Introduction

BEAST 1 does a fantastic job in performing a wide range of phylogenetic analyses. The success of BEAST 1 has meant that many researchers started using it and that demand has fuelled tremendous growth in its source code base. An unintended side-effect was that it made it hard for newcomers to learn the code. Since a lot of the work done with BEAST is at the cutting edge of phylogenetic research, parts of the code base are used to explore new ideas. Since not all ideas work out as expected, some experimental code is abandoned. However, only if you know what to look for is it clear which classes are experimental and which are production code. A partial solution to these problems in BEAST 2 is the package. A package is a library based on BEAST 2 that can be installed separately from the BEAST 2 core libraries. This way, the core BEAST 2 library remains small and all its classes are production code. This makes it easier for new developers and PhD students in phylogenetic research to learn BEAST 2. It also makes development work less cumbersome, so that BEAST 2 acts more like a platform for Bayesian phylogenetics, rather than a single monolithic code base. The platform is intended to be quite stable and individual researchers can develop and advertise packages independently of the BEAST 2 release cycle. This provides a cleaner mechanism for providing correct attribution of research, as individual packages can be published separately. Furthermore, it separates out experimental code from the core, which makes it easier to determine which classes are relevant and which are not. Users can install packages effortlessly through the package manager in BEAUti. Developers can check out code from the package repository. Some packages already available are:

1. •SNAPP for performing multispecies coalescent analysis for SNP and AFLP data;

2. •BDSKY contains a birth–death skyline tree prior;

3. •subst-BMA for Bayesian model averaging over non-contiguous partitions and substitution models;

4. •RB contains a reversible-jump substitution model and auto-partition functionality;

5. •BEASTlabs has a range of utilities such as multi-chain MCMC, some experimental methods of inference, a number of experimental likelihood cores; •MASTER is a framework for simulation studies; •MultiTypeTree contains classes for using the structured coalescent;

6. …