A long time ago, I used BEAST (Vers. 1.4.7) to estimate divergence times in two groups of snakes for my thesis. This is what I had in a draft for my thesis. I don't know if it is useful:

In order to calculate historic demographic parameters it is first necessary to determine the appropriate model of sequence evolution. Initially, MEGA was used to assess substitution pattern homogeneity by way of the Disparity Index and Tajima’s relative rate test, as well as transition/transversion bias. DnaSP was used to assess overall nucleotide diversity at synonymous and non-synonymous sites. The FindModel website (http://hcv.lanl.gov) was then used to determine the substitution model and parameters for the 12 L. triangulum sequences in the L595 contig and the 21 M. fulvius sequences in the M101 contig. Both of these contigs were chosen to mimimize the amount of missing data, which could potentially confound analyses by nullifying sites where nucleotides were missing and altering distribution frequencies of polymorphic sites.

For the Lampropeltis sequences, no significant difference was found in substitution rates between sequences and a strict molecular clock could not be falsified (c2=2.00, p = ns). All mutations (13) occurred at synonymous sites (p=0.03318). In addition, the overall transition/transversion bias (R) was found to be 301.786. The data was found to fit a General Time Reversible (GTR) substitution model with a gamma distribution (K=4, a= 0.49669). The nucleotide substitution rate parameters were aÞc=0.285268, aÞg=0.705436, aÞt=0.113178, cÞg=0.177409 and gÞt=0.058868. For the Micrurus sequences, Tajima’s Test falsified a strict molecular clock (c2=6.00, p = 0.01431). The Hasegawa-Kishino-Yano (HKY) substitution model was assigned for these sequences.

BEAST was used to calculate times to the most recent common ancestors for both of the contigs, using the parameters calculated above, and the substitution rate of 1.79%/my. The MCMC chain was set at 10,000,000 steps, allowing for a BurnIn of 10,000 steps. For Lampropeltis, besides calculating the TMRCA for the entire sequence set, calculations were performed for the nodes containing the Aiken, SC specimens and the divergent Florida grouping, as determined through phylogenetic analyses. For Micrurus, parameters were as determined above, however, with the strict molecular clock in effect.

For Lampropeltis, the mean TMRCA for all specimens was approximately 331,000 years (median= 318,000 years; 95% HPD lower = 146,000 years; 95% HPD upper = 536,000 years). For the grouping of 4 northern highland and central Atlantic specimens, the mean TMRCA was 83,920 years (73,910 years; 14,130 years; 177,000 years). The Aiken grouping diverged from a common ancestor some 42,860 years ago (36,960 years; 5,045 years; 92,940 years).

Micrurus fulvius diverged some 2.121 million years ago (median=2.041 my; 95% HPD lower = 0.942 my; 95% HPD upper = 3.439 my). A node representing the separation of populations east and west of the Mississippi River may be dated at 1.394 million years (1.328 my; 0.53 my; 2.372 my). Another node representing a divergence within the Mississippi River-east population and giving rise to the widespread haplotype observed in the haplotype tree in Figure 4 occurred perhaps 0.358 mya (0.332 my; 0.097 my; 0.665 my).

Michael Friedman

After you run the xml file, open the log.txt file in Tracer, check convergence using the ESS values, check the treeModel.rootHeight for the mean and the 95% HPD interval and subtract the age of the youngest tip from these values to get the tMRCA estimate for the root. For other monophyletic clades, you should pre-determine these in your xml file, and following beast analysis, you will have tMRCA estimates for each clade in the log file

Good luck

I also recall having to use fossil data to calibrate my divergence times and establish confidence intervals. But, that was a decade ago...

Greetings Pratanu,

From my experience (BEAST 1.8.4), working with fasta files won´t do. BEAUTi asks you to import your alignment in a nexus file format and you can easily do the transformation with several web tools or free software.

Now, when you´ve loaded your data you will see it listed in the partitions section (as in your attached screenshot 474). Afterwards, in the taxa panel you are able to define sets of taxa to enforce monophyly (which I personally do not recommend because you want to test for support of the clade) or other types of calibrations. Then, in the tips panel you are informing BEAUTi and BEAST about the sampling dates of the sequences, which by default are assumed to have been sampled at the same time (time 0), you've selected the parsing dates option and that's ok, but I recommend you delete day and month from the sample name for simplicity. Height column only depicts the age of your tips relative to your time 0 (which is 2014).

Personally, I ignore the traits tab. In the sites panel, you will define the evolutionary model settings. For this, ideally you have tested your data to determine the substitution models that fit to it, again you can use several free software like Partitionfinder. Depending on the model your data fits to, you can select the options of substitution model, base frequencies, and site heterogeneity. Then, there´s the clocks tab, for which again, ideally you have tested your data to follow an strict or relaxed clock (you can do this in Mr. Bayes). In the trees tab, you select the tree prior (coalescent, yule, exponential growth) and the model (random, UPGMA).

Regarding your screenshot 482, for you to define nucleotide partitions, first you have to state the partitions of your sequence data at the end of your nexus file. Now the priors tab, some of them are improper by default and the software will indicate this to you by making it visible. In your example, you've selected a coalescent constant size tree prior which in consequence gives you a constant.popSize parameter, now you are able to modify the values and prior distribution of this parameter and BEAUTi will provide you a graphic distribution of the values you type in. Note that you also have to change accordingly your clock rates (in your screenshot 483, given by the parameters ucld.mean and ucld.stdev) for which you have to change the improper intervals (0, infinite) as this can give you problems in convergence and mixing of the MCMC chains. This is where you have to consider and type the mean and standar deviation of the mutation rate of the genes you are using. For example, for some mitochondrial genes in birds some of us set these values: mean 0.0123 +/- 0.02, or mean=0.029, or 0.018, but of course this depends on the gene and the species you are working with. If you set these values as priors, you will obtain ages of nodes as millions of years.

If you are interested in comparing rates among the lineages you could set the relaxed clock with a fixed mutation rate of 1.0 so the rates are realtive to each other, as a result the ages of nodes of the tree are going to be estimated in units of substitutions per site.

I usually leave alone the operators tab, unless the selected substitution model demands turning off some of them. Finally, in the MCMC tab you specify the lenght of chains, usually 10,000,000 steps will do. Your log parameter and echo seem fine also, given the lenght of your chains. Afterwards you just save your xml file and run BEAST with it.

Now, as Malik says, after the process is finished you import the log.txt file into tracer and check convergence. All ESS values should be over 200 for you to say the chains have converged and you have reached stationarity, and so your analysis is good. If you click on the trace tab, you can observe the sampled values of your MCMC.

If everything went good with your ESS values, now you can use FigTree, if you check the node bars box and display "height_95%_HDP" you will observe a bar ranging from the maximum to the minimum values estimated for your node ages. Note that you also have to check the scale axis box, and the reverse axis box.

Hope this is helpful. Good luck

Thanks a lot Michael Friedman

Thanks a lot Alberto Rocha-Méndez

Thanks a lot Malik Sallam