Heritability from summary statistics of admixed and related biobank participants

Summary-statistics estimators of heritability and genetic correlation assume that participants resemble each other only through the SNPs, so analyses drop relatives and admixed participants, which costs the most precision in small ancestry groups. This work extends the SUMMIT framework so that relatives stay in the GWAS, their phenotypic resemblance is removed from the estimating equations without assumptions about its form, and the resemblance of admixed participants through shared local ancestry is modeled with a second kernel, all from summary statistics and in-sample genome-wide LD scores.