BLocPRS: locus-aware ensembles of polygenic scores provide accurate linear prediction

BLocPRS, a locus-aware polygenic score ensemble, outperformed the genome-wide MultiPRS ensemble in all 25 held-out UK Biobank analyses, with mean relative gains of 14.0% for continuous traits and 7.9% for diseases.

• Why it matters: Genome-wide PRS ensembles can miss locus-specific differences in genetic architecture across cohorts and phenotypes, limiting prediction when genetic effects vary by genomic region.
• What they did: The method learns interpretable, locus-specific weights for an ensemble of PRSs, and was evaluated across 25 held-out UK Biobank analyses (13 continuous traits, 10 diseases), including non-European participants and transfer to CARTaGENE.
• The result: Performance gains increased with training size, generalized to CARTaGENE, and reflected locus-specific recalibration and shifting source weighting, including added contributions from lower-powered non-European-derived PRSs at trait-associated loci.

AI summary of the abstract · may contain errors · not medical advice · read the paper.