Why Would Phylogeneticists Ignore Computerized Sequence Alignment?

In phylogenetic analyses, multiple sequence alignment often seems to be the poor cousin to tree building. It has long been recognized that both primary homology assessment (alignment) and secondary homology assessment (tree building) can have important effects on phylogenetic analyses (Morrison and Ellis 1997), and this has been repeatedly demonstrated for both empirical and simulated data (see references cited by Morrison 2008; also Wong et al. 2008). However, most theoretical contributions to phylogenetic analysis continue to involve tree building alone. Indeed, even most review articles continue to treat alignment as being about automated bioinformatics procedures (Wallace et al. 2005; Edgar and Batzoglou 2006; Kumar and Filipski 2007; Notredame 2007) rather than about phylogenetics (Morrison 2006; Phillips 2006). This suggests the possibility that there have been no (or few) theoretical contributions to multiple sequence alignment that practitioners considered to be useful in phylogenetics. In other words, most phylogeneticists are prepared to do tree building in a fully computerized manner but not (yet) alignment. Here, I test this hypothesis by reviewing the current practices for multiple sequence alignment in published phylogenetic analyses, particularly with respect to differences between disciplines in the way that phylogenetic analyses are used. I also provide suggestions as to why phylogeneticists are apparently dissatisfied with computerized sequence alignment and how we might deal with it. My intention is to highlight why phylogeneticists are dissatisfied with similarity-based alignment procedures and what specifically they are doing about it in practice. I conclude that there is currently no bioinformatics approach that is acceptable for phylogeneticists and that there is thus a gaping hole that needs to be filled. In order to quantify the current state of affairs in a range of biological journals, I reviewed each paper published in 2007 containing an original multiple sequence alignment that was based on empirical data and was used in a phylogenetic context. This involved 1280 papers in 26 biology journals (Table 1), although nearly 8000 papers were screened. A summary of the data for each journal is available as supplementary material at http://www.oxfordjournals.org/our_journals/sysbio/. Characteristics of the multiple alignment procedures reported in 1280 papers published in 26 biology journals during 2007. The percentages do not sum to 100 because any one paper may have included several alignment techniques (e.g., for different genes) Combined alignment + tree building, including POY and BaliPhy. Excluding the simple statement that gaps were deleted before tree building. Explicit use of a previous alignment, including those from the ARB project, the Ribosomal Database Project, and the European rRNA Database. No criterion specified. Explicitly stated criteria included the use of SOAP, adjustment to match RNA structure/codons/motifs, and minimizing the total number of mutational changes/polymorphisms/variable columns. Usually occurred for sequences with high similarity, such as within-species comparisons, and chloroplast sequences. Explicitly stated criteria included match to RNA structure/codons/motifs and recognition of events that cause sequence variation (e.g., repeats, inversions). Molecular Phylogenetics and Evolution. Arthropod Systematics and Phylogeny, Australian Systematic Botany, Biochemical Systematics and Ecology, Cladistics, Insect Systematics and Evolution, Invertebrate Systematics, Journal of Zoological Systematics and Evolutionary Research, Mycotaxon, Plant Systematics and Evolution, Protist, Systematic Biology, Systematic Botany, Systematic Entomology, Systematic Parasitology, Systematics and Biodiversity, and Taxon. American Naturalist, BMC Evolutionary Biology, Evolution, Evolution and Development, and Journal of Evolutionary Biology. Molecular Biology and Evolution and Journal of Molecular Evolution. International Journal of Systematic and Evolutionary Microbiology (first 6 months only) and Systematic and Applied Microbiology. Muscle was a close third. Characteristics of the multiple alignment procedures reported in 1280 papers published in 26 biology journals during 2007. The percentages do not sum to 100 because any one paper may have included several alignment techniques (e.g., for different genes) Combined alignment + tree building, including POY and BaliPhy. Excluding the simple statement that gaps were deleted before tree building. Explicit use of a previous alignment, including those from the ARB project, the Ribosomal Database Project, and the European rRNA Database. No criterion specified. Explicitly stated criteria included the use of SOAP, adjustment to match RNA structure/codons/motifs, and minimizing the total number of mutational changes/polymorphisms/variable columns. Usually occurred for sequences with high similarity, such as within-species comparisons, and chloroplast sequences. Explicitly stated criteria included match to RNA structure/codons/motifs and recognition of events that cause sequence variation (e.g., repeats, inversions). Molecular Phylogenetics and Evolution. Arthropod Systematics and Phylogeny, Australian Systematic Botany, Biochemical Systematics and Ecology, Cladistics, Insect Systematics and Evolution, Invertebrate Systematics, Journal of Zoological Systematics and Evolutionary Research, Mycotaxon, Plant Systematics and Evolution, Protist, Systematic Biology, Systematic Botany, Systematic Entomology, Systematic Parasitology, Systematics and Biodiversity, and Taxon. American Naturalist, BMC Evolutionary Biology, Evolution, Evolution and Development, and Journal of Evolutionary Biology. Molecular Biology and Evolution and Journal of Molecular Evolution. International Journal of Systematic and Evolutionary Microbiology (first 6 months only) and Systematic and Applied Microbiology. Muscle was a close third. I explicitly grouped the journals into 5 disciplines based on the general purpose of the phylogenetic analyses within the context of the published papers. That is, the classification refers to the journals rather than to the individual papers—it would be impractical to try to classify each paper separately. There is thus likely to be considerable overlap between the categories, especially the first 2. However, it seems unlikely that the choice of journal is arbitrary among authors, and so the categories are meaningful in a general sense. The labels for the 5 disciplines should be seen as indicative of the general fields of the journals rather than as formal definitions, as clearly some of the papers in any one journal could fit into one or more of the disciplines. The “general phylogenetics” papers focused on the phylogeny of whole organisms, with the phylogeny being presented as the primary result—there was no necessary use of the phylogeny for any subsequent purpose. The “systematics” journals also focused on the phylogeny of whole organisms, although the phylogeny may have been a means to a specific end (i.e., classification). The “evolution” journals usually focused on whole organisms, but the phylogeny was only one tool among many for studying evolutionary patterns; indeed, the majority of the papers in the evolution journals had no phylogenies at all and those that did do so usually used a previously published one. The “molecular biology” journals usually focused on gene phylogenies, with the phylogeny being used as a tool for studying molecular evolution. The “microbiology” journals have guidelines stating that a phylogenetic context is required for the publication of new taxa (e.g., Christensen et al. 2007), so that their organismal phylogenies are part of a “publication pipeline.” The discipline of “evolutionary ecology” was ex- cluded because most of the published studies involved intraspecies comparisons, where the high sequence similarity precludes the need for automated alignment; “genetics” was excluded because there were few original phylogenetic analyses; and “genomics” was excluded because the procedures for whole-genome alignment are usually quite different from those for gene alignments. For the purposes of comparison, I have treated the percentages of papers as being “significantly different between disciplines” if the respective 95% score binomial confidence intervals (Newcombe 1998) do not overlap the other means. When the disciplines differed with respect to alignment procedures, it is clear (Table 1) that they formed 3 groups: general + systematics, evolution + molecular biology, and microbiology. For example, wholly automated alignments appeared in 75% of the microbiology papers, 50–55% of the evolution and molecular biology papers, and only 20–25% of the general and systematics papers. As I think most readers will be expecting, the Clustal program dominated the automated part of most of the alignment procedures, being cited in from one-half to two-thirds of the papers (Table 1). What is perhaps more surprising is the very to the of where there were many but no clear It has been several in the few (see that and are the of the based on sequence similarity, by and However, this not to have phylogenetic as and to of the papers (Table 1). There thus seems to be a if Clustal is being similarity is being as a criterion for alignment it will often not and the similarity-based are not being used. The choice of Clustal may thus rather than any of For example, and the of review may be in the that it may be to a paper published if procedures match all previous papers. in and are to at the of many of the differences between 5 categories, in the that phylogeneticists are likely to be more in some disciplines than in in phylogenetic analysis (or of by both and is also likely to be It has also been that alignment and tree building should be into a thus primary and secondary homology this is a thus and so it is unlikely to be by those for phylogeny is but one tool among use has thus been so (Table 1). Indeed, the most used of is and it is but not to papers in the journal The use of a previous alignment as the for any subsequent alignment that the sequences has been alignment (Morrison 2006). This is an to the an alignment is This was explicitly stated to have been used by only a few practitioners (Table 1), although it may have been an part of many of the alignments. This is particularly surprising in the of as there are + and that were specifically to provide RNA alignments for to use as part of their publication This may with the of the et al. and the et al. both of are specifically to provide sequence for in the context of a phylogenetic alignment. In of automated alignment procedures, perhaps the most that I repeatedly were that sequences were any as to this was by or by Clustal as a and that sequences were of automated procedures was used. use of a program not of the at as as published are et al. the most that I than was that the alignment was the phylogenetic tree was the of of the that the most important in is that in both the general and the systematics categories the majority of the practitioners and explicitly stated that their alignment involved by the from a program or by the alignment from the The number of such practitioners was for the evolution and molecular biology and even for microbiology Molecular were as as general phylogeneticists and to their alignments in all 3 but to and tree building is a wholly computerized alignment is often perhaps on the context. That context may of many For example, alignment is to with evolutionary among the sequences. It is also that the of the data may to an alignment Wong et al. 2008). the most important context is likely to be the purpose to the tree is the tree is one of a project, it might individual For any procedures, there is an important between those where and are and those where there is an explicitly stated This is important because the with data analyses any is that they may be in any of the that is to this I that we do not that a will to the but we do that will to very data most that they will to the sequence alignments will be in this there is an criterion for the as by et al. the between a alignment (or based on a criterion and one where the criterion is not is that the is in the that an to the is likely to to the end That is, it is the majority of the in multiple sequence alignment was not based on any criterion (Table 1), of often because the were to any may an to to in each and paper rather than of specific It may also a of that alignment procedures can have a on the or perhaps an that subsequent of the data (e.g., or SOAP, were will deal with any that have phylogeneticists were more likely to a criterion than were in the other by and molecular and I this as that phylogeneticists are the most that 1) the from similarity-based alignment is for their purposes the and there is currently no program available for in the first the This may their in phylogenetic may not be by many of the use phylogenetic analysis as one of many are more likely to be of the of their than are This is likely to to and review as only a of papers the of several alignment Indeed, even the several there have been few papers where the have a of doing so (e.g., and et al. 2006; and 2006; and This very with analyses, where it is to and by more than one This is a by and recognized the of alignments and and variation in In it is that molecular phylogenetics is dominated by the of systematics, not of molecular biology (i.e., the alignment procedures in the general match those of systematics rather than those of molecular This to to the of that systematics has been to the phylogenetics demonstrated that the in is a I to some will be within the of a This is not because I have any in but because I think there is to from that and the (or of of phylogeneticists to it. In what to be the first published of variation sequences the a very considered multiple alignment based on recognition of and RNA within the This was very in with to the assessment of to phylogenetic tree building, for (Morrison 2006). computerized procedures were based on for alignment based on similarity and based on the alignment for multiple alignment and the Clustal program was procedures in a manner for (see et al. Clustal has that this approach has to computerized multiple alignments in all sequence analyses, not in phylogenetics. There are important to be about this part of the The first is that although similarity-based alignments to sequence they have not so in phylogenetics. It is a of phylogenetics that similarity and so there is no necessary that a based on similarity will an alignment that of homology between the The of to Clustal alignments that the program of multiple alignments that are considered to be for phylogenetic may provide a for alignment, but that is The to be is that sequence alignment in phylogenetics by explicitly molecular that might to sequence variation and an alignment that those The of computerized alignment procedures this not of but because there can be no to because homology is based on on the other a because it can be in many it seems that many have not the because most theoretical papers on multiple alignment with a homology and that by similarity alone. I that the of alignments is a of the that phylogeneticists do that similarity-based alignments are not likely to be homology alignments. I will to this the for the similarity alignment is (e.g., the subsequent of sequence alignment has been based on recognition of the that homology of is not necessary for many studies of molecular evolution. of this has involved the of that may have to to phylogenetic practice. I will with different alignment Morrison 1) sequence comparison, and phylogenetics. The of is to the secondary and of the gene from a gene This of they are or one based on a sequence and one on an are in and the that alignments are not the as homology alignments because may between during The has thus been to for based on not on homology and the subsequent multiple alignments this (e.g., and 2008; et al. 2008; et al. 2008). of alignments to with new alignments (see Morrison 2006; also et al. of an alignment of the gene for of and 3 The are by and the by The of the are The first a in for the at in because the in the of the the is involved in for a different at and of the with at and the homology alignment not match the alignment, as the alignment would the in and with the in The (first and The at (or is involved in both repeats, so that it is not part of an of the for of and a original data are from and The first a by by of a by The of the a in the for can in the the but not This is in the alignment to the of the and to the of the The of by and by The for homology in a and each involve and so no choice between can be there is some for the different events The of alignment for is to the between and sequences. can be any necessary of among the for this purpose have involved multiple alignments based on sequences et al. 1997), do not need to be but sequences that a of variation among the sequences. have been for multiple sequences based on such as and et al. 2007), and many of to with alignments. the use of sequences not involve homology of the but the alignments can be (e.g., and 2007; and 2008). The of alignment for sequence is to sequence (e.g., may between during as by an in and alignments are not the as homology alignments. many are being for of sequence (e.g., all the multiple alignment used in phylogenetics were for sequence This that in this might be to phylogenetic analyses, at in practice. There have been 3 1) the of the of different alignment (e.g., and and has sum of as the criterion for alignment (e.g., an and a and the and more (e.g., Muscle and can have the of similarity-based multiple sequence alignment (see references cited by Morrison 2006; also Notredame However, few of to have phylogenetic practice. The in all 3 and so that it is and more than Clustal and current use in phylogenetics is on of was the most cited of the but it often did not involve any automated sequences were based on (i.e., their but this was as likely to be adjustment of a alignment as an automated alignment of a of sequences. all alignments of and sequences were alignments based on were used in where the ARB was the most cited one. this a that the current of alignments are not particularly the and the to provide used rRNA alignments for all This may provide some to It is also clear that few published alignments are available in a that their use for (e.g., and fully although some do it seems to that phylogeneticists are in automated alignment procedures, particularly those primary sequence The for this than of on the part of is that in sequence similarity do not phylogenetic analyses because of the alignment will usually That is, similarity not homology Indeed, of alignments is so in phylogenetics that I even one paper where the it to why the computerized alignment was not that the of phylogenetic analysis is to of evolutionary homology among the should be only if they have from a homology is thus the in the of a phylogenetic alignment of the that I for it is often that can only be recognized on a phylogenetic where they as it seems that phylogeneticists might be in procedures that primary and secondary homology This a analysis rather than of an alignment and a There are to a phylogenetic analysis (see Morrison an based on alignment the POY and a that alignment and tree building by explicitly including a of with a of alignment, or There are to 2007; Morrison with the important of for in alignment et al. 2008). However, of theoretical are likely to have on phylogenetic practice. In the will be the alignments alignment in the first and an alignment in the match what phylogeneticists In both there are several based on simulated data and 2007) and empirical data et al. 2007; Morrison that currently to be Indeed, are among the multiple alignment 2008). There seems to be one for both that currently the on they are based are The into the alignment by both can currently be based only on or on of For example, alignment can or of use or or use and alignment can the or the are because many of the molecular that sequence variation on multiple and are because can on formed by many of A simple is in This being we alignment and tree building to in of any theoretical that the may be seen to are that more match molecular (i.e., sequence of arbitrary are unlikely to Indeed, a that I more than was that alignment was used because of be with and a can be about the being In it seems that we computerized alignment procedures to multiple alignments that are for phylogenetic procedures, no how they the that similarity and alignment with tree building has not for sequence There is an that sequences are to deal with than because sequences of only of However, it is to that this homology assessment not The or the to be in an alignment, but the are the alignment and the are the in those columns. an A in one is not the as an A in any other are of different one A in (i.e., As the so there is any that can this This is often not for where there may be clear for example, between and or and need to the molecular of the criteria for primary homology of such as similarity, and The for phylogeneticists is to to their by Indeed, I that many of have those is why alignment is so There is one that I have not so and it is the most important of the current computerized explicitly or the that and involve of That is, all sequence variation is to be a from the of all sequence However, this is usually in and it is the of this that is the cause of all the in phylogenetic sequence alignment. The of this simple has been recognized by and 2007; et al. 2008). A sequence is not an arbitrary of sequence alignment is many but is a with specific biological variation is by a number of and molecular such a particularly and of a by of between of a and at of a and of events will cause variation in the sequences (e.g., and and some will not (e.g., and inversions). In all the currently available variation is to and all other variation to to be an sequence variation is clearly to the in the most in are the most cause of sequence and 2007; and 2007) but also in other of as as sequence alignment should involve the most of events based on recognized rather than that all the variation at The sequences are so that they those events (Morrison 2008). That is, the events should be as the alignment rather than being the alignment is events are used in the alignment rather than being a from it. In the alignment should molecular rather than similarity-based it should be to the alignment with all the events and 2005; and 2005; et al. 2007; et al. 2007), thus the whole is likely to be a more important criterion for alignments than is computerized I that this is what alignment is all In the most cited paper this was that of although several papers (see references cited by Morrison 2008; also et al. papers provide a of guidelines for sequences based on explicitly the molecular events that have to the sequence can be by many different of and need to be as the alignment might be different for different events (see the in and the molecular are likely to with different in different sequence for example, are very in both and are most reported in sequences and This means that it is likely to be to a similarity-based to sequences that for and also sequences. there are to alignment as sequence the is that it is and there is currently no computerized for alignments of this No program this because of are explicitly to do In events all the sequences (e.g., for involved in is that alignment do those based on and This is where there is a bioinformatics hole that needs to be filled. It is to an based on this For example, that could be used to all the events in a of based on sequences (e.g., for and and of sequences (e.g., for and would a of alignments and 2006; et al. 2006). this it would be to the most based on for the different to the alignments into a alignment. an alignment would be with respect to the molecular that to sequence variation rather than with respect to However, the of all of events is likely to be an impractical A perhaps more approach would be to a alignment as a for the of events and this alignment to explicitly the This would not but it is likely to a This approach could be the of a new or it could be automated by to with each other so that the can be for phylogenetic purposes is likely to be an as is from different but there is currently no way to and that I that this is the of the current of of similarity-based at those based on In the is one of the current and Indeed, Morrison has this in a and to the that the similarity-based alignments by and are among the for has been a in although all alignment use as and the there is no as to how to the molecular but it is clearly necessary to do so if a alignment is to simple are in and if are among may a of such as how to treat of Indeed, all alignment such as can be by the or of and of an alignment of the for and several for homology that can be only if there are for events original data are from and The first The a by a The a by an The alignments to the and are No choice can be between the 3 there is some for the different events although no current the it is to do and this would the current need for and multiple alignments for use in phylogenetic analyses in a manner that should be acceptable to all The that I from is that sequence alignment for phylogenetic purposes is at the a bioinformatics is a biological more needs to be to what are the biological criteria for a phylogenetic alignment. can we for a for sequence alignments. have not with an that can and there is for those to in an automated alignment In the the to alignments and is part of the of all alignments and alignments may currently provide for a phylogenetic alignment, but that is material can be at http://www.oxfordjournals.org/our_journals/sysbio/. to and for the to at the Systematics and to and the for