Privacy-preserving chi-squared test of independence for small samples.

Area of Science:

• Genetics
• Bioinformatics
• Cryptography

Background:

• Privacy protection is crucial for personal data analyses like genome-wide association studies (GWAS).
• Chi-squared (χ²) tests are used in GWAS to identify disease-associated single-nucleotide polymorphisms (SNPs).
• Existing privacy methods for χ² tests have limitations in table size and accuracy with small sample sizes.

Purpose of the Study:

• To develop a novel anonymization method for χ² testing that addresses limitations with small sample sizes.
• To ensure the proposed method satisfies differential privacy guarantees.
• To evaluate the performance of the new method against existing approaches.

Main Methods:

• Introduction of RandChiDist, a new differentially private anonymization technique for χ² values.
• Mathematical proof of RandChiDist's adherence to differential privacy standards.
• Experimental evaluation using synthetic and real genomic datasets.

Main Results:

• RandChiDist successfully anonymizes χ² testing for I×J contingency tables, even with small sample sizes.
• The method satisfies differential privacy.
• RandChiDist demonstrated superior performance by minimizing Type II errors while controlling Type I errors compared to existing methods.

Conclusions:

• RandChiDist is a novel, differentially private method for anonymizing χ² values in studies with limited sample sizes.
• Experimental results confirm RandChiDist's effectiveness and superiority over existing methods for small sample scenarios.
• This method enhances privacy in genetic analyses without compromising accuracy.