PPPCT: Privacy-Preserving framework for Parallel Clustering Transcriptomics data.

Area of Science:

• Genomics
• Computational Biology
• Data Privacy

Background:

• Single-cell transcriptomics data offers vital patient health insights but faces significant privacy risks from data breaches.
• Genomic data attacks can compromise sensitive health information of patients and their families, with leaked data being permanent.
• Existing clustering methods for single-cell data often neglect critical privacy considerations.

Purpose of the Study:

• To introduce an efficient, fast, and privacy-preserving approach for clustering single-cell RNA-sequencing (scRNA-seq) datasets.
• To ensure data privacy, achieve high-quality clustering, handle high dimensionality, and maintain reasonable computation times for large datasets.
• To leverage secure computing environments and advanced algorithms for robust scRNA-seq data analysis.

Main Methods:

• Utilizes a map-reduce scheme for parallelized clustering to handle intensive calculations.
• Employs Intel Software Guard eXtension (SGX) processors for secure processing of sensitive code and data.
• Incorporates logarithm transformation, non-negative matrix factorization for dimensionality reduction, and parallel k-means clustering within a secure private cloud.

Main Results:

• Demonstrates superior efficacy in preserving patient privacy compared to state-of-the-art methods.
• Achieves a minimum of 7% higher Adjusted Rand Index (ARI) for clustering quality, dependent on dataset size.
• Exhibits significant efficiency, with computation times under 10 seconds for large datasets, even with privacy measures enabled.

Conclusions:

• The proposed approach effectively balances privacy preservation with high-quality clustering for scRNA-seq data.
• It offers a computationally efficient and secure solution for analyzing sensitive genomic datasets.
• The method provides a robust framework for advancing privacy-preserving machine learning in bioinformatics.