First, you must accept that it does not exist "the best". Instead, you need to take estimating divergence time of any lineage with any kind of data as a serious endeavour that requires a lot of serious attention and exploration of the data. Secondly, you have to explore both the molecular data and any kind of calibrations (mainly fossil record) available with the required attention to the details. Thirdly you need to take attention to the biology of your organisms. Finally, you can explore different programs such as MCMC but keep in mind that you have to accept none of the established models will be suitable for all cases and the differences are less between programs but between the models of evolution available in the programs.

I am no expert on this, but I am guessing that the answer could be different, for different types of organisms. The fossil record is more solid for some organisms than others. Some organisms change shape more rapidly than others, so the fossils for various species and intermediate forms are more recognizable. The genomes of organisms that produce millions of offspring evolve differently than organisms like mammals that can only produce a few. The genomes of organisms like plants and fungi, evolve differently than the genomes of vertebrates. Animals with wings, or animals that swim in the oceans, have different patterns of speciation, than plants or animals that remain in one location for millions of years. Most of those factors influence more the number of samples, how the samples are chosen, and the number and type of genes needed for analysis, than the nuts and bolts of the analysis method (multiple sequence alignment etc).

Even within a type of organism such as mammals, the research is very different if you are dating events in the past 1 million years with a lot of good fossils, dates and even some "ancient DNA" such as human evolution with Neanderthals, Denisovans, and tens of thousands of modern humans sequenced, than if you are studying the evolution of rodents over more than 70 million years.

Hi Rowan,

For genome-scale dating, MCMCtree, part of the PAML package by Prof. Ziheng Yang, is a nice option. This program can reduce the computational demand by using approximate likelihood calculation. It is able to implement age constraints ( fossil calibration) with soft bonds. I hope this is helpful for you.

All the best,

Fangluan

You need to understand the current literature on the fossil record of your taxon very, very well. That is the limiting factor.

Archaeopteryx stand-alone tool

PhyloBayes works quite well, if you have enough computing resources available. And of course, you need to have a sufficient number of good fossil calibration points to date the tree. As you have many different loci you will not be able to use published molecular rates (as this only apply to the respecive gene).

First, you must accept that it does not exist "the best". Instead, you need to take estimating divergence time of any lineage with any kind of data as a serious endeavour that requires a lot of serious attention and exploration of the data. Secondly, you have to explore both the molecular data and any kind of calibrations (mainly fossil record) available with the required attention to the details. Thirdly you need to take attention to the biology of your organisms. Finally, you can explore different programs such as MCMC but keep in mind that you have to accept none of the established models will be suitable for all cases and the differences are less between programs but between the models of evolution available in the programs.

I'm aware that this is not what you asked, but - why on earth would you want to do that?

There's a lot, really a lot, of uncertainty in dating because of the incomplete fossil record, so to me it makes no sense to date using numerous loci because there is no precision to gain. It is also a waste of time to investigate (many) more loci than necessary to resolve the phylogeny.

Having said that, the *worst* approach I could think of using gene alignments would be combining concatenated data, node dating using just the oldest fossil, and with some secondary prior on the root. Concatention causes node heights to represent ~average coalescence time rather than speciation time, which could be anything from 0 to a few Ma younger. Node dating is often done using a uniform prior with minimum hard bound, which means that som upper bound has to chosen subjectively. Finally, adding an age prior on the root is something that you just don't do, because it will inflate the age of the ingroup - this is because the outgroup typically has fewer sequences than the ingroup which means that the outgroup branch length will become underestimated and "pull" the ingroup towards an older age.

The *best* approach that I can think of is analysing a selected few genes (say, 30) with a multispecies coalescent analysis and a dating approach that takes into account the whole fossil record (i.e. not just the oldest fossil). This can be done directly using a fossilised birth-death approach (tip dates for the fossils) or as node dating where the prior distribution of node height has been estimated separately from the fossil record using e.g. PyRate. To aid convergence it is possible to constrain the species tree topology. And the secondary age prior, if used, should be put on any node but the root.

To this end I would still use BEAST2, or RevBayes.

Cheers,

Thomas