Use of "uninformed" prior

[trvrb]

The "uninformed" prior used in Table 1 is not described in detail, but it actually hides a subtle issue. For an unbounded variable like the x- and y- positions of a mapped isolate, the "uninformed" prior cannot simply be uniform on R^2, but rather is uniform on some compact subset of R^2. In other words, the authors had to assume a prior with a maximum possible x- and y-position. They don't describe what maximum bounds went into their "uniformed" prior, but this is well-known to influence the results of Bayesian analyses in general. This is a relatively minor issue -- authors should simply show how the form of his uniformed prior influences the resulting maps.

[trvrb]

We apologize for the lack of clarity. The 'Uninformed location prior' in Table 1 was described in Materials and methods as a diffuse normal prior (equation 10) with each coordinate having a variance of 10000. We've made this clearer in the text. We chose this variance based on prior knowledge of the scale of antigenic maps, i.e. around 50 units in breadth for A/H3N2 from Smith et al. 2004. In general, our use of diffuse priors limits their influence on the resulting antigenic maps. In this case, choosing a variance of 1000 gives very similar results.