Associating Viral Sequence With Epidemiological Data

Question

[Agapow]

The setup: I have a large number of sequences from a viral pathogen and the associated epidemiological data, collected during a major disease outbreak. I deeply suspect that the varied epidemiology seen across the outbreak (case severity and outcomes, transmission rate, etc.) is the result of changes in the viral sequence.

The question: So, how do I best correlate these epidemiological data with sequence data? In the crudest sense, how can I point at a SNP and say "this is associated with more severe cases"?

Complications:

• I'm concerned about phylogenetic inertia, i.e. false correlations caused by evolutionary relationship. A given sequence change may correlate with increased fatality because it was fixed in the lineage that infected a weakened group of hosts.

• Some characteristics which are technically non-heritable will behave as heritable, e.g. location.

Solutions I've considered:

• Tools from GWAS studies or similar: apart from the possible overkill of using these on such a short genome, I don't know of any GWAS tools that deal with the inertia problem..

• Comparative analysis with independent contrasts: would be the obvious choice if I was dealing with solely character data. I could hack an suitable dataset together, say by treating a SNP loci as a character, but it seems ugly. Also, the state of useful software here is not good.

• Selection: will tell me what sites are being selected for but not what might be correlated with that selection.

• Compare controls: is something I've done before, but in this case it seems that deciding what to control for is pre-emptively deciding what won't correlate.

[Meredith]

Exactly what kind of epidemiological data do you have, e.g. is it already aggregated by viral sequence, or do you have individual case data at your disposal?

[Agapow]

Individual case data, dates, outcomes, the whole paella.

Answer 1

[David Quigley]

You should look at the literature on evolution of tumors, e.g. Navin Nature 2011, Bozic PNAS 2010, Wood Science 2007. Although tumors are often called clonal populations, a more sophisticated model of tumor evolution starts with a single aberrant cell producing a heterogeneous population of offspring which mutate independently within the overall tumor mass. The problem of determining which of many somatic mutations is a strong candidate to be causal (a.k.a. a driver mutation) and which is a bystander or passenger mutation is the same as your inertia problem.

I would guess that creating a phylogenetic tree based off of the sequence alterations would be crucial for establishing causality candidates, but this isn't really my area of expertise. You might structure the question as a set of regressions, asking whether a given alteration is associated with your phenotype and then looking for progenitor alterations highest up on the tree. If you're using a regression-based statistic you can control for biases such as location. I would consult a card-carrying molecular epidemiologist to avoid re-inventing the wheel.