The files here are scripts that were used in the phylogenetic analysis of an alignment of bacterial SSU rRNA genes used to illustrate use of the tree-heterogeneous models NDCH and NDCH2. This was described in “Phylogenetic analysis that models compositional heterogeneity over the tree”.
The alignment used in the bacterial SSU rRNA Thermus-Deinococcus analyses is available at https://doi.org/10.6084/m9.figshare.13708702.v1
but it is provided here, as the file thermus.nex.
Some analyses use “screen”. To detach screen runs, do Ctr-a Ctr-d after they start up.
Model choice for the 6-taxon dataset. TIM2 is best.
Analysis of the six-taxon dataset with TIM2, JC, and GTR. We get the correct biological tree for TIM2, and similar trees for GTR and JC. The ML tree for GTR and JC both had T_ruber+Deinococcus when T_ruber+Thermus was expected. However, bootstrap supports for the splits was similar among the three models.
| Thermus + | |||
| Meiothermus + | Thermus + | Meiothermus + | |
| Model | Deinococcus | Meiothermus | Deinococcus |
|---|---|---|---|
| JC | 78 | 30 | 58 |
| GTR | 84 | 44 | 45 |
| TIM2 | 86 | 47 | 45 |
Remove the T.ruber sequence. T.ruber is now known as Meiothermus ruber
p4 sRemoveMeiothermus.py
Do model choice on the 5-taxon dataset.
Analysis of the 5-taxon dataset with GTR, JC, and TIM2. We get the “Attract” tree.
p4 sCompo.py
- Make a composition table; GC content with and without constant sites
p4 sMakeCompoTable.py - Symmetry tests with IQ-Tree
bash doSymtest.sh - Symmetry tests with P4
p4 sP4SymTests.py - Chi-squared test
p4 sChiSquaredTest.py - Compo test using simulations
p4 sCompoTestUsingSimulations.py
ML optimization, pickle Tree+Model, make likelihood table
p4 sML.py
AU test with consel. The “consel” suite of programs needs to be installed
p4 sConsel.py
Do the sims and opts for the TAMCFT, and do the test evaluations. Results go in tamcft*
p4 TAMCFTests.py
- Do two runs each with GTR and NDCH models, using screen
bash doMcmc.bash
- Plot the likelihoods
p4 sPlotLikes.py
- Evaluate posterior predictive simulations
p4 sPostPred.py
- Diagnostic checks of the MCMC
sReadMcmcCheckPoints.py
- Restart for collecting Stepping Stone likelihoods. This takes a while.
bash doSteppingStone.bash
- Combine the step stone likes from the two pairs of runs
p4 sCombineSSLikes.py
- Examine those stepping stone likes, and estimate the marginal likelihoods. This week, this needs the
digestStepStoneLikes.pyfrom p4 in your path.
bash doDigestStepStoneLikes.sh
As in H_BayesianGTRandTwoCompNDCH, but only for NDCH, and with free compLocation, ie those proposals are turned on.
As in I_ndchWithFreeCompLocation, but using an MCMCMC.
As in H_BayesianGTRandTwoCompNDCH, but fully parameterized for composition. This is not NDCH2 because it has flat priors. The compLocation proposal does not apply, as it is fully parameterized.
As above, but now using NDCH2, with 4 runs. The first two have fixed priors, and the second two have free priors.
Plot the hyperparameters for the two runs where they are free. The hyperparameter for the internal comps varies a lot. Perhaps it needs a hyperprior?
p4 sPlotHypers.py
Same but with MCMCMC.
ML analysis with NDCH, two composition vectors
p4 sML_noRootComp.py
ML with NDCH, with a third composition vector on the root
p4 sML_withRootComp.py
Run the MCMC analysis as above. Making the consensus tree shows the consensus root.
p4 sCons.py
Since rooting the bacterial SSU rRNA dataset did not go well, this example uses the NDCH2 model on simulated data. To simulate the dataset,
p4 sSim.py
To run the MCMC,
bash doMcmc.bash
To see the consensus tree, with consensus root position and support for all roots visited during the MCMC,
p4 sCons.py
The root is correct, and the root position shows mixing.