BEAST 2 Remco Bouckaert BEAST 2 BEAST 2 RB subst model BEAST Add-Ons Remco R. Bouckaert remco@cs.{auckland|waikato}.ac.nz rrb@xm.co.nz Department of Computer Science University of Auckland & University of Waikato 1
BEAST 2 What BEAST 2 does Remco Bouckaert • The kind of Bayesian analysis as per citations on the BEAST 2 BEAST 1 wiki RB subst model BEAST Add-Ons • BEAUti 2: GUI to specify analysis • Sequence generator for simulation studies • Some post processing tools: log analyser, log combiner, DensiTree • Documentation for all the above – from user to developer, XML tweaker, etc. 2
BEAST 2 What BEAST 2 does that BEAST 1 doesn’t... Remco Bouckaert ...hence why you want to use BEAST 2 BEAST 2 RB subst model BEAST Add-Ons • Can resume runs when a chain is not mixing well • BEAUti 2: reload existing specifications – reduced need for XML hacking • Logs model with trace – allows looking up where the trace comes from • Provide a platform to develop add-ons - powerful interface, easy extensible XML, templates for BEAUti. • Book pending – expected 2012 3
BEAST 2 BEAST Add-Ons Remco Bouckaert Consider BEAST 2 as a library for MCMC and phylogenetics BEAST 2 RB subst model BEAST Add-Ons A BEAST 2 add-on is a library based on BEAST 2 Why add-ons: • Making work easier citable • Making the core easier to learn – it’s a lot smaller / cleaner • Separating out stable / experimental code / dead code • ... 4
BEAST 2 Add-ons Remco Bouckaert • SnAPP - multi-species coalescent for SNP and AFLP data http://snapp.otago.ac.nz/ • beastii - utilities, parallel treelikelihood, spreadsheet BEAST 2 RB subst model http://code.google.com/p/beastii/ BEAST Add-Ons • BDSKY: birth death skyline – handles serially sampled tips http://code.google.com/p/bdssm-beast2/ • Subst-BMA - Bayes model averaging over subst. models http://code.google.com/p/subst-bma/ • Beast classic http://code.google.com/p/beast-classic • Phylogeography http://code.google.com/p/beast-geo • Protein evolution http://code.google.com/p/evoprotein • Demographic model http://code.google.com/p/coalsir • Graphics http://code.google.com/p/beast-graphics/ • RB-substitution model? ... 5
BEAST 2 Reversible-jump Based (RB) substitution model for nucleotides Remco Bouckaert β frequencies π A G - α β γ � Q = π R � � � γ � α - δ ǫ BEAST 2 � � α � ω � � R = � � δ � � RB subst model β δ - ω � ǫ transition prob. C T BEAST Add-Ons - γ ǫ ω P ( t ) = e Qt Hierarchy of models # dimensions α β γ δ ǫ ω F81 (JC69) 1 1 1 1 1 1 0 HKY85 (K80) a 1 a a 1 a 1 TN93 a b a a 1 a 2 TIM a b c c 1 a 3 new a b c d 1 a 4 GTR (SYM) a b c d 1 e 5 Increase dimension by drawing a new parameter value from Γ( 0 . 2 , 5 ) 6
BEAST 2 Drawing from Prior Remco Bouckaert BEAST 2 RB subst model BEAST Add-Ons Exponential (mean 1) prior on parameters 7
BEAST 2 Simulating from subst model Remco Bouckaert BEAST 2 RB subst model BEAST Add-Ons 6 taxa, 3000 sites, tree from ape data, Yule tree prior, no gamma, 0 prop 8
BEAST 2 Hepatitis C: HKY massively preferred Remco Bouckaert BEAST 2 RB subst model BEAST Add-Ons (None of the dataset fits F81/JC69) 44 taxa, 576 sites, Bayesian skyline, 0 gamma, 0 prop invariant 9
BEAST 2 Apes: HKY preferred, but higher order models are explored as well Remco Bouckaert BEAST 2 RB subst model BEAST Add-Ons 6 taxa, 768 sites, Yule treeprior, 0 gamma, 0 prop invariant same pattern for Darwin finches, Bird flu 10
BEAST 2 Gophers: TN93 preferred, but HKY as well Remco Bouckaert BEAST 2 RB subst model BEAST Add-Ons α β γ δ ω β A G mean rate 0.181 0.525 0.143 0.094 0.200 � � � ’old value’ 1 1 0.181 0.143 0.181 � γ � � � α � ω � � � � δ � � � ǫ C T 26 taxa, 614 sites, Yule treeprior, 0 gamma, 0 prop invariant 11
BEAST 2 Dengue virus: TN93 preferred Remco Bouckaert BEAST 2 RB subst model BEAST Add-Ons (None of the dataset fits TIM or ’new’) 17 taxa, 1485 sites, Yule treeprior, 0 gamma, 0 prop invariant 12
BEAST 2 HIV: only GTR will do Remco Bouckaert BEAST 2 RB subst model BEAST Add-Ons Same results without and with partitioning sites at codon positions {1,2},{3} or {1},{2},{3} 46 taxa, 2382 sites, Coalescent (exp), 0 Γ , 0 PI. MP , FPV, animal dna, primates. 13
BEAST 2 Effective sample size Remco Bouckaert BEAST 2 RB subst model BEAST Add-Ons for all logged statistics for Ape (HKY), HVC (HKY), Dengue (TN93), Primates (GTR) In summary: RB subst models don’t require model selection, at no significant computational cost, and generate interpretable rates. 14
BEAST 2 RB Add-On: Implementing Plug-ins Remco Bouckaert BEAST 2 RB subst model BEAST Add-Ons Everything is a Plug-in Plug-ins connect with other Plug-ins through Inputs Plug-ins and Inputs provide documentation, type 15 checking, XML file parsing/producing
BEAST 2 Creating a new RB subst model add-on Remco Bouckaert BEAST 2 RB subst model • Check out BEAST 2 from google code. BEAST Add-Ons • Create a project in your favorite IDE (Eclipse, Intellij, etc) with a dependency on BEAST2 and do some programming... • Create documentation (examples, README, etc.) • Create BEAUti template (optional) • Wrap in a zip file and publish through beast2 wiki http://beast2.cs.auckland.ac.nz/ 16
BEAST 2 Installing RB subst model add-on Remco Bouckaert BEAST 2 RB subst model BEAST Add-Ons 17
BEAST 2 Installing RB subst model add-on Remco Bouckaert BEAST 2 RB subst model BEAST Add-Ons 18
BEAST 2 Why create BEAST add-ons Remco Bouckaert BEAST 2 RB subst model BEAST Add-Ons • No need to program an MCMC framework or phylogenetics library, file IO, etc. • Requires little effort to add to GUI • Easy way to distribute new models to users 19
BEAST 2 Summary Remco Bouckaert BEAST 2 • BEAST 2 is ready to be used, most standard models RB subst model BEAST Add-Ons from BEAST 1 are available • BEAST 2 has some models not in BEAST 1, like SNAPP and RB subst model • (The RB substitution model appears to select models at no extra computational cost) • Writing BEAST add-ons is (relatively) easy • Let me know if you need help setting up your own add-on 20
Recommend
More recommend