To identify cell types underlying various phenotypes, we employed four methods (Fig. 1). LDSC (version 1.0.1) [21, 22] was first used to estimate SNP-h2 and bivariate genetic correlations across all traits. We then constructed the specificity metric Sg,c from the 10x Genomics dataset, denoting specificity of a gene g for cell type c, which was used as input for LDSC, MAGMA and DEPICT (Supplementary Methods). The specificity metric of the KI dataset was previously constructed [8]. Next, LDSC (version 1.0.1) [21, 23], MAGMA (version 1.08) [8, 24] and DEPICT (version 1, release 194) [25] were employed to test cell type enrichment using the 10x Genomics and KI datasets. We utilized the diversity in statistical approaches of these methods to strengthen the confidence in consistently enriched cell types across methods. In brief, with LDSC we investigated the top 10% cell type-specific genes for enrichment of SNP-h2; with MAGMA we tested whether trait associations linearly increased with cell type-specificity or whether the top 10% cell type-specific genes were associated with gene-level association to traits; with DEPICT we evaluated the enrichment of cell type for genes from trait-associated loci.
