While Eagle provides new levels of efficiency (and accuracy compared to fast alternatives) for phasing very large cohorts, we note a few limitations. First, Eagle relies on the IBD present within very large data sets to achieve high accuracy; on smaller data sets (e.g., N≈15,000), we recommend SHAPEIT2, which provides higher accuracy and is computationally tractable for such data sets. Second, along similar lines, we observed that when phasing all N≈150,000 UK Biobank samples together, Eagle achieved lower accuracy than SHAPEIT2 on the <10,000 samples of non-European ancestry (due to limited IBD). In practice, such samples are easily detected (e.g., by using FastPCA35 or SNPweights39) and could be phased separately with SHAPEIT2. Alternatively, a hybrid algorithm that uses the Eagle approach for most of the phasing computation but switches to the SHAPEIT2 model in segments of genome lacking IBD would be ideal; developing such an algorithm is a direction for future work. Finally, despite Eagle's speed, its computational complexity contains a quadratic term (like all other published methods) and will become daunting for million-sample data sets. Most simply, this issue
