Fast and robust detection of ancestral selective sweeps

There are few methods tailored for detecting signals of positive selection in populations directly ancestral to multiple descendent populations. We introduce the ancestral branch statistic (ABS), a four-population summary statistic for identifying selective sweeps occurring in the direct ancestor of a pair of populations. Simulations show that ABS performs at least as well as, and often better under model violations, than the complementary likelihood approach of 3P-CLR across diverse selection scenarios and parameter values. We first applied ABS to contemporary human genomic data to identify genes that may have been adaptive in ancestral East Asian populations, uncovering the well-established candidate EDAR, as well as a novel candidate SLC35F3, which encodes a putative thiamine transporter that may have been involved in adaptation to eating polished grains. Next, we performed scans with ancient European genomic data to reexamine evidence of recent positive selection in ancestral Europeans. The MCM6/LCT cluster and the SLC45A2 and HERC2 genes are strong outliers, agreeing with previous studies. Novel candidates, such as SLC30A9 and CYP1A2, may have been involved in adaptation to local nutrient sufficiency and lifestyle changes. Finally, we provide open-source software, CalcABS, which can perform genomic scans of ancestral sweeps with ABS from population allele frequency data.