ProtTest - Bioinformatics and Computational Biology Services
Contents
1. OB sodel RETTET TEEEEEEETETEETEEETTETETT JTT F Number of parametersS s 58 19 39 branch length estimates S aminoacid frequencies observed see above EOC 2999 42 0h0m5s ar S PE OE EA JTHI Number of parameters proportion of invariable sites 40 1 39 branch length estimates 0 069 lNL ssssssssooosossooososssoososso 2976 88 0h0m16s Model ccccccccccnccccnssecccccscce JTT I F gt Number of parameters 59 20 39 branch length estimates proportion of invariabl _ aminoacid frequencies 0 068 observed see above 2975 28 OhOmizs Work progress model 5 32 gt Sort Dy Overall comparison Display best tree in Newick format _ ASCII text Close Save Figure 2 ProtTest s console window ProtTest at the command line To run Prottest at the command line open a shell window terminal change to the directory where ProtTest is installed and run the script runProtTest In Windows use the script runProtTest bat instead specifying the following options A alignment file required E tree file optional default NJ tree O output file optional default STDOUT sort A B C D optional default A A AIC B BIC C ATCC D LLnL all T F If true a 7 framework comparison table is displayed default true S optimization strategy mode default 0 0 Fast optimize branch lengths am2. model averaged estimates of different parameters Posada and Buckley 2004 and calculates the importance of each of these parameters ProtTest differs from its nucleotide homolog Modeltest Posada and Crandall 1998 in that it does not include likelihood ratio tests many models implemented in ProtTest are not nested The program using ProtTest ProtTest selection of models of protein evolution ProtTest is written in java and takes advantage of the PAL library Drummond and Strimmer 2001 for manipulating trees and alignments and of the Phyml program Guindon and Gascuel 2003 for the computation of likelihoods and the estimation of parameters Given an alignment and a tree provided by the user or calculated with the BIONJ algorithm Gascuel 1997 ProtTest currently computes the likelihood for each one of 112 candidate models of protein evolution matrices WAG Dayhoff JTT mtREV MtMam MtArt VT RtREV CoREV Blosum62 LG DCMut HIVw and HIVb with the l G and F parameters Then the fit of the models can be estimated using the AIC AlCc and BIC Download and installation ProtTest works in Mac OSX Windows and Linux and requires a version of the java runtime environment equal or posterior to 1 3 read section Installing java if you don t have it ProTest is available from http darwin uvigo es After registration download the package and decompress it in any directory Some examples are included Installing java
3. there are some ways we can model these constraints we can consider that a fraction of the amino acids are invariable commonly indicated with a I code in the name of the model Reeves 1992 we can consider some different categories of change low medium high rate etc and assing each site a probability to belong to each of these categories usually indicated by a G code Yang 1993 or we can include both in the model I G Also we can use as equilibrium aminoacid frequencies those observed in the alignment at hand indicated as F Cao et al 1994 Statistics for model selection Akaike Information Criterion and others For a more detailed background on model selection the user is referred to Posada and Crandall 2001 and Posada and Buckley in press Burnham and Anderson 2003 provide a very good description of the AIC framework and its use for model averaging which they call multimodel inference The fit of a model of protein evolution M to a given data set D given a tree T and branch lenghts B is measured by the likelihood function L L P D M 7 B One could think that the best model is the one which results in the maximum likelihood but this is not necessarily true the more parameters the model includes the higher its advantage in fitting better the data but also the higher the variance for the parameter estimates So how many parameters should the best model include One way to answer
4. Phylogenetics and sample size What is the sample size of a protein alignment is very unclear ProtTest offers different criteria for sample size determination e Alignment length default e Number of variable sites e Shannon entropy summed over all alignment positions Shannon Entropy Shannon 1948 It s a way to measure the disorder or entropy ShEn X Pilog Pi In our case if a position is completely conserved it takes the value of 0 If completely disordered frequency of every aminoacid equals 1 20 it takes the value of 4 32 e Number of sequences x length of the alignment x normalized Shannon s entropy The normalized Shannon s entropy is calculated by summing the entropies over all positions dividing this quantity by the number of positions and dividing the resulting quantity by the maximum possible entropy 4 32 13 ProtTest selection of models of protein evolution e Number of sequences x length of the alignment e Users provided size Akaike weights and the relative importance of parameters The AIC or AlCc or BIC differences can be used for calculating the Akaike weights exp 1 2Ai 5 exp 1 2A i these weights can be interpreted as the probability that a model is the best AIC model Parameter _importance By summing the weights of the models that include a given parameter for example the gamma distribution we get the relative importance of such parameter R wpa
5. X w M where 1 if is in model M I M 0 otherwise Model averaged parameter estimates We can also obtain an averaged estimation of any parameter by summing the different estimates for the models that contain such parameter after multiplying them by the Akaike weight of the corresponding model For example the model averaged estimate of alpha q for R candidate models would be R 5 wily M Qa Pa i l s Wy a where R w gt Wily M and 1 if Qa is in model M ly M 0 otherwise 14 ProtTest selection of models of protein evolution Credits and acknowledgements ProtTest takes advantage of the PAL library Drummond and Strimmer 2001 for manipulating alignments and trees The core of the computation is carried out by the Phyml program slightly modified to acomplish some requirements and to include additional models Guindon and Gascuel 2003 which calculates the likelinoods and optimizes the parameters Phyml is also used for calculating BioNJ trees The code of ProtTest takes also benefit from other resources found at the WWW as indicated in the source java code Very special thanks to Stephane Guindon Phyml and Matthew Goode PAL for being so helpful and patient This work was financially supported from a grant for research in bioinformatics from the Fundacion BBVA Given that ProtTest uses intensively Phyml and PAL we encourage users to cite these programs as well when
6. optimization of model branches and topology of the tree see the Optimization strategies section below 4 Optionally you might want to restrict the set of candidate models For that you should click the models button Read first the section Restricting the set of candidate models 5 And click the Start button ProtTest selection of models of protein evolution Protlest 2 0 11 About WWW Help ProtTest Tree PAL amp Phyml based Alignment BION tree J User tree required Select file Program options Optimization strategy Fast optimiz Set of candidate models Exit Start Figure 1 ProtTest main window At this moment ProtTest starts computing the parameters for the different models Be aware that computation can take some time maybe hours or even days You can watch the progress in a new window the console see figure 2 If there s some error related with the format of the alignment or the installation of the program a warning will appear in the console When likelihood computations are finished you ll be warned and prompted to select a statistical framework AIC AlCc or BIC for determining which of the candidate models best fits your data If you select AlCc or BIC you will be prompted to specify a criterion to estimate the size of the sample Additionally you can ask ProtTest to display a comparison of different selection scenarios AIC AlCc a
7. orders J Mol Evol 47 307 322 Dayhoff M O Schwartz R M and Orcutt B C 1978 A model of evolutionary change in proteins In Atlas of Protein Sequence and Structure ed M O Dayhoff pp 345 352 National Biomedical Research Foundation Washington DC Dimmic M W Rest J S Mindell D P and Goldstein R A 2002 rtREV an amino acid substitution matrix for inference of retrovirus and reverse transcriptase phylogeny J Mol Evol 55 65 73 Drummond A and Strimmer K 2001 PAL an object oriented programming library for molecular evolution and phylogenetics Bioinformatics 17 662 663 Gascuel O 1997 BIONJ an improved version of the NJ algorithm based on a simple model of sequence data Mol Biol Evol 14 685 695 Gilbert D 2001 Guindon S and Gascuel O 2003 A simple fast and accurate algorithm to estimate large phylogenies by maximum likelihood Syst Biol 52 696 704 Henikoff S and Henikoff J G 1992 Amino acid substitution matrices from protein blocks Proc Natl Acad Sci U S A 89 10915 10919 Jones D T Taylor W R and Thornton J M 1992 The rapid generation of mutation data matrices from protein sequences Comp Appl Biosci 8 275 282 Le S Q and Gascuel O 2008 An improved general amino acid replacement matrix Mol Biol Evol 25 1307 1320 Maddison W P and Maddison D R 1992 MacClade analysis of phylogeny and character evolution Sinauer Associates Inc Sunderland Massachusetts US
8. s Shannon Entropy over the whole alignment 169 2 AICc BIC 3 sample size as align length x num sequences x averaged Sh Entropy 822 2 10 ProtTest selection of models of protein evolution We can end to some conclusions the empirical WAG matrix is clearly the one that fits best the family of L5 proteins However modifying the matrices with the observed amino acid frequencies applying F allows RtREV models to better fit the data compared to WAG Since the use of these observed frequencies add extra parameters to the model AIC AlCc and BIC interpretate this better fit of RtREV models as an over fit and penalizes them consequently Including a gamma distribution to account for different rates of change at different positions is always of some importance ranging from 0 09 in AlCc 3 to 0 43 in BIC 3 Including an invariable sites distribution alone is not But both G and I together do it better Known bugs Many users have reported errors when running ProtTest under Windows Such errors are related to filenames and file paths conflicts and can be usually circumvented by placing the ProtTest input files the alignment and optionally the tree in a lower directory such as C Program history Version 2 4 September 2009 Bug fixed in the reading of the proportion of invariable sites Version 2 2 August 2009 Some new options added to the command line version of the program E g numcat Version 2 1 June 2009
9. this question is by using the Akaike Information Criterion AIC Akaike 1973 AIC 2LnL 2K 12 ProtTest selection of models of protein evolution LnL log likelinood K number of parameters The model with lowest AIC is expected to be the closest model to the true model among the set of candidate models Since AIC is on a relative scale it is useful to present also the AIC differences or deltaAIC For the ith model the AIC difference is Ai AIC min AIC where min AIC is the smallest AIC among all candidate models The AIC might not be accurate when the size of the sample is small compared to the number of parameters For these cases it is recommended to use a second order AIC or corrected AIC AlCc in ProtTest Sugiura 1978 which includes a penalty for cases where the sample size is small 2K K 1 n K 1 AlCc AIC where n is the size of the sample see below If n is large with respect to K the second term is negligible and AlCc behaves similar to AIC The corrected AIC is recommended when the relation n K is small for example n K lt 40 being K the number of parameters of the most complex model among the set of candidate models ProtTest also calculates the Bayesian Information Criterion BIC Schwarz 1978 which is another measure of model fit The BIC is considered a good approximation of the very computationally demanding Bayesian methods and is formulated as BIC 2LInL K log n
10. 1 September 2009 ProtTest Selection of best fit models of protein evolution version 2 4 Federico Abascal Rafael Zardoya and David Posada 1 Universidad de Vigo Vigo Spain 2 Museo Nacional de Ciencias Naturales Madrid Spain abascal mncn csic es dposada uvigo es rafaz mnecn csic es e What can I use ProtTest for Introduction e The program using ProtTest o Download and installation Installing java o Running ProtTest ProtTest through its graphical Interface GUI ProtTest at the command line ProtTest at the web Optimization strategies Alignment and tree formats Restricting the set of candidate models o ProtTest s output A guided example the ribosomal L5 protein family e Known bugs e Program history e BACKGROUND o Models of protein evolution o Statistics for model selection Akaike Information Criterion and others Phylogenetics and sample size Akaike weights and the relative importance of parameters e Parameter importance e Model averaged parameter estimates e Credits and acknowledgements e References o0 o What can I use ProtTest for Introduction ProtTest is a bioinformatic tool for the selection of the most appropriate model of protein evolution among the set of candidate models for the data at hand ProtTest makes this selection by finding the model with the smallest Akaike Information Criterion AIC or Bayesian Information Criterion BIC score At the same time ProtTest obtains
11. A pp 398 Muller T and Vingron M 2000 Modeling amino acid replacement J Comput Biol 7 761 776 Nickle D C Heath L Jensen M A Gilbert P B Mullins J I and Kosakovsky Pond S L 2007 HIV specific probabilistic models of protein evolution PLoS ONE 2 e503 Posada D and Buckley T R 2004 Model Selection and Model Averaging in Phylogenetics Advantages of AIC and Bayesian approaches over Likelihood Ratio Tests Systematic Biology 53 793 808 Posada D and Crandall K A 1998 MODELTEST testing the model of DNA substitution Bioinformatics 14 817 818 16 ProtTest selection of models of protein evolution Posada D and Crandall K A 2001 Selecting the best fit model of nucleotide substitution Syst Biol 50 580 601 Reeves J H 1992 Heterogeneity in the substitution process of amino acid sites of proteins coded for by mitochondrial DNA J Mol Evol 35 17 31 Schwarz G 1978 Estimating the dimension of a model Ann Statist 6 461 464 Shannon C E 1948 A mathematical theory of communication Bell System Technical Journal 27 379 423 and 623 656 Sugiura N 1978 Further analysis of the data by Akaike s information criterion and the finite correction Comm Statist A Theory Meth 7 13 26 Thorne J L and Goldman N 2003 Probabilistic models for the study of protein evolution In Handbook of Statistical Genetics ed M B D J Balding and C Cannings pp 209 226 John Wiley amp Sons Lt
12. First of all make sure that you have a Java Virtual Machine JVM properly installed in your system To test your JVM 1 Go to http www java com en download help testvm jsp 2 Or in a terminal window type java version The JVM is also included in Java Runtime Environment JRE Java 2 Platform Standard Edition J2SE More information on obtaining the JVM in http java sun com To automatically download the JVM http java sun com webapps getjava BrowserRedirect Running ProtTest There are three different ways to run ProtTest using the graphical interface recommended using the command line version and through the web server http darwin uvigo es ProtTest through its graphical Interface GUI Just double click the jar file ProtTest jar Note Some Linux environments need to be configured to respond to jar double clicks If you are not able to set up Linux to do so you can launch ProtTest by running the runXProtTest script A window like the one in figure 1 will appear in the screen Now follow these steps 1 Input an alignment in phylip recommended or nexus sequential format more about accepted formats in the section Alignment and tree formats below 2 Optionally you can input a tree topology in newick format if not a BIONJ tree will be calculated 3 Select the strategy fast fixed tree optimization of model parameters and branch lengths this is the strategy used by Modeltest or slow
13. NS Sans ened see JTT I Number of parameters 40 1 39 branch length estimates proportion of invariable sites 0 069 LTT Saharan E E AE Talk AEE 2955 16 OhOm8s When likelihoods and model parameters are estimated for all models we will be prompted to select a statistical framework for the selection of the best model To accomplish this we should select one of the options that can be found at the bottom of the console Even if raw likelihoods are not adequate for model selection we start by selecting this option to illustrate some concepts We ll get something like this KKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKK Maximum Likelihood lnL framework KKKEKKK KKK KK KKK KKK KKK KKK KKK KKK KKK KKK KKK KKK KKK KKK KKK KKKKEKK Best model according to lnL WAG I G F KKEKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKK odel deltaAIC AIC inL AICw RtREV I Gt F 0 96 5908 73 2894 36 031 RtREV G F 2625 5910 02 2896 01 0 16 AG I G F 13 48 5921 25 2900 63 0 00 AG G F 16 46 5924 23 2903 12 0 00 AG I G 0 00 5907 77 2912 89 0 49 AG G 5x29 5913 06 2916 53 0 04 Blosum62 I G 9 84 5917 61 2917 81 0 00 CpREV I G F 51 66 5959 43 2 919 71 0 00 rest of lines omitted We can see which model has the highest likelihood the RtREV I G F Some information related with the AIC framework is also displayed but forget it by now Now we select the AIC framework and this is what we obtain kkkkx
14. Updated to a new release of Phyml Version 2 0 March 2009 Major update LG and DCmut models included Updated to Phyml v3 version Version 1 4 July 2007 HIVb and HIVw models added to ProtTest Version 1 3 January 2006 Version tracking renumbered according to the release of a new version Version 1 2 16 November 2005 Minor aesthetic change some information is printed to the console when ProtTest is launched in the console mode Version 1 2 12 July 2005 A bug in the overall comparison has been fixed thanks to Marc Elliot Version 1 2 10 April 2005 New model in ProtTest MtArt is a replacement matrix for arthropod mitochondrial proteins It has been estimated with Paml Version 1 2 8 February 2005 bug corrected so ProtTest is now java 1 3 compatible Version 1 2 6 January 2005 added the ability to specify the number of rate categories for the gamma distribution Version 1 2 4 January 2005 MtMam matrix added to ProtTest and Phynil Version 1 2 2 December 2004 some adjustments to the calculation of model averaged parameters and their importance Version 1 2 0 November 2004 models based on VT CpREV RtREV and Blosum62 added to Phyml and ProtTest Version 1 0 4 October 2004 updated to new version of Phyml 2 4 1 ProtTest now checks for taxa name duplicates this caused problems with Phyml Version 1 0 2 October 2004 A problem with spaces in the path has been corrected Some improvements in the co
15. a good candidate RtREV I G F has the highest likelihood but it has more parameters 19 from the F and two from l G plus the ones for the branch lengths and consequently is penalized by the AIC what situates WAG I G as the best fit model The worst models are the ones built upon the MtREV matrix Below the table of models we can see the relative importance of parameters There we find that including I G is very important the two best models use this distribution G seems to be also important but alone seems to describe poorly the evolution of these proteins In this example adding F has some importance 0 47 because the model RtREV I G F is the second best Below we see a model averaged estimate of parameters In this example the averaged alpha shape of the models I G has a value of 2 69 Why The alpha of the models I G mainly the top ranking WAG I G and RtREV I G F is averaged using the weight of those models Note that both the importance and the averaged estimate of alpha are separated for models G and models I G given the interdependence of and G parameters The same stands for the models I Let s try now the AlCc framework to see if the AIC model selection is affected by a not enough large sample size If we select AlCc at the bottom of the console a new window will appear prompting us to select a criterion to determine the sample size We start selecting for example the Alignment length As a resu
16. d Chichester England Whelan S and Goldman N 2001 A general empirical model of protein evolution derived from multiple protein families using a maximum likelihood approach Mol Biol Evol 18 691 699 Yang Z 1993 Maximum likelihood estimation of phylogeny from DNA sequences when substitution rates differ over sites Mol Biol Evol 10 1396 1401 17
17. ection AIC AlCc BIC or LnL If AlCc or BIC sort C and sort B respectively are selected you might want to change the default criterion for sample size interpretation To accomplish this set the sample option to one of the values shown above 0 5 The optimization strategy see the section Optimization strategies below is specified by the S option followed by a 0 or 1 fast or slow The all option is set to true by default Since the command line version of ProtTest is not interactive and you cannot play with the different frameworks once the likelihoods are calculated the all option is useful for having a table in which you can see at one sight how the best model selection is affected under seven different scenarios AIC AlCc and BIC with three different criteria for sample size corresponding to 0 1 and 2 in the sample option If t1 or t2 options are set to true the tree corresponding to the best model will be displayed in the output in newick format or ASCII representation respectively The order in which you specify the options doesn t matter Example runProtTest i alignment _file phylip S 0 sort C sampl i o results txt t1 T MtArt F MtREV F MtMam F F F By running ProtTest with these options we will find the model that fits best the alignment contained in alignment_file phylip according to the AlCc criterion and taking as sample size the nu
18. kkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkk Akaike Information Chriterion AIC framework kkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkk Best model according to AIC WAG I G kkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkk Model deltaAIC AIC AICw 1nL ProtTest selection of models of protein evolution WAG I G 0 00 5907 77 0 49 2912 89 RtREV I G F 0 96 5908 73 0 31 2894 36 RtREV G F 2 25 5910 02 0 16 2896 01 WAG G 5 29 5913 06 0 04 2916 53 Blosum62 I G 9 84 5917 61 0 00 2917 81 WAG I G F 13 48 5921 25 0 00 2900 63 Blosum62 G 14 07 5921 84 0 00 2920 92 WAG G F 16 46 5924 23 0 00 2903 12 some lines omitted MtREV G 402 75 6310 52 0 00 3115 26 MtREV I 584 89 6492 66 0 00 3206 33 MtREV 616 99 6524 76 0 00 3223 38 kkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkk Relative importance of parameters kkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkk alpha G 0 20 p inv I 0 00 alpha p inv I G 0 80 freqs F 0 47 kkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkk Model averaged estimate of parameters kkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkk alpha G 1 67 p inv I 0 07 alpha I G 2 69 p inv I G 0 05 The best model according to the AIC criterion is the WAG I G model and the probability that it is the best AIC model is 0 49 its Akaike weight The second best model is RtREV I G F and since the AIC difference is 0 96 with a 0 31 weight this model is also
19. le interpretations through a guided example the case of the ribosomal L5 C terminal domain protein family which you can find in the examples Ribosomal_L5 PF00673 directory We will use the graphical version of ProtTest ProtTest selection of models of protein evolution First we double click the ProtTest jar file If we have the proper java version a window will appear Then we enter the alignment we can use the file included in the examples folder at examples Ribosomal_L5 PF00673 alignment file We leave the other options as they are and press the Start button A new window will appear we will refer to this window as the console window to distinguish it from the main one In the console we can watch the progress of the analysis and check if everything is working properly We will see a header reporting some information about the alignment and below the results for each model as they are being optimized Something like The header bla bla Modeles canre Dubie aha TE a E ae 200 Goure a SIJ LT Number of parameters 39 0 39 branch length estimates SHUT E eee arava kee Gueth Boos where eee ates 2980 51 OhOm15s MOG iF se ehan ails ie oh be sae E S Susie tere JTT F Number of parameters 58 19 39 branch length estimates aminoacid frequencies observed s above eS TPT Mies toe ee ssa sate a a aa Mie eee eee ate aMe Ts 2977 23 OhOm6s Modekai a aA aie a A
20. lt the following will be displayed in the console ProtTest selection of models of protein evolution kkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkk Second order AIC AICc framework Sample size Total number of characters aligment length 113 00 kkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkk Best model according to second order AIC WAG G kkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkk Model deltaAICc AICc AICcw 1nL WAG I G 0 00 5956 28 0 76 2912 89 WAG G 2 34 5958 61 0 24 2916 53 Blosum62 I G 9 84 5966 12 0 01 2917 81 Blosum62 G 11 12 5967 39 0 00 2920 92 CpREV I G 19 31 5975 58 0 00 2922 54 Umm now the RtREV I G F and RtREV G F models dissapear from the top of the ranking The AlCc tells us that the sample size isn t large enough for supporting the addition of the F extra parameters in those models Using this framework we could say that WAG I G is the best fit model and that the WAG G model is also an interesting candidate What if the alignment length is an underestimation of the size of the sample If we try other sample size criteria we ll see that as sample size increases the support given by AlCc to more complex models is higher the larger the sample size the most similar the behaviour of AlCc compared to AIC is What about the BIC framework We may have a better perspective of the scenario if we click the overall comparison button in the console what resul
21. mber of sequences multiplied by the length of the alignment and multiplied by a correction factor the averaged Shannon s entropy normalized to a 0 1 scale see more details in the Phylogenetics and sample size section Results will be stored in a file named results txt The t1 T is telling ProtTest selection of models of protein evolution ProtTest to return the tree optimized for the best fit model in newick format This command line excludes MtArt MtMam and MtREV models from the analysis as well as the add on F To learn about the output of the program and its interpretation go to the section ProtTest s output A guided example New options It is now possible to restrict the set of candidate models in the command line version of ProtTest For example if we would like to discard the mitochondrial models we should type MtArt F MtREV F MtMam F In addition if we would like to discard 1 G models as well as F models just type I G F F F To discard different models proceed in a similar way ProtTest at the web ProtTest analysis can also be executed at its web site http darwin uvigo es Functionality of the web version of ProtTest is similar to the graphical one but the ability of restricting the set of candidate models and the ability to select interactivelly different model selection criteria is not provided as in the command line version Enter the web page and just in
22. nd BIC with three different criteria for sample size by clicking the overall comparison button If you want ProtTest to display the tree corresponding to the best model select between an ASCII representation and the tree in Newick format Under the fast strategy this tree becomes the topology of the initial tree BIONJ or provided by the user with branch lengths optimized under the best fit model under the current selection criterion Under the slow strategy this tree is the best ML tree both topology and branch lengths will be optimized under the best fit model under the current selection criterion ProtTest selection of models of protein evolution To learn about the output of the program and its interpretation go to the section ProtTest s output A guided example A080 ProtTest 1 0 6 About WWW Help e080 ProtTest console SE sequence 17 rL11 DROME PAL amp Sequence 18 RL11 HUMAN Sequence 19 RL11_CHLRE k Sequence 20 R111 ARATH r Alig Sequence 21 RL11 TETTH Alignment contains 21 sequences of length 113 PERE EEE EEE EEE EEE EEE TEETER EEE EEE EE EEE EEE E EE ET Observed number of invariable sites 8 Observed aminoacid frequencies gt aA 0 051 cC 0 009 D 0 061 E 0 062 F 0 070 G 0 091 H 0 019 I 0 074 K 0 078 L 0 082 7 OM 0 025 N 0 034 P 0 037 Q 0 021 R 0 087 S 0 042 T 0 052 v 0 066 W 0 006 Y 0 034 E E 39 0 39 branch length estimates 3003 21 0h0m7s
23. nsole interface Version 1 0 October 2004 First release of ProtTest Version tracking renumbered 11 ProtTest selection of models of protein evolution BACKGROUND Models of protein evolution Basically a model of protein evolution indicates the probability of change from a given amino acid to another over a period of time given some rate of change Although mecanistic models exist Thorne and Goldman 2003 models of protein evolution are preferentially based on empirical matrices for computational and data complexity reasons These matrices are constructed based on large datasets consisting of many diverse protein families The resulting matrices state the relative rates of replacement from one aminoacid to another The most common matrices which are the ones included in ProtTest are the Dayhoff Dayhoff et al 1978 JTT Jones et al 1992 WAG Whelan and Goldman 2001 mtREV Adachi and Hasegawa 1996 MtMam Cao et al 1998 VT Muller and Vingron 2000 CpREV Adachi et al 2000 RtREV Dimmic et al 2002 MtArt Abascal et al 2007 HIVb HIVw Nickle et al 2007 LG Le and Gascuel 2008 and Blosum62 Henikoff and Henikoff 1992 matrices Conservation of protein function and structure imposes constraints on which positions can change and which cannot This evolutionary information can be inferred from a multiple alignment but cannot be specified in a substitution matrix such as the empirical ones described below Fortunately
24. one of the accepted formats using programs such as MacClade Maddison and Maddison 1992 or ReadSeq Gilbert 2001 The second can be easily used through its web version http bimas dcrt nih gov molbio readseq You can get more information about alignment and tree formats at http workshop molecularevolution org resources fileformats ProtTest selection of models of protein evolution http workshop molecularevolution org resources fileformats tree formats php Restricting the set of candidate models This functionality is only available through ProtTest s graphical interface If the set of models you are interested in is a subset of those offered by ProtTest e g you want to select the best model for using a program that doesn t support the Dayhoff matrix you can restrict the set of candidate models by clicking the models button Then select which empirical matrices WAG Dayhoff JTT MtREV VT RtREV CpREV and Blosum62 and improvements l G F l G should be included in the analysis see figure 3 A ProtTest candidate models Substitution matrices Add ons Mi JTT M 1 M MtREV Mic v Dayhoff v 14 G M WAG MW F Mi RtREV M CpREV v Blosum62 MW vT Cancel OK Figure 3 Restricting the set of candidate models ProtTest s output A guided example the ribosomal L5 protein family using ProtTest version 1 2 2 In this section we ll explain the output of the program and its possib
25. p model 1 Slow optimize branch lengths model amp topology sample sample size for AICc and BIC corrections default 2 0 Shannon entropy Sum 1 Average 0 1 Shannon entropy x NxL 2 Total number of characters aligment length ProtTest selection of models of protein evolution 3 Number of variable characters 4 Alignment length x num taxa NxL 5 Specified by the user size number specifying sample size only for sample 5 EL T F If true display best model s newick tree default false E23 T F If true displya best model s ASCII tree default false verbose T F true or false default true model T F true or false default all models are set to true model JTT LG DCMut MtREV MtMam MtArt Dayhoff WAG RtREV CpREV Blosum62 VT HIVb HIVw addon T F true or false default all model addons are set to true addon F I I G G The only required option is i and it must be followed by the name of the file that contains the alignment Other options have their own values set by default which you may want to change If you want to specify a tree topology its branch lengths are of no importance they will be optimized use the t option followed by the tree file name this option doesn t matter when S 1 is also specified Set the sort option to indicate what statistic should be used for model sel
26. put an alignment and optionally a tree select the statistical criterion for model selection and the other parameters Your job will be sent to a queue and you ll be notified by e mail when the analysis is finished Optimization strategies Ideally one should optimize the tree topology its branch lengths and the model parameters for each model to assure maximum likelihood is achieved This complete optimization strategy can be performed by ProtTest when the slow option is selected in the main window However model selection seems to be quite robust to topology as long as this is a reasonable representation of the true phylogeny Posada and Crandall 2001 Therefore a faster strategy and the one implemented in the program Modeltest Posada and Crandall 1998 is to estimate a good tree and make all likelihood calculations for all models in this fixed tree This strategy is named fast in ProtTest Because it only optimizes branch lengths and model parameters it has the advantage of being much faster Alignment and tree formats ProtTest is able to read through the PAL library the following alignment formats phylip interleaved or sequential and nexus sequential The phylip format is recommended since the nexus reader has some bugs For reading trees the newick format is supported You can find examples of these formats in the formats examples directory If your data is in a different format you can convert it to
27. ts in a table like the following where the ranking of models the importance of parameters and other statistics are compared under seven frameworks model AIC AICc 1 AICc 2 AICc 3 BIC 1 BIC 2 BIC 3 WAG I G 0 49 1 0 76 1 0 86 1 0 86 1 0 78 1 0 74 1 0 57 1 RtREV 1I G F 0 31 2 0 00 19 0 00 11 0 04 3 0 00 12 0 00 12 0 00 18 RtREV G F 0 16 3 0 00 16 0 00 10 0 02 4 0 00 11 0 00 11 0 00 16 WAG G 0 04 4 0 24 2 0 13 2 0 07 2 0 22 2 0 25 2 0 42 2 Blosum62 I G 0 00 5 0 01 3 0 01 3 0 01 5 0 01 3 0 01 3 0 00 4 WAG I GtF 0 00 6 0 00 23 0 00 13 0 00 7 0 00 16 0 00 17 0 00 23 Blosum62 G 0 00 7 0 00 4 0 00 4 0 00 6 0 00 4 0 00 4 0 01 3 some lines omitted Relative importance of parameters AIC AICc 1 AICc 2 AICc 3 BIC 1 BIC 2 BIC 3 G 0 20 0 24 0 13 0 09 0 22 0 25 0 43 I 0 00 0 00 0 00 0 00 0 00 0 00 0 00 I G 0 80 0 76 0 87 0 91 0 78 0 75 0 57 F 0 47 0 00 0 00 0 06 0 00 0 00 0 00 Model averaged estimate of parameters AIC AICc 1 AICc 2 AICc 3 BIC 1 BIC 2 BIC 3 alpha G 3 02 2 01 2 01 2 05 2 01 2 01 2 02 p inv I 0 00 0 07 0 06 0 00 0 07 0 07 0 07 alpha I G 2 68 3 10 3 10 3 10 3 10 3 10 3 09 p inv I G 0 03 0 05 0 05 0 05 0 05 0 05 0 05 AIC Akaike Information Criterion framework AICc x Second Order Akaike framework BIC x Bayesian Information Criterion framework AICc BIC 1 sample size as number of sites in the alignment 113 0 AICc BIC 2 sample size as Sum of position
28. using ProtTest o ProtTest Abascal F Zardoya R Posada D 2005 ProtTest Selection of best fit models of protein evolution Bioinformatics 21 2104 2105 o Phyml Guindon S Gascuel O 2003 A simple fast and accurate algorithm to estimate large phylogenies by maximum likelihood Syst Biol 52 696 704 o PAL Drummond A Strimmer K 2001 PAL An object oriented programming library for molecular evolution and phylogenetics Bioinformatics 17 662 663 15 ProtTest selection of models of protein evolution References Abascal F Posada D and Zardoya R 2007 MtArt a new model of amino acid replacement for Arthropoda Mol Biol Evol 24 1 5 Adachi J and Hasegawa M 1996 Model of amino acid substitution in proteins encoded by mitochondrial DNA J Mol Evol 42 459 468 Adachi J Waddell P J Martin W and Hasegawa M 2000 Plastid genome phylogeny and a model of amino acid substitution for proteins encoded by chloroplast DNA J Mol Evol 50 348 358 Cao Y Adachi J Janke A Paabo S and Hasegawa M 1994 Phylogenetic relationships among eutherian orders estimated from inferred sequences of mitochondrial proteins instability of a tree based on a single gene Journal of Molecular Evolution 39 519 527 Cao Y Janke A Waddell P J Westerman M Takenaka O Murata S Okada N Paabo S and Hasegawa M 1998 Conflict among individual mitochondrial proteins in resolving the phylogeny of eutherian
Download Pdf Manuals
Related Search