Module 4: Phylogenetics
A tree is a hypothesis about history. Software will give you one whether or not your data contains any.
Learning objectives
- Build a maximum-likelihood tree.
- Explain what a bootstrap value does and does not measure.
- Choose and justify a substitution model.
- Explain why recombination breaks tree-based methods.
The alignment determines everything
Tree-building assumes column i is homologous across all sequences. If the alignment is wrong, every downstream inference is wrong — and the tree will still look perfectly reasonable.
Look at your alignment. Ragged ends and gap-rich columns are where errors live.
Substitution models
Not all changes are equally likely. Transitions outnumber transversions; base composition is uneven; sites evolve at different rates.
| Model | Assumes |
|---|---|
| JC69 | All substitutions equal (rarely true) |
| HKY | Transitions ≠ transversions |
| GTR | All six rates differ |
| +G | Rate varies across sites |
| +I | Some sites invariant |
GTR+G is a reasonable default for viral data. Use ModelFinder rather than guessing — and report what you used.
What bootstrap values mean
A bootstrap resamples alignment columns, rebuilds the tree, and asks how often each clade recurs. High values mean the data support this clade consistently.
A bootstrap of 100 on a systematically wrong alignment means the error is consistent. Bootstrap cannot detect bias — only noise.
It also says nothing about whether your model was right, your sampling was representative, or recombination occurred.
Rough reading: >95 strong, 70–95 moderate, <70 do not interpret.
Recombination breaks trees
A tree assumes one history for the whole sequence. A recombinant has two. Forcing it into a single tree produces a topology that is wrong everywhere.
If Module 3 found breakpoints, do not build one tree. Split at the breakpoints and build a tree per segment. Different segments give different — and both correct — topologies.
Trees are not epidemiology
A clade means sequences are similar. It does not mean transmission. Similar viruses cluster for many reasons: shared source, sampling, convergent evolution. Transmission inference needs dates, and usually more.
Exercises
- Build a tree with IQ-TREE and 1,000 bootstraps. Which nodes are supported?
- Compare JC69 to GTR+G. Does the topology change? Does support?
- Take a recombinant, build one tree, then per-segment trees. Compare.
- Discussion. Two sequences sit together with bootstrap 99. A colleague concludes direct transmission. What do you say?
Takeaways
- Inspect the alignment first.
- Bootstrap measures consistency, not correctness.
- Test recombination before building a tree.
- Clustering is not transmission.