I am making a tree with the whole genome of over 100 samples. They are very similar, with ~95% of sites identical. If I want to make a tree with the whole genome to compare their overall similarity, can I only use sites that are not identical to all? I mean, 95% of sites in the sequences alignment contains only identical bases, which cannot tell the differences among them.
Hi Xiaolong,
I think the short answer to your question is yes, you can.
However, it really depends on what you want to prove/demonstrates/see.
In short, the more similar your genomes are, the less your samples will look different (separated from each other on your tree). This means that if you generate a phylogenetic tree using the full genomes (95% similar to each other) you probably won't see a lot of variation in your tree.
However, if your intent is to understand HOW different these genomes are, then you can select the regions of the genome with higher genetic variability. This will allow you to focus more on that 5% variation and understand what parts of the genomes are different between your samples.
I think that the only risk of focusing solely on the variable regions is to overestimate the genetic variations of your samples. However, considering you are already aware that your samples fall within a 5% genetic variation, you won't risk any overestimation.
Furthermore, if you select only certain area of the genome you will probably be able to run your analysis in a shorter time and/or using less computational power.
If you decide to run a phylogenetic tree based only on certain region, remember to specify what these regions are if you end up making a figure for a work/publication. Since you won't be using the whole genome, the figure won't be a phylogenetic tree of the whole genomes. If I can give you my personal opinion on this, I would probably try to do both: a first phylogenetic tree including the complete genomes and a second figure focusing only on the more variable regions. This way you could explain your rationale.
Hope this helps!
What Francesco Martoni says is correct, the branching order should be the same or very similar in both trees, but the branch lengths will be very different. Most phylogenetic models and methods try to adjust for the problem of multiple mutations in the same site; a site where the nucleotide has changed from A to G to C over time so we observe an A and a C in our data, for example. When the sequences are 99% identical to each other, the odds that one site changed more than once are very low. But when the sequences are more distant the odds of this are much higher. In reality the problem goes up much more quickly in protein coding DNA and in structural RNA genes because of selection pressures to maintain the protein or RNA function, than in DNA under no selection pressure, but this is very difficult to incorporate into phylogenetic models.
One problem with stripping out the identical or uninformative sites, is that it then becomes very difficult to impossible to add a new genome to your data set without repeating the whole process again. And as Fancesco said, it becomes difficult to explain exactly how and why you removed the noninformative sites. For most phylogenetic models and methods the computational power needed to produce a tree goes up exponentially (or more) with number of species or sequences added, but only linearly with number of bases or columns in the alignment, so stripping nonimformative sites does not speed up the computation as much as pruning the tree does. Pruning to remove sequences that are very close to identical to each other, leaving only a representative few sequences from each species or clade, can greatly decrease the computation.
There are programs available both for removing noninformative sites and for pruning alignments based on pairwise percent identity or a tree, so you should not attempt to do such things "by hand".
Thank you all for your answers. I asked some professor. They said, in the phylogenetic building, using a general time-reversible (GTR) substitution model with inverse gamma rates, the program will use the sites that are identical in all samples to generate some data to build the tree. So if we only use sites that are different, it may not be called a phylogenetic tree.
Personally, I think it is possible that the overall spciation structure should be still right, but may be different from the one we use all sites. The difference may be tiny.
STRUCTURE, Treemix all use kind of "SNP" sites, which are sites that are not identical in all samples.
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology