I am currently studying the population structures of several tree species, and I use DNA sequences in my study. One part of my study involves constructing Neighbor-Joining trees to visualise the population structuring within each of the species. I looked through some articles, some build their NJ trees using population pairwise Fst data, while others build their NJ trees using Nei's Da distance. Someone once commented that it is better to use Nei's Da genetic distance to construct NJ trees as it Da distance represents "raw distance" while Fst doesn't (meaning Da distance is a more accurate measure of genetic distance), which makes sense. However, in my case, NJ trees built with Fst consistently show more reasonable population structuring patterns based on my hypothesis, than NJ trees build using Da distances. Apart from that, I also used STRUCTURE to visualise the population clustering, and the population clusters estimated using STRUCTURE agrees with the NJ trees built using population pairwise Fst matrices.
It would be great if someone could help me in this, and probably suggest some articles that explain such situations.
According to Nei (http://www.genetics.org/content/178/1/385.full.pdf) Da has better chance for obtaining correct topology of trees comparing to Fst. But, when your population/loci numbers are far from "optimal", you better look at real structure, and chose that one that you like more...
According to Nei (http://www.genetics.org/content/178/1/385.full.pdf) Da has better chance for obtaining correct topology of trees comparing to Fst. But, when your population/loci numbers are far from "optimal", you better look at real structure, and chose that one that you like more...
Thank you Dr Ignatov for the insight. If I may add some other resources:
1. http://evol.mcmaster.ca/~brian/evoldir/Answers/Tree.reliability.answers
2. http://www.biomedcentral.com/1471-2156/9/37/
No. 2 above is an article that used Fst to build NJ trees just as a representation of the relationship among populations. For me it makes sense, just as a way to view the relationship among populations from the pairwise Fst values estimated. This is because sometimes there are too many populations (hence population pairs), and a tree is just an easier way to view the results. However, some researchers do not like the idea as they prefer a tree from Da values. Da values are more accurate, yes, but I think it deviates from the initial purpose of even drawing that tree, i.e. to present the results of pairwise Fst in terms of an image. Recently I have even got comments that they do not even think using a tree would be appropriate (be it Da or Fst). So I guess it becomes a preference now.
My opinion could be wrong, so it would be great if someone could offer some more insight. Thank you!
Hi, Dr Ng, i think i am being in the same trouble with you.
Now, i am working on the speciation history of some wild rice using genomic SNP data, and i want to know the relationship between populations. However, the NJ trees built with Fst and Da matrices were totally different, and the topology of Fst tree was more consistent with my hypothesis and prior knowledge.
I do not know how to interpret the conflict between different distance matrices. Have you get more explanations for such situations? Thank you!!!
Hi Zhe Cai,
I am sorry I still do not have a conclusive answer to this situation.
But to me, Fst is after all a relative measure of genetic differentiation given the total genetic variation present in the population (throughout your sampling range), it should be more able to tell the relationship among the subpopulations you have in your study; while Da is a direct measure of genetic distance between any 2 populations without regard to the total genetic variation observed, so the resulted tree might not reflect as accurately the relationship among the subpopulations in your study.
I hope I got all my concepts right in the above. Do get a 2nd opinion about this, and let us know here. Thanks!
Wei,
there is a good comparison of data analysis employing Fst and Da matrices, but the authors used both to confirm the same idea.
http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0095436
Wei,
Quality of phylogenetic tree depends on 2 main factors (besides the quality of row data): 1) chosen genetic distance metrics;
2) algorithm of the tree construction (NJ, ME, MPP, UPGMA, etc.)
NJ is very sensitive to type of genetic distance. Have you ever tried to use other methods of grouping?
hi, Wei and Alex, thanks for your opinions.
I have tried other distance metrics (Nei's standard genetic distance, Cavalli-Sforza chord distance, Reynolds, Weir, and Cockerham's genetic distance) and different tree algorithm, NJ, ML, UPGMA. Only the NJ tree using Fst distance consist with my hypothesis.
I am considering using tools to simulate some population SNP data based my hypothesis and construct tree with different distance and tree algorithm from these simulated data. Maybe the simulation could tell us which distance metrics is better or suitable.
Thanks Alex and Zhe for the input!
Zhe: It would be great to know, maybe you could publish your findings on these comparisons, and share them here with us? All the best!
- Badges
- Science topic
- Similar topics
- Cell Biology
- Methods