Post date: May 29, 2015 7:28:30 PM
Results from the popanc model with selection are in /home/zgompert/labs/evolution/projects/popanc_sims/analysis/. Parameter estimates and true parameter values are in subdirectories in the same directory, and the diagnostics (ESS and Gelman-Rubin) are in the sims directory. In general selection didn't affect model performance. It had an effect on true ancestry and a lesser effect on inferred ancestry frequencies (again, because extreme values tend to pull towards the genome-average). 95% CIs still have poor coverage, but true and inferred population ancestry frequencies are highly correlated (e.g. 0.76 and 0.83 in the example plots). Interestingly, the scale parameter decreased every so slightly.
So, the model works alright, it could be better, but that might not be possible. The real test is whether it beats simply counting sample ancestry frequencies in structure (running now). If so, its a useful advance, if not it might not be worth pursuing.