To identify more differentiated cell types, the analysis was expanded by using the KI level 2 dataset (Figs. S20–S22, Table S13), which includes 149 cell types that were subtypes of the cell types identified in the level 1 dataset. Using both the KI level 1 and level 2 datasets, we again found that MAGMA linear mode, compared to strength of association estimates computed by LDSC, MAGMA top 10% mode and DEPICT, provided disproportionally large estimates (Figs. S23, S24) and was prone to inflated results due to GWAS sample size (Fig. S25).