1. UT DALLAS Erik Jonsson School of Engineering & Computer Science FEARLESS engineering Privacy Preserving Data Dissemination Dr. Murat Kantarcioglu (muratk@utdallas.edu) Harichandan Roy (harichandan.roy@utdallas.edu) Data Security and Privacy Lab UT Dallas

2. FEARLESS engineering What is Task 1? Given- case/control datasets Desired result- perturbed datasets Case Control Desired Result snps Perturbed snps anonymous Method applied

3. FEARLESS engineering Proposed Method: NCBI Added Noise Main Idea: Keep the significant SNPs MAFs, add noise to non-significant SNPs MAFs For significant snips –Define significant and non-significant snips based on p-value using pLink (significant if p is less than 0.05) –Keep MAF as it is in case

4. FEARLESS engineering Proposed Method: NCBI Added Noise For non-significant snips –Take variance of MAF in case –Get gaussian random values, noises, with mean = 0 and std = (a*Math.sqrt(var)), where a = 3. –Finally, add noise to NCBI MAF –If NCBI MAF does not exist, add noise to average MAF of case and control Order all perturbed MAFs as in case file

5. FEARLESS engineering Flow Diagram

6. FEARLESS engineering Results For chr2, –Used NCBI Added Noise Method –Got power ≈ 8 <20 For chr10, –Used Average Case/Control Method –Got power ≈ 8 <20 Power ≈ 8 <20 –It means desired level of privacy is preserved Results are not always same

7. FEARLESS engineering Other Tries Method AppliedResult for chr2Result for chr10 Control-maf for non-sigPower > 60Power < 10 Avg-maf for non-sigPower > 60Power < 10 NCBI-maf for non-sigPower > 40Power < 10 Control-maf/2 for non-sigPower < 5Power > 20 Avg-maf/2 for non-sigPower < 5Power > 30 NCBI-maf/2 for non-sigPower < 5Power > 10 Min-maf for sig.Power > 40Power < 10 Max-maf for sigPower > 40Power < 10 Gaussian noise for non-sigPower < 15Power <10

8. FEARLESS engineering Questions?