FPLIER: Federated Pathway-Level Information Extractor

2026-05-28

AI Summary

This work addresses the challenge of limited data sharing in clinical transcriptomics due to privacy and governance constraints, which hampers the performance of pathway-aware factorization methods like PLIER. To overcome this, we propose FPLIER—the first framework integrating federated learning into pathway-level information extraction—enabling multi-institutional collaborative training without sharing local gene expression data beyond institutional boundaries, while also incorporating public datasets. FPLIER employs secure aggregation to achieve updates algebraically equivalent to centralized training. We further identify the rank of the training matrix as a critical factor influencing vulnerability to membership inference attacks and introduce a novel mechanism that enhances privacy by increasing matrix rank through dimensionality reduction or integration of public data. Experiments demonstrate that FPLIER converges stably in simulated federated settings and reduces attack success rates to near-random levels, substantially improving privacy guarantees.

Abstract

In transcriptomics, gene-set-aware factorization methods such as the Pathway Level Information Extractor (PLIER) are most effective when trained on large, heterogeneous expression compendia. Yet, many clinically relevant cohorts cannot be pooled into a single dataset due to privacy and governance constraints. We present FPLIER, a federated extension of PLIER that enables distributed training across multiple data holders while incorporating publicly available datasets. Through secure aggregation, FPLIER produces training updates algebraically equivalent to those of a centralized pooled-data approach while keeping expression data local. We evaluate FPLIER across multiple scenarios in two simulated consortia (from the K-CLIER and MultiPLIER studies) and demonstrate stable convergence. We further conduct a systematic analysis of membership inference attacks targeting both intermediate training statistics and the released model. Our results show that privacy risk is governed by the rank of the training expression matrix. Incorporating public data or reducing data dimensionality increases this rank, moving the system toward a full-rank regime in which training and non-training samples become indistinguishable to the attacker, and membership-inference performance approaches random guessing.

Problem

• federated learning
• transcriptomics
• privacy constraints
• gene-set-aware factorization
• data heterogeneity

Innovation

• federated learning
• transcriptomics
• privacy-preserving
• membership inference
• matrix rank