Federated unsupervised random forest for privacy-preserving patient stratification

Area of Science:

• Computational biology and bioinformatics
• Machine learning in healthcare
• Precision medicine methodologies

Background:

• Effective patient stratification and disease subtyping are crucial for precision medicine, requiring advanced methods for multi-omics data analysis.
• Clinical datasets are often small and fragmented across institutions, posing challenges for big data approaches due to privacy concerns.
• Machine learning techniques are vital for medical advancements but are hindered by data sharing limitations.

Purpose of the Study:

• To develop an innovative framework for advancing precision medicine using unsupervised clustering on multi-omics data.
• To address privacy concerns in medical data sharing through the integration of federated computing.
• To enhance patient subgroup identification and disease subtyping capabilities.

Main Methods:

• A novel multi-omics clustering approach employing unsupervised random forests.
• Federated execution of the random forest methodology to ensure data privacy.
• Validation on benchmark machine learning datasets and The Cancer Genome Atlas (TCGA) cancer data.

Main Results:

• The unsupervised random forest enables identification of cluster-specific feature importance, revealing key molecular drivers of patient groups.
• The federated approach is competitive with state-of-the-art methods for disease subtyping.
• The methodology significantly enhances the interpretability of identified patient clusters.
• Federated computing demonstrates potential for improving local clustering performance.

Conclusions:

• The proposed framework offers a powerful tool for precision medicine by enabling robust patient stratification and disease subtyping.
• The integration of unsupervised random forests and federated computing provides a privacy-preserving solution for multi-omics data analysis.
• The R-package facilitates the application of these advanced clustering techniques in research and clinical settings.