Privacy in Epigenetics: Temporal Linkability of MicroRNA Expression Profiles
summary
Paper shows blood-based miRNA profiles are linkable over time: an identification attack reaches ~76% success on blood (≈30% plasma), while a global matching attack reaches up to 90% on blood (≈30–55% on longitudinal plasma). Data are 29 athletes (1 week, >1,000 miRNAs, blood+plasma) and 26 lung-cancer patients (up to 8 timepoints, 3-month intervals). Two defenses are evaluated: hiding subsets and a probabilistic sanitization (multivariate Laplacian, differential privacy); noise gives better privacy–utility, often halving linkability with <1% diagnostic accuracy loss.
AI-written summary · about
chapters (9)
- 0:00Start
- 0:20Primer
- 0:58Most Interesting Properties of Micrornas
- 1:44Comparison between Dna and Micronase
- 9:34Identification Attack
- 16:00How Does the Success Rate Evolve with the Time Period in between the Samples
- 17:02Probabilistically Sanitizing Micronic Expression Profiles
- 19:36Results
- 22:28The Probabilistic Standardization
related talks
46mFrom Health Data to Location Traces: Addressing Spatiotemporal Correlations in Differential Privacy
28mBiological Discovery and Consumer Genomics Databases Activate Latent Privacy Risk in...
01h 04mPrivacy Dependencies
01h 43mPerspective
42mDifferential Privacy
10mHealth Data Anonymization and Privacy
20mShare data without revealing personal information
09mReplication: The Effect of Differential Privacy Communication on German Users'...