dsMTL: a computational framework for privacy-preserving, distributed multi-task machine learning

Area of Science:

• Computational biology
• Machine learning
• Bioinformatics

Background:

• Differentiating reproducible and cohort-specific effects is crucial in multi-cohort machine learning.
• Multi-task learning (MTL) enables simultaneous learning across cohorts for effect differentiation.
• Analyzing geographically distributed data necessitates privacy-preserving, federated approaches.

Purpose of the Study:

• To develop a computational framework, dsMTL, for privacy-preserving, distributed multi-task machine learning.
• To enable the analysis of multi-cohort data that cannot be centrally stored.
• To facilitate the differentiation of reproducible and cohort-specific effects in distributed datasets.

Main Methods:

• Developed dsMTL, a framework with supervised and unsupervised algorithms for distributed MTL.
• Derived theoretical properties and machine learning workflows for software validation.
• Implemented dsMTL as an R package utilizing the DataSHIELD platform for federated analysis of sensitive data.

Main Results:

• Demonstrated dsMTL's applicability in distributed comorbidity modeling.
• Showcased that dsMTL outperformed conventional federated machine learning and aggregated individual models.
• Confirmed dsMTL's computational efficiency and scalability with real-world expression data.

Conclusions:

• dsMTL provides a robust solution for privacy-preserving, distributed multi-task learning.
• The framework enhances the analysis of multi-cohort, geographically dispersed datasets.
• dsMTL offers a significant advancement for federated learning in biomedical research.