Flimma

Flimma implements federated learning to perform privacy-aware differential expression analysis of RNA-seq transcriptomics data using the limma voom workflow.

Key Features:

• Federated learning framework: Performs computations across multiple sites while keeping patient-level data local to each contributing institution.
• Privacy preservation: Shares only model updates or aggregated statistics rather than raw data to support compliance with privacy regulations such as GDPR.
• Integration with limma voom: Builds on the limma voom methodology for RNA-seq differential expression, preserving voom-based linear modelling approaches.
• Robustness with imbalanced cohorts: Produces results consistent with limma voom applied to aggregated datasets even when class distributions differ across cohorts.
• Increased sensitivity via aggregation: Aggregates transcriptomics signals across sites to increase sensitivity and robustness of differential expression detection.

Scientific Applications:

• Multi-center transcriptomic studies: Enables collaborative differential expression analysis across hospitals and research centers without sharing raw patient data.
• Oncology: Facilitates large-scale RNA-seq analyses to identify cancer-related expression changes and potential therapeutic targets.
• Genomics: Supports integration of multi-site transcriptomic datasets for gene expression studies and comparative analyses.
• Personalized medicine: Enables discovery of patient-stratified expression signatures across cohorts to inform precision approaches.
• Biomarker and mechanism discovery: Supports identification of biomarkers, therapeutic targets, and investigation of disease mechanisms from multi-site transcriptomics.

Methodology:

Implements the limma voom workflow using the voom transformation to model the mean–variance relationship in RNA-seq count data, applies linear models for differential expression, and integrates these steps within a federated framework that exchanges model updates or aggregated statistics rather than raw data.