We have described a new phasing algorithm, Eagle2, which we have incorporated into the Sanger Imputation Service and the Michigan Imputation Server to offer free reference-based phasing using the N=32,470-sample Haplotype Reference Consortium panel. This service enables high-accuracy phasing even in smaller cohorts, which was not previously possible. Eagle2 achieves substantial gains in speed and accuracy over previous methods via a novel search-based algorithm employing the positional Burrows-Wheeler transform. We believe this method is timely, as large sequenced reference panels (e.g., the HRC) are now becoming available for use—but must be utilized via analyses run on central servers due to consent restrictions. We anticipate that Eagle2's phasing speed—1.5 minutes per genotyped sample—will help keep computation tractable as demand for this service increases. Additionally, we anticipate that our release of Eagle2 as open-source software will aid in future method development and integration into analysis pipelines.
