Lastly, the lung MXI1 annotation captured 31.6% (s.d. = 3.0%) of height heritability; notably, within the MXI1 gene lies a genome-wide significant variant associated with height50. Overall, we observed higher heritability enrichments for autoimmune diseases and hematological traits than for brain-related, lung-related and adrenal traits, likely reflecting the availability of relevant cell-type-specific functional data. To demonstrate the value of the IMPACT annotation strategy over functional annotations derived from single experimental assays and from machine learning models, we directly compared heritability analysis results and observed that IMPACT consistently outperforms these other annotation strategies (Methods and Supplementary Note). We show an example with cell-type-specific histone marks3 comparing the proportion of heritability captured in Fig. 2c and τ* in Extended Data Fig. 4. Since IMPACT annotations of the same cell type are correlated, we performed serial conditional analyses to identify IMPACT annotations explaining heritability independently from one another; we identified 38 complex traits whose genetic variation regulates multiple distinct cell-type-specific regulatory programs (Supplementary Note).
