why the two different versions of BEAST?

256 views
Skip to first unread message

Mike Martin

unread,
Sep 29, 2024, 2:30:07 PM9/29/24
to beast-users
I am very curious about why there are two independent BEAST projects running, e.g. BEAST X and BEAST2? It's rather confusing! 


Alexei Drummond

unread,
Sep 30, 2024, 12:38:45 AM9/30/24
to beast...@googlegroups.com
Hi Mike,

Both projects are open source, so anyone can work on (or fork) either at any point. I think if you ask two people involved, you'll get two different answers as to why we have this situation, so the following is just my understanding (and it is undoubtedly biased).

Around 2009, the original developers of BEAST 1 and a large group of Bayesian phylogenetics researchers held a series of working group meetings about the future of BEAST and Bayesian phylogenetics. These meetings led to the development of a prototype for BEAST2, primarily by Dr. Remco Bouckaert. The aim was to build a new version of BEAST that would address some problems that had been identified with the original codebase. One key goal (among others) was to make it more easily and independently extendable by third-party developers.

However, at the end of a long process, not everyone was convinced of the benefits of moving to a new codebase. Some felt that the changes in coding style were more drastic than necessary. It was harder than anticipated to keep a globally distributed team coordinated and in agreement. I believe that failing to manage this process is the single biggest failure of my career. As a result, two teams emerged: one that continued developing in the BEAST 1.x codebase (which has recently been renamed BEAST X) and another that coalesced around the BEAST2 platform.

On the positive side, these two projects have both developed in interesting and complementary ways, and diversity often provides value. I hope that in the future we will be able to have this type of diversity while also having some level of interoperability across the entire Bayesian phylogenetics landscape (i.e. including RevBayes etc as well), so that users can more easily move from one platform to the other without unnecessary esoteric differences.

Finally, I agree that the naming has become confusing as a result, and we need to address that now more than ever. If you have any suggestions, we are happy to hear them!

Cheers,
Alexei

On 23/09/2024, at 2:12 AM, Mike Martin <sameo...@gmail.com> wrote:

I am very curious about why there are two independent BEAST projects running, e.g. BEAST X and BEAST2? It's rather confusing! 



--
You received this message because you are subscribed to the Google Groups "beast-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beast-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/beast-users/857a3baa-910f-48b8-8b17-e26c4c06d2dan%40googlegroups.com.

Julian Tang

unread,
Dec 4, 2024, 12:56:40 AM12/4/24
to beast...@googlegroups.com
Thanks Alexei - hope you are well.

I've often wondered about this, so your explanation is useful.

But do the two Beast programs give similar results when applied to the same database, i.e. can we use one to check/confirm the results of the other?

Julian



Alexei Drummond

unread,
Dec 4, 2024, 1:08:56 AM12/4/24
to beast...@googlegroups.com
Hey Julian :)

If the models and prior are specified the same then they should give the same results up to Monte Carlo error (which reduces with longer chain lengths).

The trick is that they both implement models that the other codebase doesn’t, so only a subset of models are implemented in both codebases.

But something simple like constant coalescent + strict clock + HKY + G should give exactly the same results in both codebases if the same priors are also specified for all random variables.

Cheers
Alexei


Julian Tang

unread,
Dec 4, 2024, 2:33:46 AM12/4/24
to beast...@googlegroups.com
Reply all
Reply to author
Forward
0 new messages