The trick with multiple sequence alignments is not "diversity" of the sequences per se but insertions and deletions or gene order rearrangements. If no insertions or deletions happen over time, then the alignment is trivial, no gap characters need to be inserted create the alignment.

With insertions and deletions come several problems for multiple sequence alignment and phylogenetic reconstruction. One is that insertions are often duplications of surrounding sites or "stutters" of the DNA polymerase on a repetitive region of DNA sequence. This makes it impossible to say which bases were "original" and which were inserted in the duplication event. Multiple sequence alignment programs will often place the gap characters at different positions in different sequences that do not contain the insertion.

For phylogenetic analysis, a major problem is that not all insertions/deletions are equal. Just as G -> A transitions are more frequent than C -> transversions for single point mutations, so too are "stutters" of repetitive DNA TTTT -> TTTTT or GAGAGA -> GAGAGAGA more frequent than insertions at other sites. The problem is that there are so many type of insertions and deletions that phylogenetic reconstruction programs do not have settings to deal with them.

In practice, the best methods for dealing with all of this is to remove regions of uncertainty in the alignment. The idea is that the alignment has "signal" in truly homologous columns and "noise" in columns of uncertain ancestry/history. Regions of uncertainty can be removed with a "mask" sequence ++++ characters above the good columns and --- above the columns to be ignored, for example. Many programs can be used to generate the mask and many can read the mask, but not all can. The other method is ti use a tools such as "GapStreese" (http://www.hiv.lanl.gov/content/sequence/GAPSTREEZE/gap.html) or GBLOCKS (http://molevol.cmima.csic.es/castresana/Gblocks/Gblocks_documentation.html) to remove those columns from the alignment prior to feeding the alignment to a phylogenetic program that does not recognize a "mask".

The idea is to throw away "noise" more than "signal" in order to increase the overall signal:noise ratio in the data.

May be you could try cd hit , http://bioinformatics.org/cd-hit/...

The trick with multiple sequence alignments is not "diversity" of the sequences per se but insertions and deletions or gene order rearrangements. If no insertions or deletions happen over time, then the alignment is trivial, no gap characters need to be inserted create the alignment.

With insertions and deletions come several problems for multiple sequence alignment and phylogenetic reconstruction. One is that insertions are often duplications of surrounding sites or "stutters" of the DNA polymerase on a repetitive region of DNA sequence. This makes it impossible to say which bases were "original" and which were inserted in the duplication event. Multiple sequence alignment programs will often place the gap characters at different positions in different sequences that do not contain the insertion.

For phylogenetic analysis, a major problem is that not all insertions/deletions are equal. Just as G -> A transitions are more frequent than C -> transversions for single point mutations, so too are "stutters" of repetitive DNA TTTT -> TTTTT or GAGAGA -> GAGAGAGA more frequent than insertions at other sites. The problem is that there are so many type of insertions and deletions that phylogenetic reconstruction programs do not have settings to deal with them.

In practice, the best methods for dealing with all of this is to remove regions of uncertainty in the alignment. The idea is that the alignment has "signal" in truly homologous columns and "noise" in columns of uncertain ancestry/history. Regions of uncertainty can be removed with a "mask" sequence ++++ characters above the good columns and --- above the columns to be ignored, for example. Many programs can be used to generate the mask and many can read the mask, but not all can. The other method is ti use a tools such as "GapStreese" (http://www.hiv.lanl.gov/content/sequence/GAPSTREEZE/gap.html) or GBLOCKS (http://molevol.cmima.csic.es/castresana/Gblocks/Gblocks_documentation.html) to remove those columns from the alignment prior to feeding the alignment to a phylogenetic program that does not recognize a "mask".

The idea is to throw away "noise" more than "signal" in order to increase the overall signal:noise ratio in the data.

Also, there are textbooks such as:

http://www.kuleuven.be/aidslab/phylogenybook/TOC.html where you can get much more professional help than you are likely to find here.

My short answer is that Muscle has a better theoretical basis than Clustal (I'm not familiar with the other approaches). Clustal is highly sensitive to the order in which you list the sequences, because its search algorithm always depends on the order, especially the first sequence. It is one of the most commonly used alignment programs, but not because of any theoretical or empirical justification. Clustal has been around for a long time and is fairly quick, so people tend to use it.

The underlying problem with alignment is that it depends on an implicit or explicit assumption that we know the phylogeny from which the sequences originated. However, because we have no way of knowing the 'true' phylogeny - that is often our research question in the first place - some alignment programs take an iterative approach. The advantage of an iterative approach is that it can search through a variety of alignments and phylogenies, make changes to them, test if it's an improvement or not, and continue onward until it has a more optimal solution.

Muscle takes an iterative approach, which is a plus. However, I don't know whether its search strategy does a better job than Geneious or Translation. My guess is that you are getting different answers because 1) some approaches are more effective than others (sorry I can't give you an answer on that), and 2) that the default cost (or range of costs) for insertions, deletions, substitutions, etc. differ. What you're going to have to do is decide which parameters are most reasonable and which search strategy is the most justified. Hope that helps!

Thanks for the answers.

Mysore, I think you misunderstood me; the sequences I am aligning to one another are all of roughly equal length, so I don't have to cut much I don't think. But I will try Mafft, as I have not yet implemented it.

Brian, my concern with the diversity of the sequences is the level of divergence between them; I am worried about signal saturation. I will definitely look into the masking technique; it sounds good!

Karl, I have noticed that the Muscle alignments tend to look better than the alternatives, especially translation (which, I admit, I think I am poorly implementing). I am pleased to hear that it is more theoretically reliable than ClustalW.

In my opinion MUSCLE is really good (specially for me is the best). I recommend you to do the alignment in MUSCLE and after that edit the sequences aligned in BioEdit to delete in both side the mismatches due mainly to the primers and allow the sequence to start and end in the same position without mismatches (intermediary mismatches are ok). Further more, you can establish the relationships between species with WinClada software, which is one of the most common phylogenetic program used. You can free download WinClada from internet (http://www.cladistics.com/wincDownload.htm); good luck in your future work,

Hello Mark,

I am also using the Geneious program for my work (conserved prokaryotic genomes) from single genes to whole genomes, and found that MAFFT or MAUVE were most suitable for the WGA, whilst MUSCLE is far superior to CLUSTAL for short sequences or single genes.

For my WGAs - MAFFT and MAUVE gave me almost identical alignments, and I chose MAFFT as it was faster and easier for me to manipulate the sequences post-alignment.

Keeping in mind that the sequences I am working with are highly conserved, regardless of which algorithm I used to produce the alignment, manual curation of the alignment was needed every time. I could not find a single algorithm or parameter combination that could reliably deal with fragmented genes, fragmented genes with gaps or deleted genes with remnants of sequence, at least not as reliably as I could align these regions with a quick BLAST search and some post-alignment manipulation.

As your sequences are likely to be divergent - be prepared to spend some time doing manual alignment - especially for larger sequences.

Hope this helps.

To test for saturation with mutations, I like to use DAMBE which allows you to plot pairwise transitions and transversions vs pairwise distances. For all organisms including prokaryotes and viruses, transitions outnumber transversions for several biochemical and biological reasons. But there are twice as many possible transversions as transitions. G A and C T are transitions; A C, A T, G C and G T are transversions. So at evolutionary distances less than about .3 mutations per site transitions outnumber transversions, but after full saturation with mutations transversions will outnumber transisions. I have never seen a data set where the ratio of transversions to transitions got close to the theoretical level of 2.0 because sequences that are far beyond saturation become impossible to align in any meaningful way. DAMBE has MANY other useful functions too, such as allowing the user to quickly partition data in to first second and third codon positions, and build a tree based on each of these partitions or test for saturation in each separately.

http://openi.nlm.nih.gov/detailedresult.php?img=2515634_pone.0003018.g001&query=the&fields=all&favor=none&it=none&sub=none&uniq=0&sp=none&req=4&simCollection=1310896_gki995f2&npos=35&prt=3 is an example of a Ts and Tv vs phylogenetic distance plot for data that is not saturated. All distances are less than 0.1 so on average less than one mutation per 10 bases between any pair of sequences in the set.

If you are going to construct a phylogenetic tree than use MEGA5, it will give you all the features like allignment through ClustalW, clustalW (codons), Muscle, Muscle (codons) and Phylogenetic tree using Maximum LIkehood, Neighbour-Joining, Maximum evolution, UPGMA, Maximum parsimony ect. You can also perform many other tasks for yor sequences. It's works really very good.

There is no simple way to tell alternative MSAs appart. If you are working with proteins, you may be in a better situation because you may have structures. Even one or two structures can be enough to estimate the relative merits of two alternative MSAs in which these structures may be embeded. We make two such methods available on the T-Coffee web site:

iRMSD (two structures or more): http://tcoffee.crg.cat/apps/tcoffee/do:irmsd

STRIKE (one structure only): http://tcoffee.crg.cat/apps/tcoffee/do:strike

If you have no structure, or if you are working with DNA, then the only option you have is to compare your MSAs and possibly combine them into one consensus MSA. It will not necessarily be more accurate but you may use the output to identify regions where your alternative MSAs are in good agreement. You can use the mcoffee server to run you favorite aligners and combine them into one model with a display of local agreement

http://tcoffee.crg.cat/apps/tcoffee/do:mcoffee

(see the advanced mode to select your own methods)

If you want to combine your own MSAs, this is better done from the command line:

t_coffee -seq -msa , -output score_ascii, score_html, aln

You could try VectorFriends at http://www.vectorfriends.com

It supports Muscle and Clustalw and is free for academic researchers

I have tried many different MSA tools. For a good amount of time I stuck with MUSCLE, which is great if what you need is just to get some kind of an alignment, and get it fast. I then moved to MAFFT, which I wholehearteadly recommend. It is more sophisticated that MUSCLE, has several different alignement algorithms to choose from (for different types of data) and has many useful extra options to work with alignments and probably most importantly - it is still actively maintained by it's author, Kazutaka Katoh.

In a final word though, I second the concern of many other people in this thread concerning indels in MSAs. There is a really nice aligner called PRANK (from Nick Goldman's group) which is phylogeny-aware and quite uniquely, takes extra care to properly treat the indel events. The downside is that it runs quite slow and IIRC has no multi-core support, but it's still worth checking out, I think.