Attention qpAdm users! I've been testing the performance of qpAdm (a tool for modeling population admixture) using simulated data, and I am excited to share what I have learned! The manuscript is up on bioRxiv, but here’s a sneak peek of what I found: 1/9 https://www.biorxiv.org/content/10.1101/2020.04.09.032664v1

qpAdm accurately estimates admixture proportions even in cases where data quality is decreased due to low coverage, high rates of missing data, pseudo-haploid data, reduced population size, uniform levels of ancient DNA damage, or ascertainment bias. 2/9

Including pops with very different C-to-T damage rates in the same model can result in biased estimates and decreased p-values, particularly when the target and source pops have different damage rates. So be careful with models that contain both ancient and present-day pops. 3/9

qpAdm produces uniformly distributed p-values when applied to an optimal admixture model, matching theoretical expectations (yay!), but that doesn’t mean that you can rank p-values to identify the optimal admixture model when you have several plausible options. 4/9

Trying to design an optimal strategy for comparing qpAdm models? I recommend using a “rotating” strategy, where pops serve either as sources or references. Try all combinations of source pops, and when a pop isn’t serving as a source in your model, include it as a reference! 5/9

*Sidenote* Previous discussions of qpAdm have used the term “outgroup population” in a confusing way. I propose we use the term “reference population” instead. 6/9

Including too many reference pops in a single model can result in reduced p-values. We see issues when we exceed 35 reference pops, but this number is likely to depend on the dataset being used, so to be safe, keep the number of reference pops as low as possible! 7/9

Just because you are able to find a plausible admixture model using qpAdm doesn’t mean that the population history you are modeling actually involves admixture. Be sure to consider other possibilities, like continuous gene flow, too! 8/9

Finally, if you find yourself wishing that you had a straightforward user guide for qpAdm that covers basic usage, optional parameters and provides recommendations for setting up your analyses, we’ve got you covered. Check out the updated qpAdm User Guide in the supplement. 9/9