Interpreting bayesian phylogenetic model using Tracer?

Question

I am doing phylodynamic analysis of virus sequences using bayesian method using BEAUti and BEAST. As a part of my analysis I am using Tracer to check how my model is working.

I can interpret some of the results like from the Tracer like :

clock.rate - rate of nucleotide substitution

treemodel.rootheight- time when divergence occured

ESS- measures sampling quality (ideally has to be greater that 200)

But there are other many terms on this analysis which I have no idea how to explain.

Please can anybody help me to find out what other aspects of the evolution can be explained using this table.

posterior -33531.076
prior -481.982
likelihood -33049.094
treeModel.rootHeight 32.271
tmrca(Brazil) 3.853
tmrca(Canada) 16.644
tmrca(China ) 11.752
tmrca(Columbia) 1
tmrca(Thailand) 1.093
tmrca(USA) 32.271
tmrca(allseq) 32.271
constant.popSize 43.595
CP1.kappa 18.716
CP2.kappa 17.16
CP3.kappa 22.796
frequencies1 0.224
frequencies3 0.221
frequencies4 0.26 5001
CP1.mu 0.76 CP2.mu 1.221
CP3.mu 1.019
clock.rate 1.354E-3
meanRate 1.354E-3
CP1.treeLikelihood -9462.258
CP2.treeLikelihood -12520.305
branchRates 0E0
coalescent -467.376

Answer 1

When I have a moment I will fill in the blanks

You must have an RNA virus (nothing else will mutate that fast). The basal group and root is the USA. It estimates the time to that basal event is 32.3 (likely years).

"frequencies" frequency of ACGT where the missing one will be 0.25.

Its a simple model of constant population size.

ESS is the most important parameter to ensure convergence of the Bayesian simulation .

Likelihood is the second most important parameter. I'll edit again shortly.

---

I'll fill in one more blank and summarise ...

Kappa ... This will be the transition transversion ratio in HKY. I suspect it also will be transition transversion ratio in Kimuras 3-parameter model, this is because ... the transversions are further split into two mutation groups, ie. purine to pyrimidine has 2 further combinations and you end up with 2 ratios. that makes sense, transition/transversion (kappa 1), transition/transversion 1 (kappa 2), transition/transversion2 (kappa 3) ....

• CP1.kappa 18.716
• CP2.kappa 17.16
• CP3.kappa 22.796

CP2 and CP3 are either side of CP1 .. which is what you would expect on my interpretation.

These are very high values by the way, essentially there are very few transversions in this data, so ... I dunno. You need to check the data, if you see lots (and lots) of A-G and T-C mutations, then thats what I believe it is saying.

Briefly posterior and prior are the fundamental method of a Bayesian calculations, the output is probability expressed as a log, so the more negative the higher the probability (the prior looks like a random tree, or a reasonable tree with really rubbish parameters, likely the former).

---

Summary Essentially I am going to call it a day here. The real issue is that the output summarise the entire modern theory in phylogenetics, so its a very long response required to address all of the parameters and I honestly don't think it helps your analysis. What I am clear about is:

* You've omitted a gamma distribution (I think Tracer would summarise this) *

So my advice is:

• ESS MUST be homogeneous between runs - big trouble if not
• gamma distribution MUST be run

Beast can be tricky getting gamma to converge via ESS, just a word of warning.

I can't see the full robustness, because its absent... However, from the output, I think its pretty clear, that you (well me) can more or less see the model you ran.