Phylogenetic methods - computer practicals
Parsimony    Maximum likelihood    Bayesian inference    Species tree    home

Department of Botany, Charles University, Prague (2021-2024)

Bayesian analysis

Basic comments to MrBayes are here.
Comments to jModelTest 2 are here.
Notes to Newick tree format and DNA evolutionary models are here.

Training dataset is here. It consists of alignments of eight samples from the ginger family (Zingiberaceae) published by de Boer et al. 2018.
There are ITS, matK and concatenated alignments in NEXUS format.

A ZIP file with necessary software is here. Download it and unzip.
MrBayes - program for Bayesian inference and model choice of phylogenetic and evolutionary models
jModelTest 2 - program for selection of best-fit models of DNA evolution
FigTree - phylogenetic tree viewer

Other software recommended to install

• Notepad++
• Total Commander

Tasks (to work with Amomum dataset)
after you are done submit the answers using this Google form
If an advice is required, contact Tomáš Fér: tomas.fer_AT_natur.cuni.cz

1. Select the optimal evolutionary models for ITS and matK datasets

• using jModelTest 2 and AICc criterion
  Question 1: What is the best model for ITS dataset?
  Question 2: What is the best model for matK dataset?

2. Create Bayesian tree based on ITS dataset

• put the matrix file to the folder with MrBayes binary (exe file)

• double click on the MrBayes binary (i.e., mb.3.2.7-win64.exe), a black window appears

• type execute command followed by the file name in NEXUS format, e.g. execute Amomum_ITS.trimmed.nex

• set appropriate model (here K2P+I, i.e., two rate parameters and equal base frequencies, plus proportion of invariable sites) - but USE BEST MODEL ESTIMATED BY YOU, see the document with MrBayes commands how to create the commands prset and lset)
  lset nst=2 rates=propinv
  prset statefreqpr=fixed(equal)

• check the actual model by typing
  showmodel
  Question 3: How many different parameters there are?

• set run parameters (2 independent runs with 4 chains each, 100 thousand generation, sampling every 100th generation)
  mcmcp nruns=2 nchains=4 ngen=100000 samplefreq=100

•  run the analysis
  mcmc

• summarize the analysis (i.e., discard first 25%, i.e. 250, steps as burn-in; in total there were 100,000 / 100 = 1,000 generations sampled)
  sump burnin=250

• summarize the trees (again, discard burn-in fraction, then create a consensus tree)
  sumt burnin=250 contype=Allcompat

• check the screen results and the resulting files
  Question 4: What is Average standard deviation of split frequencies?
  Question 5: What is the lowest ESS value?
  Question 6: What is the harmonic mean of the fit? (check 'ALL' line in *.lstat file)

• open the resulting tree file in, e.g., FigTree, root with 'Siphonochilus' and display support values (prob)
  Question 7: How many branches are not fully supported?

• repeat the analysis of ITS dataset with the simple JC model (i.e., one rate parameter and equal base frequencies)
  Question 8: What is the harmonic mean of the fit?

3. Create Bayesian tree based on matK dataset

• repeat the steps from Task2, set the appropriate model for matK dataset (estimated by jModelTest 2)

• open the resulting tree in FigTree
  Question 9: What is the most basal taxon after the outgroup?
  Question 10: Is it different from ITS based tree?

4. Create Bayesian tree based on concatenated dataset (single partition)

• execute the concatenated dataset
• repeat the steps from Task2, set the model to GTR+I+G (see the document with MrBayes commands how to create the commands prset and lset)
  Question 11: What is the harmonic mean of the fit?

5. Create Bayesian tree based on concatenated partitioned dataset

• set characters per partition
  charset ITS = 1-593
  charset matK = 594-3154

• define partitions
  partition twoparts = 2:ITS,matK

• apply partition settings
  set partition = twoparts

• set models for each partition (here K2P+I for ITS, HKY+G for matK - but USE BEST MODELS ESTIMATED BY YOU, see the document with MrBayes commands how to create the commands prset and lset)
  lset applyto=(1) nst=2 rates=propinv
  prset applyto=(1) statefreqpr=fixed(equal)
  lset applyto=(2) nst=2 rates=gamma

• set variable rates among partitions
  prset ratepr=variable

• unlink the parameters for each partition (but keep 'topology' and 'brlens' linked - why?)
  unlink statefreq=(all) revmat=(all) tratio=(all) shape=(all) pinvar=(all)

• check the model
  showmodel
  Question 12: How many parameters are here?

• set the mcmc parameters similarly to Task2 and run the analysis, then summarize it

• open the resulting tree
  Question 13: How many branches are not fully supported (i.e., have lower support than 1.00)?

6. Compare the model fits using a Bayes factor

• compare absolute values of harmonic means of marginal log likelihood collected in the analysis of ITS dataset with K2P (or other suitable model you estimated; ln L₂ is the harmonic mean from Question 6) and JC models (ln L₁ is the harmonic mean from Question 8), using this formula in, e.g., Excel (this is equivalent to e^{ln L2 - ln L1})
  =EXP(ln L₂ - ln L₁). Do not forget that the numbers are negative (i.e., you should get the difference of absolute values of both means and pass this to EXP)!
  Question 14: What is the Bayes factor for the two models fitting the ITS dataset?
• if this Bayes factor is >3 there is an evidence for the first model, Bayes factor between 10 and 30 means strong evidence, >100 a decisive evidence
• compute Bayes factor for partitioned and non-partitioned concatenated dataset
  Question 15: What is the Bayes factor for partitioning the concatenated dataset?

Check also this overview of all analyses to be done with MrBayes within this task.

Thank you for participating in MrBayes practicals...