In practice, reference-based phasing is typically performed on a target set containing many samples, allowing the potential to improve phasing accuracy by using inferred target haplotypes to phase each other. By default, Eagle2 performs a variable number of phasing iterations chosen based on the relative size of the target (Ntarget) and the reference (Nref). This behavior is intended to allow Eagle2 to automatically benefit from increased statistical power available from larger target sample sizes. Specifically, if Ntarget < Nref/2, Eagle2 performs only one phasing iteration (phasing each target sample using only the reference haplotypes). If Nref/2 ≤ Ntarget < 2Nref, Eagle2 performs two iterations, augmenting the reference panel with the inferred target haplotypes during the second iteration. If Ntarget ≥ 2Nref, Eagle2 performs three iterations in an analogous manner. Whenever Eagle2 performs more than one iteration, all iterations prior to the final iteration use K/2 conditioning haplotypes to save time, given that the last iteration has the most impact on accuracy. The number of iterations can also be set directly via the –pbwtIters parameter.
