Hi Marine,

If I understand you correctly, the 1500 bp sequences contain the same region as the 700 bp sequences, plus additional sequence?  If that is the case, the simple answer is that, it is not appropriate to use two distinctly different classes of sequence to perform the analysis.  It's like comparing apples and oranges.  If you remove the "extra" sequence from the long ones so that your entire dataset covers only the same region then you could at least look at differences in comparable regions of the gene, rates and nature of sequence changes in that same region of the gene, etc.  I am no expert on this, but you can get excellent advice from Dr. Jessie Kissinger right there at UGa, who is a bona fide expert in these sorts of analyses.

Hi David,

Thank you so much for your fast answer!

Yes, the 1500 bp sequences contain the entire gene, while the 700 bp cover only a portion of the gene.

I guess when I filter the alignment, I can do what you suggest, which is remove the portion of the gene that doesn't contain information for all the sequences. But when I do that, I guess I lose valuable information for all the sequences that did have bases.

Your answer is very helpful, I guess there is no other choice but to keep only 700 bp for my alignment and my tree, even for the sequences that are longer.

Thank you for the tip, I will definitely get in touch with Dr Kissinger!

Marine

Hi Marine,

The answer above is correct..that you should compare sequences for such analysis is they are drastically different in length. In fact one should avoid including such sequences right in the multiple sequence alignment....this may result in ambigous alignment. There are few suggestions

1) Use only sequence of comparable length. If some taxa lack certain region of the gene and other have some excess region...there is no point in comparison... One should compare sites which are present across the samples or some indels or if evidence exist for a large indel to be a result of single event

2) You can trim sequences to a certain region which is present in all the taxa..

3) After alignment...see if 'alignment editing' improves it further

bye

Thanks a lot, Ajay!

It is so frustrating to have to "lose" information to whatever sequence is the smallest.

Would there be a way around it? Like, for instance, make a tree with the full-length sequences, and add the shorter sequences to that tree later on?

Thanks a lot for your insights

Hello Marine,

Your are definitely right that 'losing' biological information is not good. There are some option you may have to make best use of the available information for such analysis.

1) First is the way you have mentioned..add comparable sequence first and the smaller or (even more divergent later). This can be practised even during alignment....After a preliminary view of genetic divergence (and also length differences)..you can define a cut off and above that value the sequences will be added later.

2) You manually can split the alignment into several steps. Take the similar length sequences, align them and then add divergent sequences to this alignment in based on their divergence in a stepwise manner. In this way, the ambiguity die to long gaps can be reduced considerably

3) After a refined alignment file..you can actually use three options for molecular phylogenetic analysis: a) use all the sites across all the sequences for inferring relationships, b) removal of deletions in a pairwise manner to compute genetic divergence and c) remove indels across all sequeneces and compute divergence based on the regions present in all of them.Such options are avilable in several programs...I particularly use MEGA (ver 5/6) for such analysis. This will give you a lot of information to discuss in such regions.

4) As I also mentioned previously that 'the loss of information due to non-inclusion of  smaller sequences' should be viewed and interpreted in a careful manner. Whether the smaller sequences in indeed smaller in certain sequences or the information is not complete (sequence wise) will decide it. In the first case nothing can be done and the deletions can be addressed accordingly, while in the second part the region from those taxa can be sequenced quickly for the analysis.

5) Final thing will be to use multiple algorithms for analysis along with statistical scores to present your results.

Discuss your results with proper emphasis on each aspect of the sequences...it would be informative and will be appreciated well.

best of luck

Hello Marine,

assuming you have a good aligment you can use gblocks (http://molevol.cmima.csic.es/castresana/Gblocks_server.html) to eliminate the poorly aligned positions (mainly gap ped positions).

If you want to use the information of indel regions you can use a different strategy. Try to code the indels and use them as a different partition in a Bayesian Inference software (eg, MrBayes).  One of my students developed sidier, a R package to analyze substitution and indel distances (https://cran.r-project.org/web/packages/sidier/index.html). This package has functions to automatically codify indels.

All the best

Hello Marine,

The suggestions descrbed by the  two experts are right!

All the best,

Zhang

Hello Marine,

In addition to Gblocks you can also use TrimAl (http://trimal.cgenomics.org). These programs will give a high quality alignment that you can then use to build the tree. You can also use these programs in the command-line version if you are working in a computer cluster.

Hope this helps

Best

Amanda

Dear Marine,

I would tend to think that the difference in length is not necessarily the major issue, at least as opposed to sequence similarity for instance. If your sequences are shorter because of biology (the gene got truncated at some point in history), then you might indeed expect different molecular evolution and difficulty in alignment and phylogeny reconstruction. If, however, your sequences are shorter because of sequencing reasons, then the flanking gaps are missing data and not real gaps, and phylogenetic methods should be able to work with your data set (most maximum likelihood and bayesian methods allow to consider gaps as missing data). In case your are using an automatic method for filtering your alignment such as GBlocks, you might end in problems however, as such method will not be able to distringuish real gaps from missing data, unless you recode them as N for instance. If so, you may want to align and filter the two groups of sequences (short and longs) separately and combine them before performing the global phylogenetic analysis.

Hope this helps,

Julien.

Hello Marine,

There have been several great post so far for this question and you are probably bombarded with information right now. There are also plenty of examples out there in the literature where missing data did and did not affect the final typology. For me sometimes it helps to carry out an experiment on my data to truly understand how missing data will affect a tree. For example you can make a phylogeny with the follow three data sets and compare the results.

1) Only samples with complete sequence.

2) All samples cut down to 700bps.

3) All samples with missing data.

If all the topologies are the same then missing data may not be a big deal. If the topologies are different then you get a better understanding of how missing data affects your results and can make a better discussion on what to do next.   

Dear Marine,

sorry for bringing this topic up once more, especially since I do not have much to add to the previous answers: I agree with the previous posters.

However, I would like to suggest a complementary approach that may sidestep some of the issues you describe: have you thought about working with reference trees? The idea would be that you take a (curated) reference 16S tree and map your sequences to it; this does not require a multiple alignment of your sequences, but rather, every sequence is aligned to the closest matches in the reference tree (essentially, you're BLASTing against the tree).

The advantages of this approach are (i) the relative robustness to heterogenous input data (you could even provide sequences of non-overlapping regions); (ii) the "fixed" topology of the reference tree which is (ideally) curated; (iii) the option to transfer further information from (known) leaves in the reference tree.

The disadvantages are (i) reference bias – if your sequences are not well represented in the ref tree, you cannot learn much; (ii) reference dependence – you won't be able to "discover" new leaves on the tree.

There are several software tools available which can map your sequences to a tree of marker genes. I am most familiar with MLTreeMap (attached link), since it was developed in our group. It has a simple-to-use web interface (just paste your sequences, push a button, wait). But there are many other great options available.

Best,

Sebastian

http://mltreemap.org