by stratified LD score regression89, which generalizes LD score regression85 to regress χ2 statistics for each SNP against LD scores with each functional category. Fine-mapping methods can also estimate functional enrichments, although these analyses are often restricted to disease-associated loci47,52,61. Notably, all of these summary statistic based methods have been applied to a large number of overlapping functional annotations, whereas methods that analyze individual-level genotypes have only been applied to a small number of non-overlapping functional annotations88,90. In addition, stratified LD score regression is not limited by the single causal variant per block assumption of the Bayesian hierarchical model, increasing power in settings of highly polygenic traits89. Application of the method identified significant cell-type-specific enrichments for many highly polygenic traits, including enrichments for histone marks in brain for smoking behavior and educational attainment—even though the summary statistics analyzed contained only one and three genome-wide significant loci, respectively. One limitation of the method is limited power for functional categories spanning a small percentage of the genome, motivating additional work in this area. As both summary statistic and functional annotation data sets grow larger and richer, identifying enriched functional annotations using summary statistic data will likely continue to be a fruitful endeavor.
