Module 3: Genotyping & Recombination

Assigning a genotype assumes the genome has one. Recombinants do not.

Learning objectives

  1. Describe how HBV genotypes are defined.
  2. Assign a genotype by similarity to references.
  3. Explain why whole-sequence classifiers fail on recombinants.
  4. Interpret a jpHMM breakpoint plot.

What a genotype is

HBV genotypes (A–J) are defined by sequence divergence — historically >8% across the complete genome. Subgenotypes (A1, A2, D3…) subdivide further.

Genotype is not academic bookkeeping. It tracks with disease progression, treatment response, and geography. In Botswana, A1 and D3 dominate, which is itself an epidemiological signal.

The naive approach

Align a query against references, take the most similar. This works, right up until it doesn’t.

ImportantThe recombination problem

A recombinant genome is genotype A across part of its length and genotype E across another. A whole-sequence classifier must return a single label. It will return one, with high confidence, and it will be wrong — the “similarity” it reports is an average over segments that disagree.

A confident answer from a method that cannot express uncertainty is the most dangerous output in bioinformatics.

Why jpHMM is different

jpHMM (jumping profile Hidden Markov Model) does not classify the genome as a unit. It walks along the sequence, and at each position asks which genotype’s profile best explains the data — with the option to jump between genotypes.

The output is a genotype assignment per region, plus the breakpoints where it switched. For a pure genotype, one assignment spans the genome. For a recombinant, you see the mosaic directly.

This is a general lesson: the right model is the one whose structure matches the biology. Recombination is a positional process, so the model must be positional.

Reading breakpoints skeptically

Not every reported breakpoint is real. Consider:

  • Low coverage. A breakpoint where depth collapses is suspect.
  • Short segments. A 50bp “genotype E” island inside genotype A is more likely noise than biology.
  • Reference bias. If the true genotype is absent from your reference set, jpHMM assigns the closest available — and may invent a mosaic to do it.
WarningMissing references manufacture recombinants

An unrepresented lineage gets explained as a patchwork of the lineages you did include. Before reporting a novel recombinant, ask whether your reference panel could represent the truth.

Exercises

  1. Assign genotypes to workshop sequences by similarity. Where is the call close?
  2. Run jpHMM. Do any sequences show breakpoints?
  3. For one breakpoint: check depth and segment length. Do you believe it?
  4. Discussion. You find a novel A/E recombinant. What must you rule out first?

Takeaways

  • Genotype assumes a single ancestry. Recombinants violate that.
  • Match model structure to the biology.
  • Suspect your reference panel before announcing a new recombinant.