AFEI: adaptive optimized vertical federated learning for heterogeneous multi-omics data integration.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in healthcare

Background:

• Vertical federated learning (VFL) enables secure data collaboration across institutions, crucial for privacy-sensitive fields like healthcare.
• Integrating multi-omics data improves cancer prognosis prediction but faces challenges in data collection and privacy compliance (e.g., EU GDPR).
• Existing methods struggle to effectively integrate heterogeneous multi-omics data from multiple institutions while preserving patient privacy.

Purpose of the Study:

• To propose an Adaptive Optimized Vertical Federated Learning framework (AFEI) for integrating heterogeneous multi-omics data from multiple institutions.
• To enable accurate cancer prognosis prediction by leveraging distributed and encrypted multi-omics features.
• To address privacy and security concerns in multi-institutional data sharing for medical research.

Main Methods:

• Developed an adaptive optimized vertical federated learning framework (AFEI).
• Utilized distributed and encrypted multi-omics features shared by participating institutions.
• Built a joint evaluation model for cancer prognosis prediction using integrated data.

Main Results:

• AFEI achieved an average of 6.5% higher prediction accuracy compared to single omics data.
• The framework's performance closely matched direct integration of multi-omics data, demonstrating its effectiveness.
• Validated the ability to integrate encrypted multi-omics data from different institutions for improved predictive power.

Conclusions:

• AFEI provides an efficient solution for overcoming barriers in multi-institutional collaboration for cancer prognosis.
• The framework enhances cancer prognosis prediction by securely integrating heterogeneous multi-omics data.
• AFEI promotes the development of data-driven cancer research while upholding strict data privacy regulations.