AI Helps Doctors Study Genes Without Sharing Personal Data

A new privacy-preserving approach lets researchers run large genome-wide association studies across institutions and borders without sharing individual-level genetic data. Each site computes its own summary statistics, and only aggregate results are transmitted, while the computation moves to the data using standard GA4GH Task Execution Service endpoints. In a five-site simulation involving 100,000 synthetic individuals and roughly 240,000 variants for Type 2 Diabetes and Body Mass Index, the federated analysis reproduced the association signals expected from a pooled study — without centralizing any genotype data.