Federated Learning in Genetics: Extended Analysis of Accuracy, Performance and Privacy Trade-offs

TL;DR

The paper tackles privacy-preserving collaborative learning on distributed transcriptomic data for precision medicine. It benchmarked two FL frameworks, TensorFlow Federated and Flower, on AML binary and MTG brain multi-class tasks using logistic regression and a sequential deep-learning model, across varying client counts, training rounds, and privacy-noise conditions. Key findings show deep learning generally outperforms logistic regression, FL can approach centralized performance with sufficient local data, and privacy-noise can significantly reduce accuracy with framework-specific robustness differences. The work demonstrates FL’s practicality for cross-institution genomic analyses while highlighting privacy-utility trade-offs and guiding future improvements in privacy, robustness, and personalization.

Abstract

Machine learning on large-scale genomic or transcriptomic data is important for many novel health applications. For example, precision medicine tailors medical treatments to patients on the basis of individual biomarkers, cellular and molecular states, etc. However, the data required is sensitive, voluminous, heterogeneous, and typically distributed across locations where dedicated machine learning hardware is not available. Due to privacy and regulatory reasons, it is also problematic to aggregate all data at a trusted third party. Federated learning is a promising solution to this dilemma, because it enables decentralized, collaborative machine learning without exchanging raw data. In this paper, we perform comparative experiments with the federated learning frameworks TensorFlow Federated and Flower. Our test case is the training of disease prognosis and cell type classification models. We train the models with distributed transcriptomic data, considering both data heterogeneity and architectural heterogeneity. We measure model quality, robustness against privacy-enhancing noise and computational performance. We evaluate the resource overhead of a federated system from both client and global perspectives and assess benefits and limitations. Each of the federated learning frameworks has different strengths. However, our experiments confirm that both frameworks can readily build models on transcriptomic data, without transferring personal raw data to a third party with abundant computational resources. This paper is the extended version of https://link.springer.com/chapter/10.1007/978-3-031-63772-8_26.

Figures (10)

• Figure 1: Overview of horizontal and vertical federated learning.
• Figure 2: Key challenges of federated learning.
• Figure 3: AUC with respect to an increasing training rounds and clients
• Figure 4: AUC with respect to an increasing class imbalance and training rounds over all imbalance configurations
• Figure 5: AUC with respect to an increasing local Gaussian noise