Privacy-Preserving Clustering Algorithm and Parameter Selection for Collaborative Population Stratification Using Distributed PCA
Organizations: Department of Computer and Data Sciences, Case School of Engineering, Case Western Reserve University, Cleveland, OH, USA. · Department of Bioethics, School of Medicine, Case Western Reserve University, Cleveland, OH, USA.
Abstract
We propose a privacy-preserving selection layer for collaborative population stratification workflows, enabling an honest-but-curious server to recommend a strong clustering algorithm-hyperparameter configuration from privacy protected shares. Instead of using a fixed workflow such as PCA followed by K-Means with a preset number of clusters, the framework compares multiple clustering methods and parameter settings under privacy constraints. To avoid relying on a public reference basis, we construct a shared low-dimensional representation through a distributed-PCA (DPCA) procedure in which each party computes local top- principal-component information and the server aggregates only low-rank summaries to recover a global basis. Building on this representation, the framework supports three LDP-protected pipelines: PCA then Noise, Noise then PCA, and Noise-Only. The server evaluates clustering methods and parameter settings using internal validation metrics and recommends a top-performing configuration from the perturbed data; this step is pure post-processing. Experiments on openSNP and 1000 Genomes show that the server-side selection layer reliably identifies strong settings from LDP-protected data. Across both datasets, the PCA-based pipelines outperform Noise-Only in clustering utility, communication cost, and runtime, and show lower membership-inference attack (MIA) power at small privacy budgets. Overall, they provide the best utility--scalability trade-off in our setting.
Figures & tables
| Symbol | Description |
|---|---|
| S | Server |
| Dataset of each party | |
| Noisy data of each party produced as a result of Laplace Noise | |
| LDP_processed data of party | |
| Global PCA basis learned by the server from aggregated local PC summaries. | |
| epsilon, Privacy Parameter |
| Dataset | Pipeline | Algorithm | Hyperparameter | Silhouette | CH | DB | Selection Score | |
|---|---|---|---|---|---|---|---|---|
| GMM | 3.0 | k = 3 | 0.9 | 13020.26 | 0.123 | 0.70 | ||
| openSNP | Noise PCA | HC | 3.0 | k = 3 | 0.90 | 13059.83 | 0.12 | 0.70 |
| K-Means | 3.0 | k = 3 | 0.9 | 13065.05 | 0.12 | 0.70 | ||
| DBSCAN | 3.0 | Eps = 11.06, minPoint = 5 | 0.76 | 4627.74 | 1.52 | 0.56 | ||
| GMM | 3.0 | k = 3 | 0.9 | 13025.87 | 0.12 | 0.7 | ||
| openSNP | PCA Noise | HC | 3.0 | k = 3 | 0.9 | 13031.27 | 0.12 | 0.7 |
| Dataset | Pipeline | Algorithm | Hyperparameter | Silhouette | CH | DB | Performance Score |
|---|---|---|---|---|---|---|---|
| GMM | k = 3 | 0.92 | 14365.4 | 0.10 | 0.71 | ||
| openSNP | Noise PCA | HC | k = 3 | 0.92 | 14365.4 | 0.10 | 0.71 |
| K-Means | k = 3 | 0.92 | 14365.4 | 0.10 | 0.71 | ||
| DBSCAN | Eps = 11.06, minPoint = 5 | 0.81 | 5066.43 | 1.41 | 0.60 | ||
| GMM | k = 3 | 0.92 | 14365.4 | 0.1 | 0.71 | ||
| openSNP | PCA Noise | HC | k = 3 | 0.92 | 14365.4 | 0.1 | 0.71 |
Privacy-Preserving ML
Privacy-preserving machine learning develops methods to learn, share, and run models while limiting exposure of sensitive data.
See all 445 papers