FastHer: a fast and accurate estimator of local heritability from GWAS summary statistics

Local heritability estimation is essential for analysing the genetic architecture of complex traits. Although the contemporary HEELS method, by using GWAS summary statistics and linkage disequilibrium (LD) matrices, achieves accuracy comparable to the gold-standard GREML method (Genomic Restricted Maximum Likelihood), its computational complexity makes the analysis of extended genomic regions (spanning tens of thousands of SNPs) a practically intractable task. We present FastHer, a maximum-likelihood-based method that, like HEELS, retains GREML-level accuracy while overcoming its computational limitations. The key innovation lies in an analytical reformulation of the likelihood function using a single eigen-decomposition of the LD matrix, yielding orders-of-magnitude acceleration. In benchmarks, FastHer demonstrated an approximately 100-fold speedup when analysing genomic regions containing [~]10,000 SNPs. FastHer thus enables highly accurate and rapid local heritability analysis for extended genomic regions typical of biobank-scale data, making genome-wide studies computationally tractable. The method is implemented as an open-source R package.