Software

Genotype and Phenotype Encryption Optimised for Federated Quantitative Genetics

---

Abstract

We have developed a framework for encrypting genome and phenotype data that is well suited to many quantitative genetics analyses. The key idea is to apply a random orthogonal transformation to the genotype dosage matrix and the matrix of phenotypes and covariates. The usual Gaussian likelihood is invariant under orthogonal transformation, so that all inferences based on the encrypted data are identical to those using the plaintext. This is a form of Homomorphic Encryption, but which differs from that used in other fields in that there is no need to decrypt the data in order to obtain results as they are invariant. The range of analyses that can be accommodated includes (but is not limited to) the standard mixed model using a SNP-based genetic-relationship matrix, and the Bayesian Alphabet models where the likelihood is multiplied by a prior on the SNP effects, and the estimation of heritability (see PMIDs 32327562, 38085098 for details). The framework is ideally suited to federated analyses where a consortium wishes to jointly analyse their data to increase statistical power, but without decryption back to the plaintext data.We have implemented a federated version of the method in a Python package pyhegp, which we describe here. We have tested it across multiple federated animal data sets and Bayesian alphabet analyses, and verified that cyphertext analyses produces the same results as plaintext. The package is freely available from https://github.com/encryption4genetics/pyhegp and is suitable for consortia who wish to jointly analyse data without sharing raw genotypes and phenotypes.

Keywords: 2026

How to Cite:

Isaac, A., Cheng, H., Li, D., Dekkers, J. & Mott, R., (2026) “Genotype and Phenotype Encryption Optimised for Federated Quantitative Genetics”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2285651. doi: https://doi.org/10.31274/wcgalp.23767

Rights: 1