Figure 3:

Our low-pass model and ANGSD enable accurate demographic parameter inference. A&B: From data simulated under a single-population growth model, the final population size and time of growth onset (T) were accurately inferred using our low-pass model and a GATK-called AFS or using normal dadi and an ANGSD-called AFS. But they were biased if low depth was not accounted for when fitting a GATK-called AFS. (Dashed horizontal lines are simulated true values.) C: The likelihoods using the GATK AFS were similar whether or not low-pass biases were modeled. D: The fluctuations introduced into the AFS by ANGSD caused low likelihoods at low depth. E-H: For a two-population split with isolation model, similar results were found, although inferences from our low-pass model were slightly biased at 3× coverage.