Preprint quantifies real-world risk of genome re-identification using polygenic phenotype prediction

A preprint posted to bioRxiv on 10 June 2026 addresses a longstanding question in genomic data governance: to what extent can an anonymised genome be re-identified by comparing polygenic predictions of a person's observable traits against their known characteristics? This class of attack — phenotypic tracing — has been discussed extensively in the genomic privacy literature, but prior studies have been criticised for overstating its practical feasibility.

The authors developed a probabilistic framework to quantify re-identification risk under realistic conditions, drawing on improvements in polygenic score accuracy made possible by increasingly large GWAS cohorts. Their analysis attempts to provide a calibrated assessment of the threat landscape rather than a worst-case theoretical bound, which has important implications for how genomic data repositories design consent frameworks, data access tiers, and de-identification standards.

The findings are relevant to researchers working in biobank governance, data-sharing policy, and genomic informatics, as well as to institutions responsible for designing ethical frameworks for large-scale genomic datasets. The preprint has not completed peer review. No personal genomic data from identifiable individuals was used in the publicly described analysis.