I Built a Private Genomics Study with Stoffel MPC
5 hours ago
- MPC (multi-party computation) enables genomic studies to compute aggregate results without collecting participants' raw DNA.
- A proof of concept with Stoffel MPC successfully computed aggregate allele counts and a weighted cohort score from 100 simulated participants without any party seeing the complete dataset.
- The experiment used a decentralized design where each user's genotype data remains on their device, shares are sent directly to MPC parties, and no central server collects all data.
- Technical implementation involved 100 client processes, 4 Stoffel MPC parties, and a coordinator, all running on a single development machine.
- The computation completed in under a minute (45.29 seconds total) with correct results: aggregate counts for 6 SNPs and a weighted score of 498.
- Developer experience with StoffelLang was positive, making private computation feel like ordinary application development.
- Production considerations include independent party operation, mobile SDK development, query controls to prevent inference, and scaling to thousands of clients.