Consistent with SNP-h2 estimate patterns (Table S9, Figs. S9, S10, Supplementary Methods, Supplementary Results) and genetic correlations (Table S10, Figs. S11, S12, Supplementary Methods, Supplementary Results), we found that cell type association patterns of neurological disorders were distinct from psychiatric, substance use and behavioral/quantitative association patterns by hierarchical clustering (Figs. S13, S14). For brain-related phenotypes, there were no cell types in the 10x Genomics dataset to which implicated genomic loci consistently mapped using all three methods (Figs. S13–S15, Table S11). By comparing the P values of each cell type computed by LDSC, MAGMA, and DEPICT against one another, we found that MAGMA linear mode was considerably more lenient (Fig. S16) and susceptible to bias due to GWAS sample size (Fig. S17). Therefore, MAGMA linear mode was excluded from our main analyses.
