Workflow from Scientific Research

Graphical summary of typical GWAS data processing steps, inherent privacy risks and risks of reduced accuracy for centralised, meta and federated analysis. Data privacy during data collection (1) and quality control (2) can be ensured through the implementation of PETs, such as HE, DP and SMPC [ 77 ] and strict access control. There are no accuracy concerns. Data privacy standards during imputation (3) differ between international imputation servers and servers imputing locally or within a legislation border, such as EagleImp-Web (EU) [ 45 ], the Haplotype Reference Consortium (UK) [ 43 ] and TopMed [ 44 ]. Whereby local imputation can lead to a loss of accuracy in the case of obsolete references and algorithms, cross-legislation border transfers lead to additional privacy risks that can be combated through the use of HE, e.g. through the tool p-Impute . However, this still results in lower accuracy. Privacy risks and accuracy during the last three steps (4-6) depend on the GWAS study design: a centralised analysis is accurate but leads to privacy risks. A meta-analysis has a medium to low privacy risk but may suffer from reduced accuracy. A federated approach combines a medium to low privacy risk and high accuracy

DOI