########################### T-Coffee Main Documentation ########################### .. warning:: This chapter has been extensively updated in 11/2016. We improved its readibility and incorporate all the latest development of the past years (PSI/TM-Coffee for homology extension based MSAs, Pro-Coffee for functional DNA, R-Coffee/SARA-Coffee for RNA, TCS and STRIKE for evaluating MSAs, T-RMSD for structural clustering). If you have any suggestion or correction, don't hesitate to contact us. ******************* Before You Start... ******************* Foreword ======== Most of the work presented here emanates from two summer schools that were tentatively called the 'Prosite Workshops' and were held in Marseille, in 2001 and 2002. These workshops were mostly an excuse to go rambling and swimming in the creeks of Cassis (Calanques). Yet, when we got tired of lazing in the sun, we eventually did a bit of work to chill out. Most of our experiments were revolving around the development of sequence analysis tools. Many of the most advanced ideas in T-Coffee were launched during these fruitful sessions. Participants included Phillip Bucher, Laurent Falquet, Marco Pagni, Alexandre Gattiker, Nicolas Hulo, Christian Siegfried, Anne-Lise Veuthey, Virginie Leseau, Lorenzo Ceruti and Cedric Notredame. Prerequisite for using T-Coffee =============================== This documentation relies on the assumption that you have installed T-Coffee, version 9.03 or higher (preferably the latest stable version). T-Coffee is a freeware open source running on all Unix-like platforms including Mac OS X, woever, it cannot run on Windows except by using Cygwin, a freeware open source allowing to run a Unix-like command line on Windows (`download `_). A better option for Windows users would be to install a Unix-like virtualbox on your computer. All the relevant information for installing T-Coffee is contained in the previous chapter **T-Coffee Installation**. Let's have a try... =================== This manual is made to help you discover (nearly) all subtleties of T-Coffee, from default applications to sophisticated ones. All along this documentation, we expect you to use Unix-like shell commands running on prepared example files we have created. For now you can find all these files on our github repository for now: `example files `_; however in the future they will be included in the T-Coffee distribution (~/tcoffee/Version.../examples/). All T-Coffee commands indicated with a tag "$$:" are currently up and running; just copy/paste these commands from the documentation (without the tag "$$:") to your terminal provided you previously downloaded the example files. We encourage you to try all these examples which are controlled cases, and also to try these commands with your own data files. All T-Coffee commands indicated with a tag "$#:" are currently under maintenance. Also, be aware that all the values/results given in this manual were obtained with different versions of T-Coffee; depending on the version of T-Coffee you are using, these results may slightly differ. .. important:: Using T-Coffee package via "command lines" is the best/only way **if you need to do something sophisticated and/or highly reproducible**, but if you don't want to bother with command line you can use our `web server `_ or different links on the `Cedric Notredame's lab homepage `_. We have tried to put as many of T-Coffee functionalities on the webserver and we plan to incorporate even more in the future, but for now some options are not available or limited. ******************* What is T-Coffee ? ******************* What is T-Coffee? ================= T-Coffee stands for **"Tree based Consistency Objective Function For alignment Evaluation"**. It is mainly/primarily a Multiple Sequence Alignment (MSA) method but it also provides a collection of useful/powerful tools presented in this manual. Before going deep into the core of the matter, here are a few words to briefly explain the things T-Coffee can do !!! What does it do? ---------------- T-Coffee has two components allowing you to perform different tasks: - mainly it is a Multiple Sequence Alignment method. Given a dataset of sequences previously gathered using database search programs (like BLAST, Ensembl, etc...), T-Coffee will produce a Multiple Sequence Alignment (MSA) (refer to the section **Building Your Multiple Sequence Alignments** for more details). - also it is a collection of tools (usually called with the flag **-other_pg "tools"**) and third party software (cf. section **Integrating External Methods in T-Coffee**) able to perform a wide range of operations such as aligning accurately, reformatting data, evaluating results, comparing methods and many more... What can it align? ------------------ T-Coffee will align nucleic acid (DNA and RNA) and protein sequences alike. T-Coffee is also able to use other type of information such as secondary/tertiary structure information (for protein or RNA sequences with a known/predicted structure), sequence profiles, trees... .. Hint:: To have a rough idea, on an average computer, T-Coffee can easily align up to a 200 sequences, about 1000 amino acid long in about ~20min. It is better to avoid aligning more than 1000 sequences without using a cluster or T-Coffee fast modes !!! How can I use it? ----------------- T-Coffee is not an interactive program. It runs from your Unix or Linux command line and you must provide it with the correct parameters and syntax. If you do not like typing commands, you can still use the latest T-Coffee webserver `here `_. .. Tip:: Installing and using T-Coffee requires a minimum acquaintance with the Linux/Unix operating system. If you feel this is beyond your computer skills, we suggest you use one of the available online servers. Is there an online webserver? ----------------------------- Yes, at `Cedric Notredame's lab homepage `_ which contains all the necessary links to our latest `webserver `_ ! Is T-Coffee different from ClustalW? ------------------------------------ According to several benchmarks, T-Coffee is on overall much more accurate than ClustalW, but this increase in accuracy comes at a price: **T-Coffee (default mode) is slower than ClustalW** (about N times for N Sequences). Still, if you want to align closely related sequences, **T-Coffee can be used in a fast mode, much faster and about as accurate than ClustalW** (cf. section **Building Multiple Sequence Alignment**). If you are familiar with ClustalW or if you run a ClustalW server, you will find that we have made some efforts to ensure as much compatibility as possible between ClustalW and T-Coffee. Whenever it was relevant, we have kept the flag names and the flag syntax of ClustalW. Yet, you will find that T-Coffee also has many extra possibilities... Is T-Coffee very accurate? -------------------------- T-Coffee belongs to the category of consistency based aligners which currently corresponds to the most accurate algorithms available (e.g. ProbCons, MSAprobs...). In addition, T-Coffee can combine (many) methods and therefore be as accurate (and hopefully more) as the methods it combines. For instance, the "accurate" mode of T-Coffee is very slow but also very accurate; on average this mode was shown to be 10 % more accurate than normal aligners on sequences less than 30% similar. If you need a very accurate alignment go to section **Building Multiple Sequence Alignment**. What T-Coffee can and cannot do for you ... =========================================== What T-Coffee can't do ---------------------- To be honest, a short answer will be that there is only one thing T-Coffee cannot do for you: **T-Coffee can NOT fetch sequences for you**. You must select the sequences you want to align beforehand and prepare your own dataset. We suggest you use any BLAST server and format your sequences in FASTA so that T-Coffee can use them easily. The `ExPASy BLAST server `_ provides a nice interface for integrating database searches. What T-Coffee can do -------------------- T-Coffee is not only just an aligner program, it comes with multiple tools and third party software increasing the range of its possibilities; here is a non exhaustive list of tasks T-Coffee can perform: - **T-Coffee can compute (or at least try to compute!) accurate Multiple Sequence Alignments of DNA, RNA or Protein sequences**. Several modes and options are available and will be presented all along this manual. The default T-Coffee accepts any kind of sequence, although some modes are specific to a given type of sequence. - **T-Coffee can help you to reformat, trim, clean, cut, color your input (sequences, structures...) or output (alignments, trees...) data**; meaning that once you have your data and/or results ready, you can always modify them at will. - **T-Coffee allows you to combine results obtained with several alignment methods** (see the section **FAQ for T-Coffee** and **Building Multiple Sequence Alignment** for more details). T-Coffee can virtually combine all these MSAs you have to produce a new Multiple Sequence Alignment having the best agreement with all these methods you tried. - **One of the most important improvement of T-Coffee is to let you combine sequences and structures**, so that your alignments are of higher quality. You need to have the SAP package installed to fully benefit of this facility (or to use another structural alignment method). - **T-Coffee allows you to extract a collection of repeats from a single sequence or a set of sequences** using MOCCA. In other words, if you know the coordinates of one copy of a repeat, you can extract all the other occurrences. MOCCA needs some time to compute a library and then prompt you with an interactive menu. You just have to follow the instructions. - **T-Coffee can be used to measure the reliability of your Multiple Sequence Alignment**. If you want to find out about that, read the section **FAQ for T-Coffee** or the **Technical Documentation** (-output flag). More details will be given anyway in this manual in the section **How Good Is Your Alignment?**. - **T-Coffee can be used to compare alternative alignment**; in case you generate several alignments of the same sequences, you can compare these alignments using the most common scores (Sum-of-Pairs or Column Score). In case you have reference alignments, you can directly benchmark your method by comparing your MSAs to your references. And probably many more options we will discover together all along this manual ! .. warning:: Some options are carried out using the function "wget". If "wget" is not installed on your system, you can get it for free from `wget download `_. Just type **wget** to be sure it is installed. How does T-Coffee alignment works? ================================== If you only want to make a standard Multiple Sequence Alignment, you may skip these explanations. But if you want to do more sophisticated things, these few indications may help before you start reading the documentation and the different articles. When you run T-Coffee, the first thing it does is to compute a library. The library is a list of pairs of residues that could be aligned...it is like a christmas list: you can ask anything you fancy, it doesn't imply you will get it. Given a standard library, it is nearly impossible to have all the residues aligned at the same time because all the lines of the library may not agree. For instance: :: Line 1 says: Residue 1 of seq A with Residue 5 of seq B, ... Line 100 says: Residue 1 of seq A with Residue 29 of seq B, Each of these constraints comes with a weight and in the end, the T-Coffee algorithm tries to generate the multiple alignment that contains constraints whose sum of weights yields the highest score. In other words, it tries to make happy as many constraints as possible (replace the word constraint with, friends, relatives, collaborators... and you will know exactly what we mean). You can generate this list of constraints the way you like. You may even provide it yourself, forcing important residues to be aligned by giving them high weights (see **FAQ for T-Coffee**). For your convenience, T-Coffee can generate (by default) its own list by making all the possible global pairwise alignments, and the 10 best local alignments associated with each pair of sequences. Each pair of residues observed aligned in these pairwise alignments becomes a line in the library. .. note:: Be aware that nothing forces you to use a given library and that you could build it using other methods. In protein language, **T-Coffee is synonymous for freedom, the freedom of being aligned however you fancy** (I was probably a Tryptophan in some previous life). ******************* Preparing Your Data ******************* .. important:: About the syntax of T-Coffee command lines, just for you to know that it is quite flexible, for instance you can use any kind of separator you want (i.e. , ; =). The syntax used in this document is meant to be consistent with that of ClustalW. However, in order to take advantage of the automatic filename completion provided by many shells, you can replace '=' and ',' with a space. Also, T-Coffee tools/modes are called using different flags to specify input/output files, parameters, modifiers, etc...Some flags are not always mandatory, however, if you use a correct/strict flag usage T-Coffee will always work fine; you just have some degrees of freedom ;-). The reformatting utility: seq_reformat ====================================== General introduction -------------------- Nothing is more frustrating than downloading important data and realizing you need to format it before using it. In general, you should avoid manual reformatting: it is by essence inconsistent and will get you into trouble. It will also get you depressed when you realize that you have spent the whole day adding carriage return to each line in your files. T-Coffee comes with several tools to reformat/trim/clean/select your input data but also your output results, especially a very powerful reformatting utility named **seq_reformat**. You can use **seq_reformat** by invoking the t_coffee shell: :: $$: t_coffee -other_pg seq_reformat This will output the online flag usage of seq_reformat meaning a complete list of things seq_reformat can do for you. The seq_reformat is a reformatting utility so it recognizes automatically the most common formats (FASTA, Swiss-Prot,ClustalW, MSF, Phylip...). It reads the input file(s) via the **"-in"** and **"-in2"** flags and outputs in whatever specified format via the **"-output"** flag. In the meantime, you can use the flag **"-action"** to perform a wide range of modification on your data. In this section we give you quite a lot of different examples of you can do with **"-other_pg seq_reformat"**. .. danger:: After the flag **-other_pg**, the common T-Coffee flags are not recognized anymore; it is like if you were using a different program. .. tip:: When using T-Coffee seq_reformat, command line may become quite long...a practical way to handle this is to create in your .bashrc an alias to call directly seq_reformat. For example, write in your .bashrc: **alias reformat='t_coffee -other_pg seq_reformat'**. You can now call seq_reformat by tiping **reformat**. Modification options -------------------- In order to perform different modifications on your data (residues/sequences/columns...), the **seq_reformat** utility can be followed by the flag **-action** and one or several modifiers listed here (this list is not exhaustive): :: Options: - +upper : to uppercase your residues - +lower : to lowercase your residues - +switchcase : to selectively toggle the case of your residues - +keep : to only keep the residues within the range - +use_cons +keep : to only keep the columns within the range - +remove : to remove the residues within the range - +convert : to only convert the residues within the range - +grep : to select a given string of character - +rm_gap : to remove columns containing gaps - etc... Using a "cache" file -------------------- What is a cache in T-Coffee? ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Several options can be performed easily by using what we call a cache (or cache file). In T-Coffee, a cache is a file containing an alternate version of your alignment where each position of the alignment is replaced by an alternative coding scheme. For instance each residue can be replaced by a score previously evaluated: this score can be the T-Coffee CORE index (cf. section **Evaluating Your Alignment**) or a matrix based evalution (blosum62mt or identity matrix). Then, when performing any modification or reformatting of your alignments, you can just specify the range of positions to be modified according to their respective scores within the cache. We will see some example especially regarding the modification of format of a given alignment; it is not mandatory to use a cache but it is quite practical. To generate a cache before any reformatting using a given evaluation score, you can use one of the following possible option: :: Evaluating the T-Coffee CORE index during the alignment procedure: $$: t_coffee sample_seq1.fasta -output=score_ascii Evaluating the T-Coffee CORE index of a given alignment (-infile is mandatory): $$: t_coffee -infile sample_seq1.aln -mode evaluate Using an identity matrix: $$: t_coffee -other_pg seq_reformat -in sample_seq1.aln -action +evaluate \ idmat -output=score_ascii Using a substitution matrix: $$: t_coffee -other_pg seq_reformat -in sample_seq1.aln -action +evaluate \ blosum62mt -output=score_ascii Preparing a sequence/alignment cache ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ The following command will convert your alignment according to the given parameters: all A/a will be turned into 1, the gaps "-" will be conserved and all the other symbols (#) will be turned into 0. The flag **-action +convert** indicates the actions that must be carried out on the alignment before it is output into cache. :: 1) Initial alignment: CLUSTAL FORMAT A CTCCGTgTCTAGGagt-TTACGTggAGT B CTGAGA----AGCCGCCTGAGGTCG--- D CTTCGT----AGTCGT-TTAAGAca--- C -TTAAGGTCC---AGATTGCGGAGC--- 2) Command line: $$: t_coffee -other_pg seq_reformat -in=sample_dnaseq3.aln -output=clustalw_aln -out=cache.aln \ -action +convert 'Aa1' '.-' '#0' 3) Cache generated: CLUSTAL W (1.83) multiple sequence alignment A 0000000000100100-00100000100 B 000101----100000000100000--- D 000000----100000-00110101--- C -001100000---101000000100--- Other alternative are possible, for instance generating a cache for unaligned sequences (**-in** can refer to an alignment or unaligned sequences): :: 1) Command line: $$: t_coffee -other_pg seq_reformat -in=sample_dnaseq3.aln -output=fasta_seq -out=cache.seq \ -action +convert 'Aa1' '.-' '#0' 2) Generate the file cache.seq: >A 0000000000100100000100000100 >B 0001010000100000000100000000 >D 0000000000100000000110101000 >C 0001100000000101000000100000 Preparing a library cache ^^^^^^^^^^^^^^^^^^^^^^^^^ The library is a special format used by T-Coffee to declare relationships between pairs of residues. The cache library format can also be used to declare for instance the color of specific residues in an alignment. For instance, the following file ``sample_dnaseq3.tc_lib`` declares which residue of which sequence will receive which color. Note that the sequence number and the residue index are duplicated, owing to the recycling of this format from its original usage. It is also possible to use the BLOCK operator when defining the library (see **Technical Documentation**). The number right after BLOCK indicates the block length (10). The two next numbers (1 1) indicate the position of the first element in the block. The last value is the color. :: ! TC_LIB_FORMAT_01 4 A 27 CTCCGTgTCTAGGagtTTACGTggAGT B 21 CTGAGAAGCCGCCTGAGGTCG C 21 TTAAGGTCCAGATTGCGGAGC D 20 CTTCGTAGTCGTTTAAGAca #1 1 +BLOCK+ 10 1 1 3 +BLOCK+ 5 15 15 5 #3 3 6 6 1 9 9 4 ! CPU 240 ! SEQ_1_TO_N Modifying the format of your data ================================= Keeping/Protecting your sequence names -------------------------------------- Only few programs support long sequence names, and sometimes, when going through some pipeline the names of your sequences can be truncated or modified. To avoid this, **seq_reformat** contains a utility that can automatically rename your sequences into a form that will be machine-friendly, while making it easy to return to the human-friendly form. 1) **Create a code list**: The first thing to do is to generate a list of names that will be used in place of the long original name of the sequences: :: $$: t_coffee -other_pg seq_reformat -in proteases_large.fasta -output \ code_name > proteases_large.code_name 2) **Code your data**: This will create a file where each original name is associated with a coded name (Cxxx). You can then use this file to either code your dataset using the following command: :: $$: t_coffee -other_pg seq_reformat -code proteases_large.code_name -in \ proteases_large.fasta > proteases_large.coded.fasta 3) **Decode your data**: Then you can work with the file sproteases_large.coded.fasta and when you are done, you can decode the names of your sequences with the following command line: :: $$: t_coffee -other_pg seq_reformat -decode proteases_large.code_name -in \ proteases_large.coded.fasta Changing the sequence format ---------------------------- Sometimes it may be necessary to change from one format to another, for instance when using another software which recognize only a given format. T-Coffee recognizes most common alignment formats and you can find the list of all input or output format recognized by simply typing **t_coffee -other_pg seq_reformat** without any input file(s). It is possible to reformat unaligned or aligned sequences alike although changing the alignment format is probably more interesting in order to use other applications; unaligned sequences format flags are generally preceded by the suffix **"_seq"** and aligned sequences flags by the suffix **"_aln"**. This also allows you to transform any alignment into unaligned sequences by removing the gaps. Here are some examples on how to change the format of your data: :: For unaligned sequences (e.g. FASTA to PIR): $$: t_coffee -other_pg seq_reformat -in proteases_small.fasta -output pir_seq > \ proteases_small.pir_seq For alignements (e.g. ClustalW to MSF): $$: t_coffee -other_pg seq_reformat -in proteases_small.aln -output fasta_aln > \ proteases_small.fasta_aln From aligned to unaligned sequences: $$: t_coffee -other_pg seq_reformat -in proteases_small.aln -output fasta_seq > \ proteases_small.fasta .. Warning:: Format recognition is not 100% full proof; occasionally you will have to inform the program about the nature of the file you are trying to reformat with **-input msf_aln -output fasta_aln** for instance. Changing the case ----------------- Changing the case of your sequences ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ If you need to change the case of your sequences, you can use different modifiers embedded in **seq_reformat**. They are accessed via the **-action** flag. For instance, to write your sequences in lower case: :: $$: t_coffee -other_pg seq_reformat -in proteases_small.aln -action +lower \ -output clustalw .. tip:: No prize for guessing that **+upper** will do exactly the opposite... Changing the case of specific residues ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ If you want to change the case of a specific residue, you can use the flag: **+edit_residue **. If you have more than one residue to modify, write all the coordinates in a text file (one coordinate per line) as spans are not yet supported; then give the file to T-Coffee :: $$: t_coffee -other_pg seq_reformat -in sample_seq1.aln -action +upper \ +edit_residue hmgb_chite 10 lower $$: t_coffee -other_pg seq_reformat -in sample_seq1.aln -action +upper \ +edit_residue sample_seq1.list .. warning:: If you give a list of coordinates, it has to be a Unix text file (not a word document). Changing the case with a cache ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ If you want to change the case depending on the score, you must either evaluate your alignment, or provide a cache. For example, this command line will upper the case of all residue then lower the case of every residue more than 50% identical to other residues in the same column: :: Using a cache on the fly: $$: t_coffee -other_pg seq_reformat -in sample_dnaseq2.aln -action +upper \ +evaluate idmat +lower '[5-9]' Using a cache file previously computed (2 steps): $$: t_coffee -other_pg seq_reformat -in sample_dnaseq2.aln -action +evaluate \ idmat -output=score_ascii > sample_dnaseq2.cache $$: t_coffee -other_pg seq_reformat -in sample_dnaseq2.aln -struc_in sample_dnaseq2.cache \ -struc_in_f number_aln -action +lower '[5-9]' Coloring/Editing residues in an alignment ----------------------------------------- Changing the default colors ^^^^^^^^^^^^^^^^^^^^^^^^^^^ Colors are hard coded in the program, but if you wish, you can change them by simply creating a file named ``seq_reformat.color`` that is used to declare the color values. The name of the file (``seq_reformat.color``) is defined in ``programmes_define.h``, COLOR_FILE and can be changed before compilation. By default, the file is searched in the current directory. For example, the following line written in ``seq_reformat.color`` indicates that the value 0 in the cache corresponds now to #FFAA00 in html, and in RGB 1, 0.2 and 0. :: Format for hard coded colors in T-Coffee: 0 #FFAA00 1 0.2 0 Coloring specific types of residues/nucleic acids ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ You can color all the residues of your sequences on the fly; for instance, the following command line will color all the a's in color 0 (blue): :: $$: t_coffee -other_pg seq_reformat -in sample_seq1.aln -action +3convert a0 \ -output color_html > sample_seq1_color.html .. warning:: This option is case sensitive so the case of the residues or nucleotides should be the same in the command line (in this command line, only a lower case will be colored). Coloring a specific residue of a specific sequence ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ If you want to color a specific residue/nucleotide, you can use the flag **+color_residue **. If you have more than one residue to color, you can put all the coordinates in a file, (one coordinate per line). Spans are not yet supported. :: $$: t_coffee -other_pg seq_reformat -in sample_seq1.aln -action +color_residue \ hmgb_chite 10 1 -output color_html > sample_seq1_single.html $$: t_coffee -other_pg seq_reformat -in sample_seq1.aln -action +color_residue \ sample_seq1_color.list -output color_html > sample_seq1_all.html .. warning:: If you give a list of coordinates, it has to be a Unix text file (not a word document). Coloring according to the conservation ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Use the +evaluate flag if you want to color your alignment according to its conservation level or using the boxshade scoring scheme: :: Conservation color scheme: $$: t_coffee -other_pg seq_reformat -in sample_seq1.aln -action +3evaluate pam250mt \ -output color_html > color_cons.html Boxshade color scheme: $$: t_coffee -other_pg seq_reformat -in sample_aln1.aln -action +3evaluate boxshade \ -output color_html > color_cons_box.html Coloring an alignment using a cache ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ If you have a cache alignment or a cache library, you can use it to color your alignment and either make a post script, html or PDF output. For instance, if you use the file cache.seq: :: Produces a html file: $$: t_coffee -other_pg seq_reformat -in=sample_dnaseq3.aln -struc_in=sample_dnaseq3.cache \ -struc_in_f number_fasta -output=color_html -out=color_dnaseq3.html Produces a pdf file: $*: t_coffee -other_pg seq_reformat -in=sample_dnaseq3.aln -struc_in=sample_dnaseq3.cache \ -struc_in_f number_fasta -output=color_pdf -out=color_dnaseq3.pdf Produces an output using a library: $$: t_coffee -other_pg seq_reformat -in=sample_dnaseq3.aln -struc_in=sample_dnaseq3.tc_lib \ -output=color_html -out=color_dnaseq3_lib.html .. warning:: The script **ps2pdf** must be installed on your system for the pdf options. Modifying the data itself... ============================ Modifiying sequences in your dataset ------------------------------------ Converting residues ^^^^^^^^^^^^^^^^^^^ It is possible for instance to selectively convert all given characters in a sequence (residues or nucleic acids alike) into another one, for example all G's having a score between 5 and 9 by using the command line: :: $$: t_coffee -other_pg seq_reformat -in sample_dnaseq2.aln -struc_in sample_dnaseq2.cache \ -struc_in_f number_aln -action +convert '[5-9]' GX Extracting sequences according to a pattern ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ You can extract any sequence by requesting a specific pattern to be found either in the name (NAME), the comment (COMMENT) or the sequence (SEQ) using the modifier is '+grep'. For instance, if you want to extract all the sequences whose name contain the word HUMAN, the flag NAME/COMMENT/SEQ indicates that the modification is made according to the sequences names, the comment section or the sequence itself, and the flag KEEP/REMOVE means that you will keep/remove all the sequences containing the string HUMAN. Here are some examples: :: To keep sequences containing HUMAN in the name: $$: t_coffee -other_pg seq_reformat -in proteases_large.aln -action +grep NAME \ KEEP HUMAN -output clustalw To remove sequences containing HUMAN in the name: $$: t_coffee -other_pg seq_reformat -in proteases_large.aln -action +grep NAME \ REMOVE HUMAN -output clustalw To keep sequence which contain sapiens in the comment: $$: t_coffee -other_pg seq_reformat -in proteases_large.fasta -action +grep COMMENT \ KEEP sapiens -output clustalw To remove sequences containing the pattern [ILM]K: $$: t_coffee -other_pg seq_reformat -in proteases_large.aln -action +grep SEQ \ REMOVE '[ILM]K' -output clustalw .. important:: You should know that the pattern can be any perl legal regular expression, you can visit this `page `_ for some background on regular expressions. .. caution:: This option is case sensitive (Human, HUMAN and hUman will not yield the same results). Be careful !!! Extracting/Removing specific sequences by names ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ If you want to extract (command 1) or remove (command 2) several sequences in order to make a subset, you can specify a list of sequences by their full name: :: Command 1: keep sequences $$: t_coffee -other_pg seq_reformat -in proteases_small.fasta -action +extract_seq_list \ 'sp|P29786|TRY3_AEDAE' 'sp|P35037|TRY3_ANOGA' Command 2: remove sequences $$: t_coffee -other_pg seq_reformat -in proteases_small.fasta -action +remove_seq \ 'sp|P29786|TRY3_AEDAE' 'sp|P35037|TRY3_ANOGA' .. note:: Note the single quotes (') are mandatory as they are meant to protect the name of your sequence and prevent the Unix shell to interpret it like an instruction. Once sequences are extracted or removed, some columns may remain containing only gaps, but it is possible to simply remove empty columns from the resulting dataset (command 3), and even extract specific blocks for the selected sequences either keeping the exact same name (command 4) or the name of the specific blocks extracted (command 5): :: Command 3: removing empty columns $$: t_coffee -other_pg seq_reformat -in proteases_small.aln -action +extract_seq_list \ 'sp|P29786|TRY3_AEDAE' 'sp|P35037|TRY3_ANOGA' +rm_gap Command 4: keeping the initial name after extracting specific blocks and removing empty columns $$: t_coffee -other_pg seq_reformat -in proteases_small.aln -action +keep_name \ +extract_seq 'sp|P29786|TRY3_AEDAE' 20 200 'sp|P35037|TRY3_ANOGA' 10 150 +rm_gap Command 5: renaming sequences according to the extracted blocks and removing empty columns $$: t_coffee -other_pg seq_reformat -in proteases_small.aln -action +extract_seq \ 'sp|P29786|TRY3_AEDAE' 20 200 'sp|P35037|TRY3_ANOGA' 10 150 +rm_gap .. hint:: The tag **+keep_name** must come BEFORE the tag **+extract_seq**. Extracting the most informative sequences ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Large datasets are problematic because they can be difficult to align and analyze, MSA programs tend to become very slow and inaccurate. In short, the best size for an MSA dataset would be between 20 to 40 sequences to have enough sequences to see the effect of evolution, but in the same time small enough so that you can visualize your alignment and recompute it as many times as needed. More important than its size, a good dataset have to be informative, when each sequence contains information the others do not have. The most informative sequences are the sequences that are as different as possible to one another, within your dataset. You can extract the most informative sequences using flag **+trim** followed by the number of sequences you wish to keep ("n" for a number and "N" for a perrcentage). The following commands will extract the 10 most informative sequences (command 1) or the 20% of most informative sequences (command 2): :: Command 1: $$: t_coffee -other_pg seq_reformat -in proteases_large.fasta -action +trim _seq_n10 \ -output fasta_seq Command 2: $$: t_coffee -other_pg seq_reformat -in proteases_large.fasta -action +trim _seq_N20 \ -output fasta_seq .. hint:: The argument to trim include **_seq_**, it means your sequences are provided unaligned. If your sequences are already aligned, you do not need to provide this parameter. It is generally more accurate to use unaligned sequences. .. note:: For very large dataset, **seq_reformat** will compute the similarity matrix between your sequences once only. It will then store it in its cache to be reused any time you run on the same dataset. In short this means that it will take much longer to run the first time, but be much faster if you need to rerun it. Extracting/Removing sequences with the % identity ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ **Removing too identical sequences (redundant)** Removing the most similar sequences is often what people have in mind when they talk about removing redundancy. You can do so using the **+trim** option. For instance, you can generate a dataset where no pair of sequences has more than 50% identity either from a dataset of unaligned sequences (command 1) or from any given alignment (command 2). If you start from unaligned sequences, the removal of redundancy can be slow. If your sequences have already been aligned using a fast method, you can take advantage of this by replacing the **"_seq_"** with **"_aln_"**. Just run the following command lines to see the difference un runtime: :: Command 1: unaligned sequences $$: t_coffee -other_pg seq_reformat -in proteases_large.fasta -action +trim _seq_%%50_ Command 2: aligned sequences $$: t_coffee -other_pg seq_reformat -in proteases_large.fasta -action +trim _aln_%%50_ .. note:: Using aligned sequences results in a fastest trimming, however, it also means that you rely on a more approximate estimation of sequence similarity. **Removing too different sequences (outliers)** Sequences that are too distantly related from the rest of the set (called outliers) may have very negative effects on the overall alignment; to prevent this, it is advisable not to use them. The next command line will lead to the removal of all the sequences where no pair of sequences has less than 30% average accuracy with all the other sequences in the dataset (the symbol "_O" stands for Outliers) and more than 80% identity: :: $$: t_coffee -other_pg seq_reformat -in proteases_large.fasta -action +trim _seq_%%80_O30 .. hint:: This particular option is quite powerful as it allows you to decide both inferior and superior tresholds for trimming your dataset based on pairwise identity score, and therefore you can dissect your dataset according to different ranges of identity values. Be careful not to remove too many sequences ;-) **Forcing specific sequences to be kept** Sometimes you want to trim based on identity while making sure specific/important sequences remain in your dataset. You can do so by providing a pattern (**"_f"** for field) : it will keep all the sequences whose name contains the given string (**"_fNAME"**, **"_fCOMMENT"** or **"_fSEQ"**). Here are some examples corresponding to the different protected fields while removing all sequences above 50% identity: :: Keep all HUMAN sequences $$: t_coffee -other_pg seq_reformat -in proteases_large.fasta -action +trim \ _seq_%%50_fNAME HUMAN Keep all sequences containing ".apiens" $$: t_coffee -other_pg seq_reformat -in proteases_large.fasta -action +trim \ _seq_%%50_fCOMMENT '.apiens' Keep all sequences containing residues $$: t_coffee -other_pg seq_reformat -in proteases_large.fasta -action +trim \ _seq_%%50_fSEQ '[MLV][RK]' You can also specify the sequences you want to keep by giving another fasta file containing the name of these sequences via the flag **-in2**: :: $$: t_coffee -other_pg seq_reformat -in proteases_large.fasta -in2 proteases_small.fasta \ -action +trim _seq_%%40 Chaining important sequences ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ In order to align two distantly related sequences, most multiple sequence alignment packages perform better when provided with many intermediate sequences that make it possible to 'bridge' your two sequences. The modifier **+chain** makes it possible to extract from a dataset a subset of intermediate sequences that chain the sequences you are interested in. For instance, let us consider the two sequences "sp|P21844|MCPT5_MOUSE" and "sp|P29786|TRY3_AEDAE" having 26% identity. This is high enough to make a case for a homology relationship between them, but this is too low to blindly trust any pairwise alignment. With the names of the two sequences written in the file sproteases_pair.fasta, run the following command: :: $$: t_coffee -other_pg seq_reformat -in proteases_large.fasta -in2 proteases_pair.fasta \ -action +chain > proteases_chain.fasta This will generate a dataset of 21 sequences, with the following chain of similarity between your two sequences. This is probably the best way to generate a high quality alignment of your two sequences when using a progressive method like ClustalW, T-Coffee, MUSCLE or MAFFT. :: N: 21 Lower: 40 Sim: 25 DELTA: 15 #sp|P21844|MCPT5_MOUSE -->93 -->sp|P50339|MCPT3_RAT -->85 -->sp|P50341|MCPT2_M\ ERUN -->72 -->sp|P52195|MCPT1_PAPHA -->98 -->sp|P56435|MCPT1_MACFA -->97 -->sp\ |P23946|MCPT1_HUMAN -->81 -->sp|P21842|MCPT1_CANFA -->77 -->sp|P79204|MCPT2_SH\ EEP -->60 -->sp|P21812|MCPT4_MOUSE -->90 -->sp|P09650|MCPT1_RAT -->83 -->sp|P5\ 0340|MCPT1_MERUN -->73 -->sp|P11034|MCPT1_MOUSE-->76 -->sp|P00770|MCPT2_RAT --\ >71 -->sp|P97592|MCPT4_RAT -->66 -->sp|Q00356|MCPTX_MOUSE -->97 -->sp|O35164|M\ CPT9_MOUSE -->61 -->sp|P15119|MCPT2_MOUSE -->50 -->sp|Q06606|GRZ2_RAT -->54 --\ >sp|P80931|MCT1A_SHEEP -->40 -->sp|Q90629|TRY3_CHICK -->41 -->sp|P29786|TRY3_A\ EDAE Modifying columns/blocks in your dataset ---------------------------------------- Removing gapped columns ^^^^^^^^^^^^^^^^^^^^^^^ You can also remove all the columns containing a given proportion of gaps; for instance the following command will delete all the residues occurring in a column that contains 50% or more gaps (use 1 to delete residues from columns having 1 gap or more): :: $$: t_coffee -other_pg seq_reformat -in sample_dnaseq3.aln -action +rm_gap 50 Extracting specific columns ^^^^^^^^^^^^^^^^^^^^^^^^^^^ Extracting portions of a dataset is something very frequently needed. You may need to extract all the sequences that contain the word human in their name, or you may want all the sequences containing a simple motif. We show you here how to do a couple of these things. To do this, you need an evaluation file that may have been generated with T-Coffee, either running a *de novo* alignment (command 1) or evaluating a preexisting alignment (command 2): :: Command 1: $$: t_coffee sample_seq1.fasta -output score_ascii, aln Command 2: $$: t_coffee -other_pg seq_reformat -in sample_seq1.aln -action +evaluate \ blosum62mt -output score_ascii > sample_seq1_blosum62.score_ascii This generates a score_ascii file that you can then use to filter out the bad bits in your alignment considering the individual score of each residue to trigger the filtering (command 3), or according to the whole column score by simply add the **+use_cons** flag (command 4). This last command can also be run on the fly with command 5. The commands 3/4/5 will keep only residues and/or columns having a score between 6 and 9. :: Command 3: $$: t_coffee -other_pg seq_reformat -in sample_seq1.aln -struc_in sample_seq1.score_ascii \ -struc_in_f number_aln -action +keep '[6-9]' Command 4: $$: t_coffee -other_pg seq_reformat -in sample_seq1.aln -struc_in sample_seq1.score_ascii \ -struc_in_f number_aln -action +use_cons +keep '[6-9]' Command 5 $$: t_coffee -other_pg seq_reformat -in sample_aln1.aln -action +evaluate blosum62mt \ +use_cons +keep '[6-9]' .. warning:: Don't forget the simple quotes ('), it's mandatory !!! Extracting entire blocks ^^^^^^^^^^^^^^^^^^^^^^^^ In case you want to extract a specific block of your alignment, for instance to remove poorly resolved regions, delimit your alignments boundaries or to extract domains, you can do so with the option **+extract_block**. In command 1, the option **cons** indicates that you are counting the positions according to the consensus of the alignment (i.e. the positions correspond to the columns # of the alignment). If you want to extract your block relatively to a specific sequence, you should replace cons with this sequence name (command 2). :: Command 1: extract block from MSA $$: t_coffee -other_pg seq_reformat -in proteases_small.aln -action +extract_block \ cons 150 200 Command 2: extract_block relative to a give sequence of the MSA $$: t_coffee -other_pg seq_reformat -in proteases_small.aln -action +extract_block \ 'sp|Q03238|GRAM_RAT' 10 200 .. tip:: It may be sometimes difficult to know where starts the blocks you are interested in except by counting manually the number of column. You can also make some tries by modifying the boundaries until you get the block you want and then redirect the result into the output file name of your choice. Concatenating blocks or MSAs ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ If you have extracted several blocks generated using the previous command and you want to glue them together, you can use the **+cat_aln** modifier: :: $$: t_coffee -other_pg seq_reformat -in proteases_small.aln -action +extract_block \ cons 100 120 > block1.aln $$: t_coffee -other_pg seq_reformat -in proteases_small.aln -action +extract_block \ cons 150 200 > block2.aln $$: t_coffee -other_pg seq_reformat -in block1.aln -in2 block2.aln -action +cat_aln .. note:: The alignments do not need to have the same number of sequences and the sequences do not need to come in the same order. Manipulating DNA sequences ========================== Translating DNA sequences into protein sequences ------------------------------------------------ If your sequences are DNA coding sequences, it is often safer and more accurate to align them as proteins (as protein sequences are more conserved than their corresponding DNA sequence). The **seq_reformat** options make it easy for you to translate your sequences: :: $$: t_coffee -other_pg seq_reformat -in proteases_small_dna.fasta -action \ +translate -output fasta_seq Back-translation with the *bona fide* DNA sequences --------------------------------------------------- Once your sequences have been aligned, you may want to turn your protein alignment back into a DNA alignment, either to do phylogeny, or maybe in order to design PCR probes. To do so, use the following command: :: $$: t_coffee -other_pg seq_reformat -in proteases_small_dna.fasta -in2 \ proteases_small.aln -action +thread_dna_on_prot_aln -output clustalw Finding the *bona fide* sequences for the back-translation ---------------------------------------------------------- Use the online server `ProtoGen `_. Manipulating RNA Sequences ========================== Producing a Stockholm output: adding predicted secondary structures ------------------------------------------------------------------- Producing/Adding a consensus structure ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Given an RNA multiple sequence alignment, it is possible to compute (command 1) or add (command 2) the alifold (Vienna package) consensus secondary structure and output in stockholm: :: Command 1: $$: t_coffee -other_pg seq_reformat -in sample_rnaseq2.aln -action +aln2alifold \ -output stockholm_aln Command 2: $$: t_coffee -other_pg seq_reformat -in sample_rnaseq2.aln -action +add_alifold \ -output stockholm_aln Adding a precomputed consensus structure to an alignment ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ The file sample_rnaseq2.alifold contains the raw output of the alifold program produced via the RNAalifold `webserver `_ or captured with the command "RNAalifold sample_rnaseq2.alifold". It is possible to add this secondary structure to an alignment (command 1) and to stack Stockholm formatted secondary structures (command 2): :: Command 1: $$: t_coffee -other_pg seq_reformat -in sample_rnaseq2.aln -in2 sample_rnaseq2.alifold \ -input2 alifold -action +add_alifold -output stockholm_aln Command 2: $$: t_coffee -other_pg seq_reformat -in sample_rnaseq2.aln -in2 sample_rnaseq2.cons.stk \ -action +add_alifold -output stockholm_aln .. warning:: The alifold structure and the alignment MUST be compatible. The function makes no attempt to thread or align the structure, it merely stacks it below the MSA. Analyzing a RNAalifold secondary structure prediction ----------------------------------------------------- The following commands can either be applied on a Stockholm or a standard MSA. In the second case (standard MSA) the secondary structure will be automatically recomputed by alifold. Visualizing compensatory mutations ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ The following command will output a color coded version of your alignment with matching columns indicated as follows: - I: incompatible pair (i.e. at least one pair is not WC) - N: pairs are Gus or WC - W: all pairs are Watson - c: compensatory mutations - C: WC compensatory mutations :: Standard alignment: $$: t_coffee -other_pg seq_reformat -in sample_rnaseq2.aln -action +alifold2analyze aln Color coded alignment: $$: t_coffee -other_pg seq_reformat -in sample_rnaseq2.aln -action +alifold2analyze color_html .. warning:: Handling gapped columns: by default gapped column are ignored but they can be included by adding the tag **-usegap**. Analyzing matching columns ^^^^^^^^^^^^^^^^^^^^^^^^^^ The option **+alifold2analyze** will estimate the number of pairs of columns that are perfect Watson and Crick pairings, those that are neutral (including a GU) and those that include correlated mutations (command 1). The WCcomp are the compensated mutations maintaining WC base pairing. Other arguments can given, to display the list of paired positions and their status (compensated, Watson, etc...) use command 2: :: Command 1: $$: t_coffee -other_pg seq_reformat -in sample_rnaseq2.stk -action +alifold2analyze stat Command 2: $$: t_coffee -other_pg seq_reformat -in sample_rnaseq2.stk -action +alifold2analyze list Comparing alternative folds --------------------------- The folds associated with alternative alignments can be compared. This comparison involves counting how many identical pairs of residues are predicted on each sequence in one fold and in the other. The top of the output (@@lines) summarizes the results that are displayed on the input alignment; if the provided alignment do not have a fold, this fold will be estimated with alifold. The folds can be provided as Stockholm alignments: :: $$: t_coffee -other_pg seq_reformat -in sample_rnaseq2.cw.stk -in2 sample_rnaseq2.tcoffee.stk \ -action +RNAfold_cmp Phylogenetic Trees Manipulation =============================== Producing phylogenetic trees ---------------------------- The seq_reformat is NOT a phylogeny package, yet over the time it has accumulated a few functions that make it possible to compute simple phylogenetic trees, or similar types of clustering. Given a multiple sequence alignment, it is possible to compute either a UPGMA or an NJ tree. The following commands use an identity matrix to compare your sequences and will output an unrooted NJ tree in Newick format (command 1) or a rooted UPGMA tree (command 2): :: Command 1: $$: t_coffee -other_pg seq_reformat -in sample_seq1.aln -action +aln2tree -output newick \ -out sample_seq1_tree_nj.nwk Command 2: $$: t_coffee -other_pg seq_reformat -in sample_seq1.aln -action +aln2tree _TMODE_upgma \ -output newick -out sample_seq1_tree_upgma.nwk If your data is not data sequence, but a matrix of 1 and Os (i.e. SAR matrix for instance), you can use a different matrix to compute the pairwise distances (command 3), and all these parameters can be concatenated (command 4): :: Command 3: $#: t_coffee -other_pg seq_reformat -in sample_seq1.aln -action +aln2tree \ _MATRIX_sarmat -output newick Command 4: $#: t_coffee -other_pg seq_reformat -in sample_seq1.aln -action +aln2tree \ _TMODE_upgma_MATRIX_sarmat -output newick .. warning:: Bootstrap facilities will also be added at some point...We recommend you to use `Phylip `_ or any other specific phylogenetic software (PhyML, RAxML, MrBayes, etc...) if you need some serious phylogeny ! Comparing two phylogenetic trees -------------------------------- A real interesting option is the ability to compare two trees (unrooted) returning some ofthe most common scores used for this including Robinson-Foulds :: $$: t_coffee -other_pg seq_reformat -in sample_tree1.dnd -in2 sample_tree2.dnd -action \ +tree_cmp -output newick #tree_cmp|T: 33 W: 20.00 L: 14.88 RF: 2 N: 9 S: 5 #tree_cmp_def|T: ratio of identical nodes #tree_cmp_def|W: ratio of identical nodes weighted with the min Nseq below node #tree_cmp_def|L: average branch length similarity #tree_cmp_def|RF: Robinson and Foulds #tree_cmp_def|N: number of Nodes in T1 [unrooted] #tree_cmp_def|S: number of Sequences in T1 About the output scores in more details: - T: Fraction of the branches conserved between the two trees. This is obtained by considering the split induced by each branch and by checking whether that split is found in both trees. - W: Fraction of the branches conserved between the two trees. Each branch is weighted with MIN, the minimum number of leaf on its left or right. - L: Fraction of branch length difference between the two considered trees. - RF: is the standard Robinson-Foulds value when comparing trees. The last line contains a tree where distances have been replaced by the number of leaf under the considered node: - Positive values indicate a node common to both trees and correspond to MIN. - Negative values indicate a node found in tree1 but not in tree2. - The higher this value, the deeper the node. .. tip:: You can extract this tree for further usage by typing **cat outfile | grep -v 'tree_cmp'** Scanning phylogenetic trees --------------------------- It is possible to scan an alignment and locally measure the similarity between an estimated local tree and some reference tree provided from an external source (or computed on the fly) using the following command: :: $# : t_coffee -other_pg seq_reformat -in sample_seq1.aln -in2 sample_seq1_tree.nwk -action \ +tree_scan _MODE_scan__W_10_ > ph_tree_scan.txt For each position of the alignment, W*2 blocks of size 2*1+1 up to W*2+1 will be extracted, for each of these block a tree will be estimated and the similarity of that tree with the reference tree will be estimated with cmp_tree. For each position, the tree giving the best fit will be reported, along with the size of the block leading to that tree: :: P: Pruning phylogenetic trees -------------------------- Pruning removes leaves from an existing tree and recomputes distances so that no information is lost. To do this with T-Coffee you need two input files: a tree file in the Newick format and a FASTA-like file where sequences can be omitted, but you can also leave them, at your entire convenience. The second file is merely a list of the sequences to be kept when pruning the tree. The resulting tree will contain only the sequences specified in the list. :: Tree file: "sample_3Dseq1.tree" ((((TNFR10-2:0.04546,TNFR16-2:0.06640)...,TNFR4-2:0.05255) 45:0.00848); \ FASTA-like file: "sample_3Dseq1.fake" >TNFR2-2 >TNFR4-4 ... Pruning the tree: $$: t_coffee -other_pg seq_reformat -in sample_3Dseq1.tree -in2 sample_3Dseq1.fasta -action \ +tree_prune -output newick Tree Reformatting for MAFFT --------------------------- The most common format for phylogenetic trees is newick, but some alternatives exist, including the one used by MAFFT for very large trees, in which the sequence names are replaced by their original index (1..N) in the sequence file. Seq_reformat can be used to cross convert these two formats: :: $$: t_coffee -other_pg seq_reformat -in sample_3Dseq1.tree -in2 sample_3Dseq1.fasta -action +newick2mafftnewick And the reverse operation when processing a mafft guide tree :: $$: t_coffee -other_pg seq_reformat -in sample_3Dseq1.mafftnewick -in2 sample_3Dseq1.fasta -action +mafftnewick2newick Manipulating structure files (PDB) ================================== Extracting a structure ---------------------- There are many reasons why you may need a structure. T-Coffee contains a powerful utility named **extract_from_pdb** that makes it possible to fetch the PDB coordinates of a structure or its FASTA sequence without requiring a local installation. By default, the option **extract_from_pdb will** start looking for the structure in the current directory; it will then look it up locally (PDB_DIR) and eventually try to fetch it from the web (via a wget to the `PDB `_). All these settings can be customized using environment variables (see next paragraph). For instance if you want to fetch the chain E of the PDB structure 1PPG and/or its sequence in FASTA format, you can use: :: Fetch the structure: $$: t_coffee -other_pg extract_from_pdb -infile 1PPGE Fetch the correpsonding sequence: $$: t_coffee -other_pg extract_from_pdb -infile 1PPGE -fasta Adapting extract_from_pdb to your own environment ------------------------------------------------- If you have the PDB installed locally, simply set the variable PDB_DIR to the absolute location of the directory in which the PDB is installed. The PDB can either be installed in its divided form or in its full form. If the file you are looking for is neither in the current directory nor in the local PDB version, extract_from_pdb will try to fetch it from rcsb. If you do not want this to happen, you should either set the environment variable NO_REMOTE_PDB_DIR to 1 or use the **-no_remote_pdb_dir** flag: :: Setting up the environment: ##: export NO_REMOTE_PDB_FILE=1 Running PDB extract: $$: t_coffee -other_pg extract_from_pdb -infile 1PPGE -fasta -no_remote_pdb_file By default, T-Coffee also requires two important PDB files declared using the two following variables. These variables do not need to be set if the considered files are in the cache directory (default behavior): :: ##: export PDB_ENTRY_TYPE_FILE= (Found at: ftp://ftp.wwpdb.org/pub/pdb/derived_data/pdb_entry_type.txt) and ##: export PDB_UNREALEASED_FILE= (Found at: http://www.rcsb.org/pdb/rest/getUnreleased) .. warning:: Since the file ``unreleased.xml`` is not part of the PDB distribution, T-Coffee will make an attempt to obtain it even when using the NO_REMOTE_PDB_DIR=1 mode. You must therefore make sure that the file PDB_UNREALEASED_FILE is pointing to is read and write. *********************** Aligning Your Sequences *********************** General comments on alignments and aligners =========================================== What is a good alignment? ------------------------- This is a tricky question, a good answer would be **"a good alignment is an alignment that makes it possible to do good biology"**. In practice, the alignment community has become used to measuring the accuracy of alignment methods using structures. Structures are relatively easy to align correctly, even when the sequences have diverged quite a lot. The most common usage is therefore to compare structure based alignments with their sequence based counterpart and to evaluate the accuracy of the method using these criterions. Unfortunately it is not easy to establish structure based standards of truth. Several of these exist and they do not necessarily agree. To summarize, the situation is as roughly as follows: - **Above 40% identity**, all the reference collections do agree with one another and all the established methods give roughly the same results. These alignments can be trusted blindly. - **Below 40% identity**, all the reference collections stop agreeing and the methods do not give consistent results. In this area of similarity it is not necessarily easy to determine who is right and who is wrong, although most studies indicate that consistency based methods (T-Coffee, ProbCons, MAFFT-slow or MSAProbs) have an edge over traditional methods. When dealing with distantly related sequences, the only way to produce reliable alignments is to use structural information. T-Coffee provides many facilities to do so in a seamless fashion. Several important factors need to be taken into account when selecting an alignment method: - **The best methods are not always the best**. Given a difficult dataset, the best method is only more likely to deliver the best alignment, but there is no guaranty it will do so. It is very much like betting on the horse with the best odds. - **The difference in accuracy between all the available methods is not incredibly high** (as measured on reference datasets). It is unclear whether this is an artifact caused by the use of 'easy' reference alignments, or whether this is a reality. The only thing that can change dramatically the accuracy of the alignment is the use of structural information. - **Keep in mind that these methods have only been evaluated by comparison with reference alignments (benchmarks)**. This is merely one criterion among many. In theory, these methods should be evaluated for their ability to produce alignments that lead to accurate trees, good profiles or good models. Unfortunately, these evaluation procedures do not yet exist. The main methods and their scope -------------------------------- .. note:: There are many MSA packages around, the most common ones being ClustalW, MUSCLE, MAFFT, T-Coffee and ProbCons; amongst the latest ones, you can find phylogeny aware aligners (PRANK and SATé) and modifed/improved consistency based aligners (MSAProbs). You can almost forget about the other packages, as there is virtually nothing you could do with them that you will not be able to do with these packages. All these packages offer a complex trade-off between speed, accuracy and versatility. ClustalW is really everywhere... ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ClustalW is still the most widely used Multiple Sequence Alignment package. Yet things are changing fast and different tests have consistently shown that ClustalW is neither the most accurate nor the fastest package around. This being said, ClustalW is everywhere and if your sequences are similar enough, it should deliver a fairly reasonable alignment. MAFFT/MUSCLE to align big datasets ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ If you have many sequences to align MUSCLE or MAFFT are the obvious choice. MAFFT is often described as the fastest and the most efficient. This is not entirely true, in its fast mode (FFT-NS-1), MAFFT is similar to MUSCLE and although it is fairly accurate, about 5 points less accurate than the consistency based packages (ProbCons and T-Coffee). In its most accurate mode (L-INS-i) MAFFT uses local alignments and consistency, however, it becomes much more accurate but also slower, and more sensitive to the number of sequences. More recently, we have seen growing the number of **(ultra) large scale** aligners such as Clustal Omega, PASTA, UPP, and we hope soon the large scale version of T-Coffee (called MEGA-Coffee). **Suitable for**: - Distance-based phylogenetic reconstruction (NJ trees) - Secondary structure prediction **Not suitable for**: - Profile construction - Structure modeling - 3D prediction - Function analysis T-Coffee/ProbCons, slow but accurate !!! ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ T-Coffee works by first assembling a library and then by turning this library into an alignment. The library is a list of potential pairs of residues. All of them are not compatible and the job of the algorithm is to make sure that as many possible constraints as possible find their way into the final alignment: it is very much like trying to choose a meeting date, and each one says something like 'I need my Monday morning', 'I can't come on Thursday afternoon', and so on. In the end you want a schedule that makes everybody happy, if possible. The nice thing about the library is that it can be used as a media to combine as many methods as one wishes. It is just a matter of generating the right constraints with the right method and compile them into the library. ProbCons and MAFFT (L-INS-i) uses a similar algorithm, but with a Bayesian twist in the case of ProbCons. In practice, however, ProbCons and T-Coffee give very similar results and have similar running time. MAFFT is significantly faster. **Suited for**: - Profile reconstruction - Structure modeling - Function analysis - 3D prediction Choosing the right package (without flipping a coin !) ------------------------------------------------------ Each available package has something to go for it, it is just a matter of knowing what you want to do !! T-Coffee is probably the most versatile, but it comes at a price, its default aligner being currently slower than many alternative packages. In the rest of this tutorial we give some hints on how to carry out each of these applications within the T-Coffee framework. ================= ====== ===== ======== ======== ======== Packages MUSCLE MAFFT ProbCons T-Coffee ClustalW ================= ====== ===== ======== ======== ======== Accuracy ++ +++ +++ +++ \+ <100 Seq. ++ ++ +++ +++ \+ >100 Seq. +++ +++ \- \+ \+ Remote Homologues ++ +++ +++ +++ \+ MSA vs Seq. \- \- +++ +++ +++ MSA vs MSA \- \- \- +++ +++ >2 MSAs \- \- \- +++ \- Seq. vs Struc. \- \- \- +++ \+ Splicing Var. \- +++ \- +++ \- Reformat \- \- \- +++ ++ Phylogeny \- \- \- \+ ++ Evaluation \- \- \+ \+++ \- Speed +++ +++ \+ \+ ++ ================= ====== ===== ======== ======== ======== Table 1. Relative possibilities associated with the main packages. In any of the situations corresponding to each table line, (+++) indicates that the method is the best suited, (++) indicates that the method is not optimal but behaves reasonably well, (+) indicates that it is possible but not recommended (-) indicates that the option is not available. ===================== ====== ===== ======== ======== ======== Packages MUSCLE MAFFT ProbCons T-Coffee ClustalW ===================== ====== ===== ======== ======== ======== Dist Based Phylogeny +++ +++ ++ ++ ++ ML or MP Phylogeny ++ +++ +++ +++ ++ Profile Construction ++ +++ +++ +++ ++ 3D Modeling ++ ++ ++ +++ \+ 2D Predictions +++ +++ ++ ++ ++ ===================== ====== ===== ======== ======== ======== Table 2. Most Suitable Appplications of each package. In any of the situations corresponding to each table line, (+++) indicates that the method is the best suited, (++) indicates that the method is not optimal but behaves reasonably well, (+) indicates that it is possible but not recommended (-) indicates that the option is not available. Protein Sequences ================= General considerations ---------------------- T-Coffee aligner is by default parallelized, meaning that it can use multiple cores when running on a cluster or a computer. By default, T-Coffee will use all available processors to run, but you can parallelize the differents steps and allocate the number of cores you want with the flag **-multi_core** or **n_core**. For more details, refer to the chapter **T-Coffee Technical Documentation**, subsection **CPU control**. :: Using a single core: $$: t_coffee -in sample_seq1.fasta -multi_core=msa Using 12 cores: $$: t_coffee -in sample_seq1.fasta -n_core=12 Also when running T-Coffee, it displays a lot of information directly on screening while running from general information, options, results, warnings...if you want, you can reduce the display using **-no_warning** to remove all the warnings, or even more strict using **-quiet** removing any display while running T-Coffee. Computing a simple MSA (default T-Coffee) ----------------------------------------- T-Coffee default mode will simply compute a Multiple Sequence Alignment of the sequences you provided in input (command 1). It will display the final MSA on the screen and in several files according to the format you asked with command 2 (by default, the MSA is stored in a file .aln in ClustalW format). The headline of the alignment file contains important information such as the version of T-Coffee used, the CPU time, the overall consistency score (normalized to 100 or 1000 depending on the version of T-Coffee) and the total length of the MSA: it is quite practical to have a quick glance at the result. :: Command 1: default MSA $$: t_coffee proteases_small.fasta Command 2: default MSA, multiple output files $$: t_coffee proteases_small.fasta -output=clustalw,fasta_aln,msf Each time you run T-Coffee, 3 files are always generated: - ``proteases_small.aln``: the alignment in ClustalW format - ``proteases_small.dnd``: the guide tree in Newick format - ``proteases_small.html``: the colored MSA in html format .. warning:: The guide tree is not a phylogenetic tree, it is used in the alignment process for clustering the sequences. .. tip:: You can visualize the colored html file with any browser/software you prefer. The display of the sequences should be aligned and formatted; if not, use another browser, it works quite well with Firefox, Safari, etc... If you need to do more sophisticated modifications on your MSA, we recommend to use `Jalview `_ which incorporate the T-Coffee color scheme. Aligning multiple datasets/Combining multiple MSAs -------------------------------------------------- If your sequences are spread across several datasets, you can give all the files you want (the limit is 200) via the flag **-seq**, and in any format you want. Just know that 1) if you give an alignment, the gaps will be reset and your alignment will only provide sequences, 2) sequences with the same name between two files are assumed to be the same sequence, 3) ff their sequences differ, they will be aligned and replaced by the consensus of that alignment (process known as sequence reconciliation). To align multiple datasets: :: $$: t_coffee -seq=proteases1_small.fasta,proteases2_small.aln -output=clustalw,fasta_aln,msf You may also have a bunch of alignments (with the same sequences) that you have either precomputed, assembled manually or received from a colleague. You can also combine these alignments. For instance, let us imagine we generated 4 alignments with ClustalW using different gap penalties. To combine them into ONE single alignment, use the **-aln** flag. The final score indicates a high level of consistency (91%) between all these MSAs, meaning that the final MSA is probably correct. :: Your 4 different MSAs: ##: clustalw -infile=proteases_small.fasta -gapopen=0 -outfile=g0.aln ##: clustalw -infile=proteases_small.fasta -gapopen=-5 -outfile=g5.aln ##: clustalw -infile=proteases_small.fasta -gapopen=-10 -outfile=g10.aln ##: clustalw -infile=proteases_small.fasta -gapopen=-15 -outfile=g15.aln Combining multiple MSAs: $$: t_coffee proteases_small.fasta -aln g0.aln g5.aln g10.aln g15.aln -output=clustalw,html Estimating the diversity in your alignment ------------------------------------------ It is easy to measure the level of diversity within your MSA with the **-output** option of **seq_reformat**, it will output all the pairwise identities, as well as the average level of identity between each sequence and the others. There are two possibilities given that your input are unaligned sequences or not: **-output sim_idscore** realign your sequences pairwise so it can accept unaligned or aligned sequences alike; **-output sim** computes the identity using the sequences as they are in your input file so it is only suited for MSAs. You can after redirect, sort and grep the output in order to select the sequences you are interested in. :: $$: t_coffee -other_pg seq_reformat -in sample_seq1.aln -output sim Comparing alternative alignments -------------------------------- If you change the parameters, you will end up with alternative alignments. It can be interesting to compare them quantitatively. T-Coffee comes along with an alignment comparison module named **aln_compare**. You can use it to estimate the amount of difference between your two alignments either using the Sum-of-Pair score or the column score using the flag **-compare_mode** (sp or column). By default aln_compare returns the SoP score: :: $$: t_coffee -other_pg aln_compare -al1 b80.aln -al2 b30.aln -compare_mode sp This comparison will return the following result: :: ***************************************************** seq1 seq2 Sim [ALL] Tot b80 19 33.6 94.2 [100.0] [79686] The interpretation of this output is as follow: b80 is the reference MSA, it contains 19 sequences with an average identity of 33.6%, and is 94.2% identical to the second MSA b30.aln (79686 pairs to be precise). Of course, this does not tell you where are the good bits, but you can get this information for instance residues that have lost more than 50% of their pairing partner between the two alignments are in lower case (command 1) or converted in any character you want (command 2). The sp score is defined as the number of aligned residue pairs (i.e. not gaps) that are common between the reference (-al1) and the test (-al2) divided by the total numbers of pairs in the reference, while excluding gaps. Other modes of comparison are available including the TC score :: $$: t_coffee -other_pg aln_compare -al1 b80.aln -al2 b30.aln -compare_mode tc The tc score is defined as the number of columns in the test MSA (-al2) that are common between the reference and the test, divided by the total number of columns in the reference. This measure, originally reported in Balibase can be problematic as it treats equaly highly and sparcely populated columns. For this reason the :: $$: t_coffee -other_pg aln_compare -al1 b80.aln -al2 b30.aln -compare_mode column The column score is defined as the total number pairs occuring in columns entirely identical between the reference (-al1) and the test (-al2) divided by the total number of pairs in the reference (excluding gaps). This makes the column score the equivalent of a weighted TC score. It is also possible to print out the conserved positions between MSAs using the following commands. :: Command 1: $$: t_coffee -other_pg aln_compare -al1 b30.aln -al2 p350.aln -output_aln \ -output_aln_threshold 50 Command 2: $$: t_coffee -other_pg aln_compare -al1 b30.aln -al2 p350.aln -output_aln \ -output_aln_threshold 50 -output_aln_modif x .. note:: It is important to take into account that the comparison between -al1 and -al2 is not symetrical. It returns the fraction of -al1 residues pairs - by index - that are recovered in -al2. If -al1 is the reference alignment, this number is more or less the equivalent of a sensistivity measure (i.e. how much True positives are recovered). If -al1 is the tartget alignment, this measure becomes the equivalent of a specificity measure (i.e. what is the fraction of residue pairs in the target that are part of the reference). .. tip:: aln_compare is particularly interesting if you are modifying the default parameters of T-Coffee and want to monitor the effects of your modifications. Comparing subsets of alternative alignments ------------------------------------------- It is also possible to restrict the comparison to a specific subset of residues within the target alignment (for instance the core residues if using a structural reference, or the catalytic residues if using functional annotation alignment). This requires a cache in which residues are annotated as being part of the core or not. The cache is a plain text ascii file that can be produced in any way convenient to you. We show below how to produce a cache based on the consistency score and how to use this cache for further comparisons. :: $$: t_coffee b11.fa -output aln score_ascii This command generates two output files - ``b11.aln``: the alignment in ClustalW format - ``b11.score_ascii``: a local reliability score with every residue having a score between 0 and 9 The target file must then be reformatted so as to decide which residues will be part of the evaluation (c=> core) and which ones will be ignored. In this example we will set the threshild at 7 which means that the comparison will ignore all residues that do not have 70% or more support from the T-Coffee library: :: $$: t_coffee -other_pg seq_reformat -in b11.score_ascii -input number_aln -action +convert 0123456i 789c -output fasta_seq > b11.cache One can then produce a target file with another method :: $$: t_coffee b11.fa -method mafft_pair -outfile b11.tcmafft With the cache, it is then possible to compare the original alignment (b11.aln) while only considering pairs of core residues :: $$: t_coffee -other_pg aln_compare -al1 b11.aln -al2 b11.tcmafft -st b11.cache pep -io_cat '[c][c]=[core]' This comparison will return the following result: :: ***************************************************** seq1 seq2 Sim [core] Tot b11 11 32.8 99.5 [ 63.6] [26044] The output can be interpreted as follows: the core residues - defined as c-c pairs constitute 63.6% of the -al1 alignment. 93.5% of these pairs are recovered in the -al2 alignment. Note that the -io_cat makes it possible to declare as many categories and combinations of categories as one wishes to have. For instance: :: $$: t_coffee -other_pg aln_compare -al1 b11.aln -al2 b11.tcmafft -st b11.cache pep\ -io_cat '[c][c]=[core];[c][i]=[mixed];[ci][ci]=[all1];[*][*]=[all2]' Modifying the default parameters of T-Coffee -------------------------------------------- .. note:: The main parameters of T-Coffee are similar to those of ClustalW, including a substitution matrix and some gap penalties. In general, T-Coffee's default is adequate. If, however, you are not satisfied with the default parameters, we encourage you to change the following parameters. Interestingly, most of what we say here holds reasonably well for ClustalW. Can you guess the optimal parameters? ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Here is another tricky question...and the general answer is NO. The matrix and the gap penalties are simplistic attempts at modeling evolution. While the matrices do a reasonable job, the penalties are simply inappropriate: they should have a value that depends on the structure of the protein and a uniform value cannot be good enough. Yet, since we do not have better we must use them...In practice, this means that parameter optimality is a very *ad hoc* business. It will change from one dataset to the next and there is no simple way to predict which matrix and which penalty will do better. The problem is also that even after your alignment has been computed, it is not always easy to tell whether your new parameters have improved or degraded your MSA. There is no systematic way to evaluate an MSA. In general, people visually evaluate the alignment, count the number of identical columns and consider that one more conserved column is good news. If you are lucky you may know a few functional features that you expect to see aligned. If you are very lucky, you will have one structure and you can check the gaps fall in the loops. If you are extremely lucky, you will have two structures and you can assess the quality of your MSA. An advantage of T-Coffee is the fact that the overall score of the alignment (i.e. the consistency with the library) is correlated with the overall accuracy. In other words, if you alignment score increases, its accuracy probably increases also. All this being said, consistency is merely an empirical way of estimating the change of parameters and it does not have the predictive power of a BLAST E-Value. Changing the substitution matrix ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ T-Coffee only uses the substitution matrix to make the pairwise alignments that go into the library. These are all the global alignments of every possible pair of sequences, and the ten best local alignments associated with every pair of sequences. - By default, these alignments are computed using a Blosum62 matrix, but you can use any matrix you fancy instead, including: pam120mt, pam160mt, pam250mt, pam350mt, blosum30mt, blosum40mt, blosum45mt, blosum50mt, blosum55mt, blosum62mt, blosum80mt, or even user-provided matrices in the BLAST format (see **T-Coffee Technical Documentation**). - PAM matrices: These matrices are allegedly less accurate than the Blosum. The index is correlated to the evolutionary distances, you should therefore use the pam350mt to align very distantly related sequences. - Blosum matrices: These matrices are allegedly the most accurate. The index is correlated to the maximum percent identity within the sequences used to estimate the matrix. you should therefore use the blosum30mt to align very distantly related sequences. Blosum matrices are biased toward protein core regions, explaining why these matrices tend to give better alignments, since by design, they can capture the most evolutionary resilient signal contained in proteins. Unless you have some structural information available, the only way to tell whether your alignment has improved or not is to look at the score. For instance, if you compute the two following alignments: :: $$: t_coffee proteases_small.fasta -matrix=blosum30mt -outfile=b30.aln $$: t_coffee proteases_small.fasta -matrix=blosum80mt -outfile=b80.aln You will get two alignments that have roughly the same score but are slightly different. You can still use these two alternative alignments by comparing them to identify regions that have been aligned identically by the two matrices. These regions are usually more trustworthy. Changing gap penalties ^^^^^^^^^^^^^^^^^^^^^^ .. important:: Gap penalties are the core of the matter when it comes to MSAs. An interesting feature of T-Coffee is that it does not really need such penalties when assembling the MSA, because in theory the penalties have already been applied when computing the library. This is the theory, as in practice penalties can help improve the quality of the alignment. The penalties can be changed via the flags **-gapopen** for the gap opening penalty and via **-gapext** for the gap extension penalty. The range for gapopen are [-500,-5000], the range for the extension should rather be [-1,-10]. These values do not refer to a substitution matrix, but rather to the values range of the consistency estimation (i.e. ratio) normalized to 10000 for a maximum consistency. The default values are **-gapopen=-50, -gapext=0**. The reasons for these very low values are that they are meant to be cosmetic only, since a trademark of T-Coffee (inherited from Dialign) is not to need explicit penalties. Yet, we know for a fact that alignments with higher gap penalties often look nicer (for publications) and are sometimes more accurate. For instance, you can try: :: $$: t_coffee proteases_small.fasta -gapopen -100 -gapext -5 This gap penalty is only applied at the alignment level (i.e. after the library was computed). If you want to change the gap penalties of the methods used to build the library, you will need to go deeper...Two methods are used by default to build the library (command 1). One does global pairwise alignments and is named slow_pair, the other is named lalign_id_pair and produces local alignments. These methods are specified via the **-method** flag. Usually you do not need to write it because it is the default, but if you want to change the default parameters of the constituting methods (command 2), you will need to do so explicitly (the default parameters are for lalign_id_pair **GOP=-10, GEP=-4, MATRIX=blosum50mt** and for slow_pair **GOP=-10, GEP=-1 and MATRIX=blosum62mt**. Using the command 2, the library is now computed using the Blosum62mt with lalign, rather than the Blosum50mt; the good news is that when using this matrix, the score of our alignment increases from 48 to 50. We assume this new alignment is therefore more accurate than the previous one. :: Command 1: default T-Coffee $$: t_coffee proteases_small.fasta -method=lalign_id_pair,slow_pair Command 2: modifiying the parameters $$: t_coffee proteases_small.fasta -method lalign_id_pair@EP@MATRIX@blosum62mt, \ slow_pair -outfile proteases_small.b62_aln .. warning:: It only makes sense to compare the consistency score of alternative alignments when these alignments have been computed using the same methods (lalign_id_pair and slow_pair for instance). Aligning (very) large datasets ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ The consistency based default method is not really suited for more that 200 sequences. Above this number, it is recommanded to use the `regressive agorithm implemented in T-Coffee `_. Protein sequences using 2D and/or 3D information ================================================ Using structural information when aligning sequences is very useful. The reason is that structures diverge slower than sequences. As a consequence, one may still find a discernable homology between two sequences that have been diverging for a long time beyond recognition using their corresponding structure. Yet, when assembling a structure based MSA, you will realize that these sequences contain key conserved residues that a simple alignment procedure was unable to reveal. We show you in this section how to make the best of T-Coffee tools to incorporate structural information in your alignment. Using 3D structures: Expresso/3D-Coffee --------------------------------------- Requirements to run Expresso/3D-Coffee ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Expresso needs BLAST you provide with the tag **-blast** and it also requires the PDB database you can specify via the flag **-pdb_db**. The good news is that it is not mandatory to have BLAST or the PDB installed locally as T-Coffee can automatically fetch the structures directly from RCSB (the home of PDB) using EBI BLAST web service. How does Expresso/3D-Coffee work? ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Expresso/3D-Coffee is probably one of the most important improvement of T-Coffee. The principle is simple: 1) it first runs a BLAST for every sequence in your dataset against the PDB and finds (or not) a structure similar in sequence (35% identity by default) to be used as a template for structurally aligning your sequence, 2) the correspondence between the query sequences and the templates are stored in a template file which is automatically generated by Expresso. Here lies the difference between 3D-Coffee and Expresso: when running Expresso, fetching structures and creating the template file are automated (so you can reuse it for other applications) while using 3D-Coffee is a bit more tricky as it requires the name of the sequences to correspond to the structure file name (and it does not fetch or create anything for you). Of course, you can create and use your own template with the tag **-template_file** with the same format presented here. At the end, whenever there are enough templates (minimum of two obviously), it will align sequences using the structural information, otherwise sequences will be aligned using the standard T-Coffee aligner. Of course, if your dataset only contains structures, your alignment becomes a structural alignment. :: >sp|P08246|ELNE_HUMAN _P_ 1PPGE >sp|P20160|CAP7_HUMAN _P_ 1AE5 ... .. tip:: The PDB files used as a template should be in the current directory, otherwise you have to declare in the template file the whole path to find your templates. Running Expresso/3D-Coffee ^^^^^^^^^^^^^^^^^^^^^^^^^^ Both Expresso (command 1) and 3D-Coffee (command 2) are modes of T-Coffee you can call with the flag **-mode**; the correspond to a preestablished configuration with default parameters. The template selection is indicated with the flag **-template_file** followed by either **PDB** (means BLAST will run remotely on the PDB) or **_SELF_P_** (means that the PDB identifier is the name of the sequence so there is no need to run BLAST) or a template file of your choice. As you can see in commands 1 and2, SAP (sap_pair) is the default structural aligner but you can choose alternative aligner(s) installed. You can give any combination of methods with the flag **-method**, but at least one has to be a structural aligner (command 3). You can also specify local installations of BLAST and PDB (command 4) **which is highly recommended if you want reproducible results**. .. note:: you may need to install cpanm Mozilla::CA [sudo cpanm Mozilla::CA] to run the EBI web server :: Command 1: two ways of running Expresso $$: t_coffee three_pdb_two_seq.fasta -mode expresso $$: t_coffee three_pdb_two_seq.fasta -method sap_pair,slow_pair -template_file PDB Command 2: two ways of running 3D-Coffee $$: t_coffee three_pdb.fasta -mode 3dcoffee $$: t_coffee three_pdb.fasta -method sap_pair,slow_pair -template_file _SELF_P_ Command 3: Choosing your own templates and methods $$: t_coffee three_pdb_two_seq.fasta -method mustang_pair,slow_pair -template_file \ three_pdb_two_seq_pdb1.template_list Command 4: Running Expresso using a local BLAST/PDB ##: t_coffee three_pdb_two_seq.fasta -mode expresso -blast=LOCAL -pdb_db= \ -pdb_type d -pdb_min_sim 95 -pdb_min cov 90 -cache $PWD .. tip:: By default, structures and structural pairwise alignments are stored in your local ~/.t_coffee/cache/ allowing Expresso to run faster if you reuse similar structures; you can choose to have all these files directly in your working directory by using **-cache=$PWD**. Don't forget to empty your cache directory from time to time otherwise your folder is just getting bigger and bigger (similar comment can be done for any template based mode of T-Coffee). Template search paramaters ^^^^^^^^^^^^^^^^^^^^^^^^^^ To use Expresso, you have different option from an entirely automated procedure to tailored procedure, by selecting either your own structures or by defining different criteria for the template selection. You can have an exhaustive list in the **T-Coffee Technical Documentation** (subsection **Template based T-Coffee modes**) yet the most important parameters for the template selection are the following: - **-pdb_type** : type of structure ("d" for diffraction/XRAY or "n" NMR structures or "e" EM structures) - **-pdb_min_cov** : minimum coverage between query sequence and template (from 0-100%) - **-pdb_min_sim** : minimum identity between query sequence and template (from 0-100%) Aligning sequences and structures --------------------------------- Mixing sequence profile and structure templates ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ If you want to go further, and be even slower, you can use the accurate mode that will combine profile and structural information. If no structure is available, the template will be a profile (similar to PSI-Coffee, see subsection **Aligning Profiles**). It is probably one of the most accurate way of aligning sequences currently available as it tries to get as much information as possible. :: $$: t_coffee sample_seq1.fasta -mode accurate Aligning profile using structural information ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ If you have two profiles to align, an ideal situation is when your profiles each contain one or more structures. These structures will guide the alignment of the profiles, even if they contain very distally related sequences. All you need is a template file that declares which sequences have a known structure. If you only want to align sequences, you can try: :: $$: t_coffee -profile=profile1_pdb1.aln, profile2_pdb2.aln -method sap_pair \ -profile_template_file two_profiles.template_file Using secondary structure predictions ------------------------------------- T-Coffee can be used to predict secondary structures and transmembrane domains. For secondary structure predictions, the current implementation is only able to run GOR on either single sequences or on a bunch of homologues found by BLAST. Single sequence prediction ^^^^^^^^^^^^^^^^^^^^^^^^^^ To make a secondary structure prediction T-Coffee used the GOR software. In the command line **-template_file=SSP** is a hard coded mode which prompts the computation of predicted secondary structures. Used this way, the method will produce for each sequence a secondary prediction file (``.ssp``). GOR run on single sequences with a relatively low accuracy, which can be increased by coupling it with BLAST (command 2). When doing so, the predictions for each sequence are obtained by averaging the GOR predictions on every homologue as reported by a BLAST against nr. By default the BLAST is done remotely at the NCBI using the BLASTPGP web service of the EBI. Transmembrane structures can also be carried out simply, or following the same previous strategy (command 3 and 4). :: Command 1: $#: t_coffee sample_seq1.fasta -template_file SSP Command 2: $#: t_coffee sample_seq1.fasta -template_file PSISSP Command 3: $$: t_coffee sample_seq1.fasta -template_file TM Command 4: $$: t_coffee sample_seq1.fasta -template_file PSITM Incorporation of the prediction in the alignment ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ It is possible to use the secondary prediction (command 1) or the transmembrane domains prediction (command 2) in order to reward the alignment of similar elements: :: Command 1: $#: t_coffee sample_seq1.fasta -template_file PSISSP -method_evaluate_mode ssp -method \ lalign_id_pair Command 2: $$: t_coffee sample_seq1.fasta -template_file PSITM -method_evaluate_mode tm -method \ lalign_id_pair The overall effect is very crude and can go up to overweighting by 30% the score obtained when matching two residues in a similar secondary structure state. The net consequence is that residues in similar predicted states tend to be aligned more easily. .. hint:: In that case, the flag **-method** can only accept one single method Using other secondary structure predictions ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ If you have your own predictions, you can use them to run T-Coffee providing you give your own template file, where the file containing the secondary structure prediction is declared along with the sequence (with _E_). :: Command: $#: t_coffee sample_seq1.fasta -template_file sample_seq1_ssp.template -method_evaluate_mode \ ssp -method lalign_id_pair Format of the template file: >hmgl_wheat _E_ hmgl_wheat.ssp >hmgb_chite _E_ hmgb_chite.ssp >hmgl_trybr3 _E_ hmgl_trybr3.ssp ... Format of the prediction file: >hmgl_wheat CCCCCCCCCCCCHHHHHHHCCCCCCCCCHHHHHHHHHHHHHHHCCCCHHHHHHHHHHHHHHHCE Output of the prediction ^^^^^^^^^^^^^^^^^^^^^^^^ You can output a color coded version of your alignment using the secondary predicted structure or transmembrane regions predictions: :: Secondary structure prediction: $#: t_coffee sample_seq1.fasta -template_file PSISSP -output sec_html Transmembrane regions prediction: $$: t_coffee sample_seq1.fasta -template_file PSITM -output tm_html RNA sequences using 2D and 3D Structure ======================================= The main property of RNA sequences is to have a secondary structure that can be used to guide the alignment. While the default T-Coffee has no special RNA alignment method incorporated in, we have developped specific modes and tools for RNA alignment and analysis (see subsection **Manipulating RNA Sequences** for more details). RNA sequences using 2D structure (R-Coffee) ------------------------------------------- Introduction ^^^^^^^^^^^^ R-Coffee is the special mode of T-Coffee developped to handle specifically RNA sequences. It has been proven far more accurate than T-Coffee default because of its specific design. It can be run as a standalone aligner (using secondary structure prediction) or using third party software. R-Coffee: aligning RNA sequences using secondary structures ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ R-Coffee uses predicted secondary structures via the software RNApfold, in order to improve RNA alignments. Running R-Coffee by default is rather simple (command 1) but as for T-Coffee, you can also specify the methods you prefer (command 2): :: Command 1: R-Coffee default $$: t_coffee sample_rnaseq1.fasta -mode rcoffee Command 2: R-Coffee selected methods $$: t_coffee sample_rnaseq1.fasta -mode rcoffee -method lalign_id_pair,slow_pair Improving R-Coffee ^^^^^^^^^^^^^^^^^^ There are two modes we proposed to improve R-Coffee alignments: 1) using the best method for RNA sequences (namely Consan), 2) combining many methods to achieve a better reliability (RM-Coffee). :: Using Consan (best): $#: t_coffee sample_rnaseq1.fasta -mode rcoffee_consan Using multiple methods: $$: t_coffee sample_rnaseq1.fasta -mode rmcoffee .. tip:: In order to know if a RNA alignment is better than another one, the best is to visualize the compensatory mutations of the secondary structure: look at the subsection **Manipulating RNA Sequences**. RNA Sequences using 2D and 3D structures (SARA-Coffee and RSAP-Coffee) ---------------------------------------------------------------------- SARA-Coffee is a structure based multiple RNA aligner. This is a new algorithm that joins the pairwise RNA structure alignments performed by SARA with the multiple sequence T-Coffee framework. Since setting up the SARA-Coffee dependencies (T-Coffee, `SARA `_, `X3DNA `_, `Numpy `_, `Biopython `_, Perl, Python 2.7) can be tricky we provide a self-contained Vagrant VM, which downloads and configures all the required pieces of software for you. This procedure can be complex, and if you cannot install SARA, we recommand using RSAP-Coffee instead (cf next paragraph). RSAP-Coffee is a an an alternative experimental mode of SARA-Coffee using protein structral aligners. It processes RNA PDB files so that the C3' Carbons are treated as C-alpha in a protein chain. This procedure is suitable for any protein structural aligner that only considers alpha carbons in the main chain. By default, rsapcoffee uses the sap_aligner that ships with T-Coffee default distribution. .. note:: SARA-Coffee and RSAP-Coffee use very similar command line. In practice, the only difference is the use of the method sa_pair instead of the sara_pair methiod when buolding the T-Coffee library. Running SARA/RSAP-Coffee ^^^^^^^^^^^^^^^^^^^^^^^^ The procedure is similar to Expresso/3Dcoffee and requires providing a template file in the PDB files are matched with the corresponding sequences :: > _P_ > _P_ ... .. note:: PBD can be either a file PDB identifier or a PDB file. The file can be either in your homew dir or on a location specified by the environement variable **PBD_DIR**. If you use an identifier T-Coffee will be able to download the file from RCSB. .. note:: If in your PDB the SEQRES and the ATOM yield a different sequence, T-Coffee will use a sequence estimated from ATOM. If tghis sequence does not match your FASTA, they will be reconciliated using an alignment procedure to establish residues equivalencies. You then need to run :: $$: t_coffee t_coffee -seq rna3D.fa -mode rsapcoffee -template_file rna3D.template Or $$: t_coffee t_coffee -seq rna3D.fa -mode saracoffee -template_file rna3D.template By Default, RSAP-Coffee only runs sap, but you can also run other structural methods supported by T-Cofee. like Mustang or TM-Align :: $#: t_coffee -mode rsapcoffee -method sap_pair, mustang_pair, TMalign_pair -template_file