Department of Computer Engineering
CS 590 SEMINAR
Genetic Indistinguishability: Differential Privacy Approach for Genome Privacy Preserving
Computer Engineering Department
Nowadays genome sequencing not only enjoys a huge decrease in price, but also it is more available for individuals: they can sequence their genomes with cost of a few hundred dollars via companies like 23andme. This has lots of benefits for drug companies and health related researchers to work on genomes. However, it raises some concerns about the privacy of the individuals. For example, it's more beneficial for health insurance companies to deny insurance to someone who has the APOE gene (the gene related to Alzheimer disease); other corporations prefer to hire healthy individuals based on their genetics. Moreover, the genetic has very sensitive information about the individuals, and their families. Our purpose is to build a framework which provides genome data for scientists, while preserving the individual's privacy. In order to do this, we started our work with two antithetical approaches. First, we tried to be the adversary and we inferred the missing genomes of individuals and his families, using belief propagation and higher order correlations. Then we continued to build a method for revealing the genetic information while preserving privacy. For this purpose, we used differential privacy framework. Actually, we tried to keep the change of adversary's knowledge about sensitive genetics beyond a specific bound while revealing other genomes.
DATE: 28 Kasım, 2016, Monday @ 16:40