The analyses were performed with two independent chains and 20 million generations and chains were sampled every 1000 generations with a burn-in of 2 million generations.
Each run consisted of a burn-in of 2 × 10 iterations, followed by 10 iterations.
The consensus tree ran for 10 generations with a burn-in of 2 × 10.
The analysis was performed with two independent chains and 20 million generations; chains were sampled every 1000 generations with a burn-in of 2 million generations.
These periods consisting of lifestyle modification ranging from 8 to 10 weeks and a placebo run-in of 2 weeks were used to identify stable HbA1 levels not affected by external conditions.
Based on initial trial runs we found that a heated chain approach encompassing 30 chains (parameters g1 = 0.7 and g2 = 0.4), an initial burn-in of 2 × 10 MCMC steps followed by 2 × 10 steps yielded convergence.
We finally performed three independent runs of 5 × 108 steps with a burn-in of 2 × 106 steps for the set of eight microsatellite markers, and three independent runs of 2 × 107 steps with a burn-in of 1 × 106 steps for the control region sequence dataset, using narrow priors after results of the preliminary analyses.
For each group number prior, three independent MCMC runs of 20,000,000 iterations each, sampling every 1000th step with a burn-in of 2,000,000 steps were performed.
For Bayesian analysis, six chains were run for 10,000,000 generations one of which was heated; after a burn-in of 2,500,000 generation every 100th tree was sampled only after posterior likelihoods reached stationary values.
Results shown are from 5 independent 20-million-generations runs with a burn-in of 2,000,000 (2000 trees) each.
