This paper presents bioETH-PRS, a method to compute polygenic risk scores (PRSs) without ever revealing raw genetic data or the underlying model weights to a third party. The authors combine fully homomorphic encryption (FHE) with smart contracts on an FHE-enabled blockchain (fhEVM). That lets the blockchain’s deterministic contract logic perform the PRS calculation on encrypted inputs, so neither patient genotypes nor genome-wide association study (GWAS) weights are exposed during computation.

At a high level, the system replaces a trusted “evaluator” server with four auditable smart contracts. A Genomic Registry holds encrypted genotype references and per-address access lists. A Model Marketplace publishes GWAS weight vectors and quantisation metadata. A PRS Compute Engine runs the encrypted dot-product between a patient’s genotype vector and a model’s weight vector. Finally, an on-chain Oracle controls how results are released, returning an encrypted noisy-score handle and a small publicly decryptable category rather than a raw score. Access to decrypted outputs is controlled by an on-chain access-control list (ACL).

Two implementation points are important. The fhEVM used here supports TFHE, an integer-exact FHE scheme, so signed floating-point model weights must be converted to unsigned integers. The team designed a three-step fixed-point quantisation scheme that packs signed GWAS floats into unsigned 64-bit integers and, on their test fixtures, reconstructed scores with very small error (reported correlation r>0.999 and mean-squared error <2.3×10−6). They also implemented two execution styles: a classic chunked path and a streaming path. In their prototype the streaming path reduced mock-measured gas by 37% compared with the chunked path. Prototype runs used real GWAS fixtures with 100–5,000 single-nucleotide polymorphisms (SNPs) and a 5,000-SNP proof-of-principle for on-chain FHE computation.