Regardless, one trend within Figure 4A and 4B is clear: with an expanded reference panel containing nearly 2,000 chromosomes, it is possible to improve imputation accuracy substantially over what is attainable with 120 chromosomes. For example, the IMPUTE v2 runs with k = 40 (solid red line) achieved best-guess discordance rates of 3.40% and 0.86% in Figure 4A and 4B, respectively, whereas the rates for IMPUTE v1 (which had access to only the HapMap reference panel; blue line) were 5.42% and 1.62%. BEAGLE (green), fastPHASE (black), and IMPUTE v2 (red) were all able to increase accuracy with the expanded reference panel, but the improvements for fastPHASE were smaller. BEAGLE (solid green line) and IMPUTE v2 with k = 40 (solid red line) yielded similar results: for BEAGLE, the best-guess discordance rates in Figure 4A and 4B were 3.46% and 0.93%. For IMPUTE v2, increasing the number of conditioning states used for phasing updates to k = 80 further reduced the discordance rates to 3.07% and 0.78%.
