I've been doing phylogenetics with large (hundreds) 16S rRNA sequences lately.

Usually I'm focusing on one order, and using a combination of trees and sequence similarity to assess stuff like 'is this a genus', 'is this a separate genus', 'is this sequence actually in the order of interest'. It feels like a lot of what I do is repetitive and could be automated more. In particular, it feels like there should be a simpler way for having an assemble view at sequence similarity rather that calculating a Percentage Identity Matrix and then putting it in Excel with conditional formatting. I'm imagining something like this clustering diagram for species (https://www.researchgate.net/publication/360738216/figure/fig4/AS:1182363421806594@1658908815491/The-t-SNE-clustering-of-bacterial-genomes-in-the-code-RA-space-at-different-taxonomic.png), but for sequences instead (using sequence similarity values, at least). Still, I can't seem to find something like that for 16S rRNA that's as visually compelling (or even as visually helpful as my Excel sheet). There are a number of programs for clustering OTUs, but the ones I've encountered don't seem to have a graphical / interactive interface.

Does something like this exist? Thank you for your time!


1 Answer 1


My recommendation is MegaX. It's the closest thing in phylogenetics to an all encompassing GUI.

  1. This has a nice visual alignment editor which highlights conserved, variable and parsimonious sites.
  2. It will produce distance matrices based on any number of criteria, e.g. % homology, Jukes-Cantor, Kimura-2 parameter etc ...
  3. Good for bootstrapping distance-calculations as well as can perform ML based tree building methods with full parameterisation.
  4. Nice tree visualisation, which can be edited, switched to viewing bootstrapping (if performed), or to branch genetic distances (again under any distance criteria, e.g. %age homology etc ..). It does loads of other stuff, which in this case are not needed.

You can export the tree and import into FigTree to produce collapsed clades represented by coloured triangles. In summary for easy GUI stuff MegaX, then FigTree.

There are no PCA/MDA based analysis for 16S sequence data because eiganvalue-based calculations would override lots of theory involved in point mutation modelling and the resulting trees. I could replicated the PCA output in the question with sequence data, its not hard to do - but it's not publishable.

  • 1
    $\begingroup$ Thank you. I'm not sure I understand the last part. (I'm not really familiar with PCA/MDA/eigenvalue) Is it difficult for some reason to represent sequences as blobs, with distances between them based on sequence similarity values? (and potentially different colors based on tresholds)? I don't think that sequence similarity alone is a good idea, but it has provided helpful supporting data to trees in the past (e.g. 'yes, these genera are likely closely related even though support values are abysmal' or 'no, there is no support for these genera being closely related, this is probably LBA') $\endgroup$
    – Laura
    Feb 27, 2023 at 13:59
  • $\begingroup$ @Laura, the 'difficulty' is the diagrams you showed were a PCA analysis (axis were 1+2 components). The problem is not the practicality but the theoretical model of what a 'blob diagram' represents. The theory you are using is modelling point mutation, Jukes-Cantor correction and all that. You can't replace Jukes-Cantor with PCA - completely different. Also you can't do PCA on a distance matrix because there's no multivariates. Sure, I could code a 'blob diagram' but the theory used would be controversial - thats the problem. $\endgroup$
    – M__
    Feb 27, 2023 at 14:07
  • $\begingroup$ Converting a distance matrix to a 'blob' diagram - it's better to use clustering analysis, i.e. make a tree. Clustering algorithms (you normally use Neighbor-joining but it's one amongst many) are the best approach to analysis of a distance matrix. LBA + support ... if could put that as another question might be able to help. $\endgroup$
    – M__
    Feb 27, 2023 at 14:11
  • $\begingroup$ We can chat briefly here chat.stackexchange.com/rooms/58858/bioinformatics but I will simply explain that the model used isn't applicable to blob diagrams $\endgroup$
    – M__
    Feb 27, 2023 at 14:38
  • 1
    $\begingroup$ Thank you. Sorry. I lack the vocabulary. I guess I wasn't really aiming for a PCA, just a graph where I could show that different points (in my case, sequences) are similar by how close they are together. Sort of, sequence A and B and C would be close together due to high similarity values (~97%), and D would be a bit off to the side (but closer to C than A or B). I'm not sure if this makes sense. I'm also confused as to how sequence similarity values necessarily imply an evolutionary model or evolutionary distance (blastn has a scoring matrix of +1match/-3gap, but does that imply J-C? @M__♦ $\endgroup$
    – Laura
    Feb 27, 2023 at 14:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.