# Differentially Private Genomic Data Release For GWAS Reproducibility

**Authors:** Yuzhou Jiang, Tianxi Ji, Erman Ayday  
**arXiv:** [2209.06327](https://arxiv.org/abs/2209.06327)  
**Date:** 2022-09-13  
**Summary page:** https://www.summarizepaper.com/en/paper/2209.06327v1/

## Key points

- Researchers face challenges in sharing genomic datasets due to privacy concerns and inference attacks
- A team of scientists has developed a differentially private sharing mechanism for genomic datasets
- The mechanism protects sensitive information while allowing for reproducibility
- The researchers have improved data utility of GWAS statistics using optimal transport on SNP values
- Experiments with real genomic datasets showed superior protection against attacks and higher GWAS utility
- This research provides a secure and efficient way to share genomic datasets while preserving privacy
- It facilitates reproducibility and collaboration among researchers in genomics research.

## Summary

In the rapidly evolving field of genomics, researchers have been developing various approaches and algorithms to advance scientific understanding. However, a major challenge arises when it comes to sharing the genomic datasets used in these studies. Due to privacy concerns and the risk of inference attacks, researchers are hesitant to publish their datasets for others to reproduce and validate their methods. To address this issue and promote collaborative research, a team of scientists has proposed a groundbreaking solution. They have developed a novel differentially private sharing mechanism specifically designed for genomic datasets. This mechanism ensures that the entire genomic dataset is protected under differential privacy, safeguarding sensitive information while allowing for reproducibility. Additionally, the researchers have focused on improving the data utility of Genome-Wide Association Studies (GWAS) statistics. They have developed a post-processing scheme that employs optimal transport (OT) on the empirical distributions of Single Nucleotide Polymorphism (SNP) values. This scheme not only enhances data utility but also maintains privacy during distribution creation. To evaluate the effectiveness of their approach, the team conducted experiments using real genomic datasets. The results demonstrated that their method offers superior protection against both genomic and machine learning-based membership inference attacks compared to baseline approaches. Furthermore, it provides higher GWAS utility, enabling more accurate analysis and interpretation of genetic data. Overall, this innovative research addresses critical challenges in genomics by providing a secure and efficient way to share genomic datasets while preserving privacy. By facilitating reproducibility and collaboration among researchers, this approach paves the way for advancements in genomics research and its applications in personalized medicine and population health studies.