Avian Influenza

Objective

Evaluate the role of wild waterfowl migration in AIV transmission.

• Characterize and map spatial patterns

• Determine dominant drivers of transmission accross space and their predictive ability.

Two Approaches to Model Drivers of Transmission

• Save decision on approach until group meeting.
• Option 1 is a subset of Option 2 because defines mechanisms to include.

Analysis plan: exponential random graph models or Bayesian mixed effect logistic regression (guess having not seen the data)

Quantify the probability of transmission between two nodes (watersheds) based on a set of predictors.

• Define the transmission network with watersheds as nodes the movement of virus as edges between them as data.

• The predictors we will start with include properties of the edges and nodes: number of birds moving between watershed 1 and 2 during a time window (Figure); age or species specific number of birds; distance/direction between watershed 1 and 2; temperature; watershed quality measurements.

• Estimate probability of the transmission network given the predictors and paramaters defining their relationship.

Data used to create predictors: (left) the number of banded and (right) recovered birds aggregated temporally.

Benifits:

• Can be used for model selection to evaluate the best predictors of transmission and clear methods exist to evaluate the goodness of fit.

• Can be used to predict forward.

Questions

• What does the data avaliable look like? How many? How are they related?

• For these objectives, do you have an opinion on using a transmission tree/network vs. phylogeny?
  • Jombart et al. (2011) argue for a network approach where all samples or watersheds are nodes. Transmission is defined based on timing and having the highest pairwise identity and a clustering algorithm, called Affinity Propogation (Hill et al. 2016)
  • Phylogenitic approaches are also used, where after ancestral reconstruction, transmission is defined based on (still black-box to me) Bayes factor tests and Bayesian stochastic search variable selection (Tian et al. 2015).

• Given these objectives and the data, do you recommend thinking about each subtype separately?

• Mixed infections - what proportion of samples came from birds with more than one infection?