an alignment gap and a question mark '?' means an undetermined nucleotide. Sites at which one or more sequences involve '-' are NOT excluded from the analysis. Therefore, gaps are treated as unknown character (like '?') on the grounds that ''we don't know what would be there if something were there'' (J. Felsenstein, PHYLIP documentation). Finally, standard ambiguity characters for nucleotides are accepted (Table 1).
  
  

  Table 1 - Nucleotide character coding Character Nucleotide A Adenosine G Guanine C Cytosine T Thymine U Uracil M A or C R A or G W A or T S C or G Y C or T K G or T B C or G or T D A or G or T H A or C or T V A or C or G N or X or ? unknown

  Table 2 - Amino-Acid character coding Character Amino-Acid A Alanine R Arginine N or B Asparagine D Aspartic acid C Cysteine Q or Z Glutamine E Glutamic acid G Glycine H Histidine I Isoleucine L Leucine K Lysine M Methionine F Phenylalanine P Proline S Serine T Threonine W Tryptophan Y Tyrosine V Valine X or ? unknown

• Data type
  This indicates if the sequence file contains DNA or amino-acids. The default choice is to analyse DNA sequences.
• Sequence format
  The input sequences can be either in interleaved (default) or sequential format, see "Sequences" above.
• Number of data sets
  Multiple data sets are allowed, e.g. to perform bootstrap analysis using SEQBOOT (from the PHYLIP package). In this case, the data sets are given one after the other, in the formats above explained. For example (with three data sets):

  5 60
  Tax1        CCATCTCACGGTCGGTACGATACACCTGCTTTTGGCAGGAAATGGTCAATATTACAAGGT
  Tax2        CCATCTCACGGTCAGTAAGATACACCTGCTTTTGGCGGGAAATGGTCAACATTAAAAGAT
  Tax3        CCATCTCCCGCTCAGTAAGATACCCCTGCTGTTGGCGGGAAATCGTCAATATTAAAAGGT
  Tax4        TCATCTCATGGTCAATAAGATACTCCTGCTTTTGGCGGGAAATGGTCAATCTTAAAAGGT
  Tax5        CCATCTCACGGTCGGTAAGATACACCTGCTTTTGGCGGGAAATGGTCAATATTAAAAGGT
  
  5 60
  Tax1        CCATCTCACGGTCGGTACGATACACCTGCTTTTGGCAGGAAATGGTCAATATTACAAGGT
  Tax2        CCATCTCACGGTCAGTAAGATACACCTGCTTTTGGCGGGAAATGGTCAACATTAAAAGAT
  Tax3        CCATCTCCCGCTCAGTAAGATACCCCTGCTGTTGGCGGGAAATCGTCAATATTAAAAGGT
  Tax4        TCATCTCATGGTCAATAAGATACTCCTGCTTTTGGCGGGAAATGGTCAATCTTAAAAGGT
  Tax5        CCATCTCACGGTCGGTAAGATACACCTGCTTTTGGCGGGAAATGGTCAATATTAAAAGGT
  
  5 60
  Tax1        CCATCTCACGGTCGGTACGATACACCTGCTTTTGGCAGGAAATGGTCAATATTACAAGGT
  Tax2        CCATCTCACGGTCAGTAAGATACACCTGCTTTTGGCGGGAAATGGTCAACATTAAAAGAT
  Tax3        CCATCTCCCGCTCAGTAAGATACCCCTGCTGTTGGCGGGAAATCGTCAATATTAAAAGGT
  Tax4        TCATCTCATGGTCAATAAGATACTCCTGCTTTTGGCGGGAAATGGTCAATCTTAAAAGGT
  Tax5        CCATCTCACGGTCGGTAAGATACACCTGCTTTTGGCGGGAAATGGTCAATATTAAAAGGT
  

• Substitution model
  A nucleotide or amino-acid substitution model.
  
  For DNA sequences, the default choice is HKY85 (Hasegawa et al., 1985). This model is analogous to K80 (Kimura, 1980), but allows for different base frequencies. The other models are JC69 (Jukes and Cantor, 1969), K80 (Kimura, 1980), F81 (Felsenstein, 1981), F84 (Felsenstein, 1989), TN93 (Tamura and Nei, 1993) and GTR (e.g., Lanave et al. 1984, Tavaré 1986, Rodriguez et al. 1990). The rate matrices of these models are given in Swofford et al. (1996).
  
  For amino-acid sequences, the default choice is LG (Le and Gascuel 2008) which tends to improve other models with standard globular proteins. The other models are WAG (Whelan and Goldman, 2001), (Dayhoff et al., 1978), mtREV (as implemented in Yang's PAML), JTT (Jones, Taylor and Thornton, 1992), DCMut (Kosiol and Goldman, 2005), RtREV (Dimmic et al.), CpREV (Adachi et al., 2000), VT (Muller and Vingron, 2000), Blosum62 (Henikoff anf Henikoff, 1992) and MtMam (Cao, 1998).
• Equilibrium frequencies
  For nucleotide sequences, optimising nucleotide frequencies means that the values of these parameters are estimated in the maximum likelihood framework. For protein sequences, the stationary amino-acid frequencies are either those defined by the substitution model or those estimated by counting the number of different amino-acids observed in the data.
• Transition / transversion ratio
  With DNA sequences, it is possible to set the transition/transversion ratio, except for the JC69 and F81 models, or to estimate its value by maximising the likelihood of the phylogeny. The later makes the program slower. The default value is 4.0. The definition of the transition/transversion ratio is the same as in PAML (Yang, 1994). In PHYLIP, the ''transition/transversion rate ratio'' is used instead. 4.0 in PhyML roughly corresponds to 2.0 in PHYLIP.
• Proportion of invariable sites
  The default is to consider that the data set does not contain invariable sites (0.0). However, this proportion can be set to any value in the 0.0-1.0 range, but it should not be larger than the proportion of constant sites in the alignments. This parameter can also be estimated by maximising the likelihood of the phylogeny. The later makes the program slower.
• Number of substitution rate categories
  The default is having all the sites evolving at the same rate, hence having one substitution rate category. A discrete-gamma distribution can be used to account for variable substitution rates among sites, in which case the number of categories that defines this distribution is supplied by the user. The higher this number, the better is the goodness-of-fit regarding the continuous distribution. The default is to use four categories, in this case the likelihood of the phylogeny at one site is averaged over four conditional likelihoods corresponding to four rates and the computation of the likelihood is four times slower than with a unique rate. Number of categories less than four or higher than eight are not recommended. In the first case, the discrete distribution is a poor approximation of the continuous one. In the second case, the computational burden becomes high and an higher number of categories is not likely to enhance the accuracy of phylogeny estimation.
• Gamma shape parameter
  The shape of a gamma distribution is defined by this numerical parameter. The higher its value, the lower the variation of substitution rates among sites (this option is used when having more than 1 substitution rate category). The default value is 1.0. It corresponds to a moderate variation. Values less than say 0.7 correspond to high variations. Values between 0.7 and 1.5 corresponds to moderate variations. Higher values correspond to low variations. This value can be fixed by the user. It can also be estimated by maximising the likelihood of the phylogeny.
• Starting tree(s)
  Used as the starting tree(s) to be refined by the maximum likelihood algorithm. The default is to use a BIONJ distance-based tree. You can use additional random starting trees with the SPR and SPR & NNI options. It is also possible to supply one or several trees in NEWICK format, one per line in the file, which must be written in the standard parenthesis representation (NEWICK format). Trees can be either rooted or unrooted and multifurcations are allowed. Taxa names must, of course, match the corresponding sequence names. Labels on branches (such as bootstrap proportions) are supported. Therefore, a tree with four taxa named A, B, C, and D with a bootstrap value equal to 90 on its internal branch, should look like this:
  (A:0.02,B:0.004,(C:0.1,D:0.04)90:0.05);
  If you give several trees and analyse several data sets the two numbers must match.
• Type of tree improvement
  PhyML is able to perform two kinds of tree topology improvement: NNI (Nearest Neighbor Interchange) and SPR (Subtree Pruning and Regrafting). You can ask PhyML to compute only NNI (faster) or only SPR (a bit slower) or the both SPR & NNI and keep the best.
• Optimise starting tree(s) options
  You can optimise the starting tree(s) in three ways:
  • You can optimise the topology, the branch lengths and rate parameters (transition/transversion ratio, proportion of invariant sites, gamma distribution parameter).
  • You can keep the topology and optimise the branch lengths and model parameters (it is not possible to optimise the tree topology and keep the branch lengths).
  • You can ask for no optimisation, PhyML just returns the likelihood of the starting tree(s).
• aLRT (approximate Likelihood-Ratio Test) for branches
  aLRT is a statistical test to compute branch supports. It applies to every (internal) branch and is computed along PhyML run on the original data set. Thus, aLRT is much faster than standard bootstrap which requires running PhyML 100-1,000 times with resampled data sets. As with any test, the aLRT branch support is significant when it is larger than 0.90-0.99. With good quality data (enough signal and sites), the sets of branches with bootstrap proportion >0.75 and aLRT>0.9 (SH-like option) tend to be similar.
• Perform bootstrap and number of resampled data sets
  When there is only one data set you can ask PhyML to generate resampled bootstrap data sets from this original data set. PhyML then returns the bootstrap tree with branch lengths and bootstrap values, using standard NEWICK format.

---

References

• "A simple, fast and accurate algorithm to estimate large phylogenies by maximum likelihood."
  Guindon S. & Gascuel O.
  Systematic Biology 52, 696–704 (2003).
• "RAxML-VI-HPC: Maximum likelihood-based phylogenetic analyses with thousands of taxa and mixed models."
  Stamatakis A.
  Bioinformatics 22, 2688–2690 (2006).
• "Genetic algorithm approaches for the phylogenetic analysis of large bio-logical sequence datasets under the maximum likelihood criterion."
  Zwickl D.
  Ph.D. thesis, The University of Texas at Austin (2006).
• "Evolution of protein molecules."
  Jukes T. & Cantor C.
  In Munro, H. (ed.) Mammalian Protein Metabolism, vol. III, chap. 24, 21–132 (Academic Press, New York, 1969).
• "A simple method for estimating evolutionary rates of base substitutions through comparative studies of nucleotide sequences."
  Kimura M.
  Journal of Molecular Evolution 16, 111–120 (1980).
• "Evolutionary trees from DNA sequences: a maximum likelihood approach."
  Felsenstein J.
  Journal of Molecular Evolution 17, 368–376 (1981).
• "PHYLIP (PHYLogeny Inference Package) version 3.6a2."
  Felsenstein J.
  (Distributed by the author, Department of Genetics, University of Washington, Seattle, 1993).
• "Dating of the Human-Ape splitting by a molecular clock of mitochondrial-DNA."
  Hasegawa M., Kishino H. & Yano, T.
  Journal of Molecular Evolution 22, 160–174 (1985).
• "Estimation of the number of nucleotide substitutions in the control region of mitochondrial DNA in humans and chimpanzees."
  Tamura K. & Nei M.
  Molecular Biology and Evolution 10, 512–526 (1993).
• "A new method for calculating evolutionary substitution rates."
  Lanave C., Preparata G., Saccone C. & Serio G.
  Journal of Molecular Evolution 20, 86–93 (1984).
• "Some probabilistic and statistical problems on the analysis of DNA sequences."
  Tavaré S.
  Lectures on Mathematics in the Life Sciences 17, 57–86 (1986).
• "An improved general amino-acid replacement matrix."
  Le S. & Gascuel O.
  Mol. Biol. Evol. 25(7):1307-20 (2008).
• "A general empirical model of protein evolution derived from multiple protein families using a maximum-likelihood approach."
  Whelan S. & Goldman N.
  Mol. Biol. Evol. 18, 691–699 (2001).
• "A model of evolutionary change in proteins."
  Dayhoff M., Schwartz R. & Orcutt B.
  In Dayhoff, M. (ed.) Atlas of Protein Sequence and Structure, vol. 5, 345–352 (National Biomedical Research Foundation, Washington, D. C., 1978).
• "The rapid generation of mutation data matrices from protein sequences."
  Jones D., Taylor W. & Thornton J.
  Computer Applications in the Biosciences (CABIOS) 8, 275–282 (1992).
• "Amino acid substitution matrices from protein blocks."
  Henikoff S. & Henikoff J.
  PNAS 89, 10915–10919 (1992).
• "MOLPHY version 2.3. programs for molecular phylogenetics based on maximum likelihood."
  Adachi J. & Hasegawa M.
  In Ishiguro, M. et al. (eds.) Computer Science Monographs, 28 (The Institute of Statistical Mathematics, Tokyo, 1996).
• "rtREV: an amino acid substitution matrix for inference of retrovirus and reverse transcriptase phylogeny."
  Dimmic M., Rest J., Mindell D. & Goldstein D.
  Journal of Molecular Evolution 55, 65–73 (2002).
• "Plastid genome phylogeny and a model of amino acid substitution for proteins encoded by chloroplast DNA."
  Adachi J.P.W., Martin W. & Hasegawa M.
  Journal of Molecular Evolution 50, 348–358 (2000).
• "Different versions of the Dayhoff rate matrix."
  Kosiol C. & Goldman N.
  Molecular Biology and Evolution 22, 193–19 (2004).
• "Modeling amino acid replacement."
  Muller T. & Vingron M.
  Journal of Computational Biology 7, 761–776 (2000).
• "Conflict among individual mitochondrial proteins in resolving the phylogeny of eutherian orders."
  Cao Y. et al.
  Journal of Molecular Evolution 47, 307–322 (1998).
• "BIONJ: an improved version of the NJ algorithm based on a simple model of sequence data."
  Gascuel O.
  Molecular Biology and Evolution 14, 685–695 (1997).
• "Modeltest: testing the model of DNA substitution."
  Posada D. & Crandall K.
  Bioinformatics 14, 817–918 (1998).
• "Prottest: selection of best-fit models of protein evolution."
  Abascal F., Zardoya R. & Posada D.
  Bioinformatics 21, 2104–2105 (2005).
• "Markov-modulated Markov chains and the covarion process of molecular evolution."
  Galtier N. & Jean-Marie A.
  Journal of Computational Biology 11, 727–733 (2004).

Webmaster: Vincent Lefort, LIRMM.