For all analyses, Markov Chain Monte Carlo (MCMC) samples were based on 100,000,000 generations, logging every 2,500 steps, with the first 10,000,000 generations discarded as the burn-in.
We implemented two runs in BEAST with MCMC chain length of 350 million generations logging parameters every 3,500 samples.
We ran the analysis for 30 million generations, logging every 1000 generations, and discarded the first 3 million generations (10%) as burn-in.
We sampled the posterior distribution with an MCMC of 80,000,000 generations, logged every 1,000 steps for a total of 80,000 data points, the first 20,000 were discarded as burn-in.
To replicate the findings from the F1 generation, log-linear models with repeated measures of eczema were used in F2 generation analyses.
There were highly significant differences between snail lines in the % infection prevalence following exposure to five miracidia by the F1 generation (log-linear analysis: χ2 = 93.83, d.f. = 2, P = 0.001) and this difference remained significant in all subsequent generations (F1 to the F6 generation used here).
We examined the output plot of generation versus log likelihood to find that stationarity was reached.
The Markov chain Monte Carlo sampling was run for 10 generations and logged every 1,000 generations.
Bayesian analysis was then performed on the alignment in BEAST 1.6.1 [ 91], running 10,000,000 MCMC generations and logging every 1,000 generations, with the following departures from the default priors: constant population size; GTR+G substitution model with starting substitution matrix and gamma parameter values chosen using ModelTest.
