Partitioning heritability with linkage disequilibrium score regression

Summary statistics for 64 GWAS were obtained from a variety of sources in either “raw” or preprocessed forms. “Raw” GWAS summary statistics were downloaded and processed using the “munge_sumstats.py” script (LDSC v1.0.0). Data sources and specific command parameters used to process the data are listed in Supplemental Table S5. Note, processed summary statistics from the CLOZUK SCZ GWAS (Pardiñas et al. 2018) needed minor modifications after processing, and the GWAS summary statistics for Alzheimer disease (Marioni et al. 2018) have been modified since analysis (see Marioni et al. 2018 and Supplemental Table S5). Annotation files and LD score files needed for analysis were created using the “make_annot.py” and “ldsc.py” scripts included in the LDSC software using standard parameters. The source for software and data downloaded to run S-LDSC can be found in Supplemental Table S16. Each ATAC-seq sample was added onto the baseline model and heritability enrichment was calculated individually (also referred to as S-LDSC). The analysis was performed with the “ldsc.py” script using standard parameters with the “‐‐h2” flag.

Results for each phenotype were aggregated, and the P-values for each annotation were calculated in R. The P-values for regression Z-scores are based on a one-sided test for the regression coefficient being greater than 0, so the P-values for each annotation were calculated using the regression coefficient Z-scores and the “pnorm” function with the following parameters: “lower.tail = FALSE”. For more information, see LDSC publications (Finucane et al. 2015, 2018) and the LDSC website (https://github.com/bulik/ldsc). Partitioned heritability calculations for all traits were combined and analyzed in R. Plots were created using custom R scripts. The level of significance was set for LDSC results as the Bonferroni corrected P-value when taking into account all summary statistics and cell populations tested (0.05/[27*64] = 0.00002894; −log₁₀[P] = 4.53857). For more details and a description of peak subset analyses, see Supplemental Methods.