Privacy preserving for genomic data analysis and sharing

Genomic data is a rich source of knowledge that can reveal important details about a person and their ancestry. However, accessing and sharing genetic datasets is essential for advancing research and enhancing healthcare outcomes. Allowing sharing, accessing, and analyzing the genomic data while preserving users’ privacy is challenging. Differential privacy (DP) is a promising lightweight technique for privacy preservation in the genomic domain which relies on adding noise to the data or the
trained model weights. However, relying on adding a static privacy budget to the models and query results no longer benefits the adaptive mode for genomic analysis as the sensitivity of the query can change over time, and queries can be correlated with each other in genomic analysis.
In this research, we aim to add an adaptive noise scaling to help strike a better balance between privacy and utility in DP under genomic analysis and model training so we can better quantify a privacy loss over time for each query with a privacy filter formulation.

Faculty Supervisor:

Ziad Kobti

Student:

Partner:

Karlsruher Institut für Technologie

Discipline:

Computer science

Sector:

Education

University:

University of Windsor

Program: