international online community for dna barcoding professionals
I am working on DNA barcoding of Dalbergia in India. Currently, I am analysing the different loci using MEGA 5. My question is while constructing NJ tree which option should I select, uncorrected p-distance or kimura-2-parameter? I have seen papers using either of them and I also read that there is no considerable difference in output. Can anybody help me to understand both the parameters correctly?
Thanks in anticipation
Hmmmm.....good question. Please tell me you solved this problem and what did you conclude?
I am clear on when to use Maximam likelihood, NJ and Minimum evolution options but I am not sure about Kimura 2. Care to share the light?
In this case, I will suggest you to use kimura-2 parameter. These are the substitution models (at least for kimura, I know for sure) which differs for different sequences. It is possible to choose best model for your multiple sequence alignment in MEGA. Just check, model test/selection something like that in analysis or phylogeny menu. It will test all models and rank them. The best model for your MSA is the top one which will pop-up once you run the model test.
Hey thanks for your reply.
I tried the suggestion given by you. I found a best model for my data, but that option is not there in model/selection when I construct NJ or ML. So is it the case that I have to come down as 1st model if not then second then third etc..
NJ algorithm, you can run using JTT (for DNA as well as protein sequences) or PAM (Dayhoff) model for protein sequences. For ML, you should chose best model. This is a standard protocol. So first decide which is the best model for you alignment using model test option. Then you run ML algorithm with these parameters (Model and Sites Rate etc). It is better to go through some online tutorials on these methods or MEGA documentation. application of these methods on your data is relatively easy since many programs like MEGA are available but you should know about them well before actual use.
I am interested in hearing what you finally came up with and how you went about it.
For me, First I construct different types of trees using the options available to me then cross them off one by one. I never use NJ especially when its a phylogenetic tree (as opposed to a dendrogram) that I am constructing just because I have never find it robust enough-in the sense that it gives fewer options for analysis, in addition to using the fewest steps to arriving at the most convenient tree. Therefore that leaves me with ML and Min Evol. I try these with diff param.
ML is good in that the options seem almost endless! I think the software I use is called JAR, to select the best param for the dataset. "Worst" case scenario- Bayesian!