Collaborative analysis for drug discovery by federated learning on non-IID data

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and development
• Machine learning and artificial intelligence in healthcare

Background:

• Large-scale Quantitative Structure-Activity Relationship (QSAR) datasets are crucial for drug discovery.
• Centralized analysis of QSAR data faces privacy and security challenges.
• Federated learning (FL) enables collaborative model training without raw data sharing, but struggles with non-independent and identically distributed (non-IID) data.

Purpose of the Study:

• To propose a novel federated learning framework for collaborative drug discovery on non-IID datasets.
• To enable joint training of robust predictive models while preserving data privacy across multiple institutions.
• To overcome the limitations of FL in handling non-IID data in drug discovery.

Main Methods:

• Developed a federated learning framework for collaborative drug discovery.
• Addressed non-IID data challenges by globally sharing a small data subset among institutions.
• Leveraged FL to distribute model training across local devices, avoiding direct data exchange.

Main Results:

• The proposed framework achieved competitive predictive accuracy compared to centralized analysis on 15 benchmark datasets.
• The method successfully preserved the privacy of individual institutional data.
• Demonstrated benefits including reduced data transmission and enhanced scalability.

Conclusions:

• The novel federated learning framework effectively supports collaborative drug discovery on non-IID datasets.
• The approach balances predictive performance with essential data privacy requirements.
• The framework is suitable for large-scale, privacy-preserving drug discovery initiatives.