A privacy-preserving HLA imputation method with homomorphic encryption

In recent years, several HLA imputation methods on local servers have been developed, using single nucleotide polymorphisms (SNPs) as inputs to predict Human Leukocyte Antigen (HLA) genotypes. However, these methods require HLA reference panels, which are often unavailable and memory-intensive. Cloud-based outsourced HLA imputation overcomes these limitations by utilizing built-in reference panels on the cloud server, removing local storage needs. However, uploading genotype data online raises privacy concerns. Although secure from third parties, the uploaded data can be misused by cloud server administrators. Additionally, reference panels and their HLA imputation model on the cloud servers must be safeguarded against malicious clients. To address these privacy issues, we developed privateHLA, the first secure HLA imputation method using homomorphic encryption. privateHLA securely performs HLA imputation on an outsourced server, protecting both the client's data and enhancing model privacy. privateHLA outperformed SNP2HLA but had slightly lower accuracy than CookHLA, both of which are plaintext-based methods.