Secure multiparty computation for privacy-preserving drug discovery.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Quantitative structure-activity relationship (QSAR) and drug-target interaction (DTI) prediction are crucial in drug discovery.
• Inter-institutional collaboration can enhance QSAR and DTI prediction performance.
• Data privacy and intellectual property concerns hinder collaboration in drug discovery.

Purpose of the Study:

• To develop novel algorithms for privacy-preserving collaborative drug discovery.
• To enable high-quality collaboration without divulging sensitive drug-related information.

Main Methods:

• Developed two algorithms, QSARMPC and DTIMPC, utilizing secure multiparty computation (MPC).
• QSARMPC is a neural network model for scalable, privacy-preserving QSAR prediction.
• DTIMPC integrates heterogeneous network data for confidential DTI prediction.

Main Results:

• QSARMPC demonstrates good scalability and performance for large-scale QSAR prediction.
• DTIMPC shows significant performance improvement over baselines and predicts novel DTIs with literature support.
• Both algorithms exhibit feasible scalability for growing drug discovery data.

Conclusions:

• QSARMPC and DTIMPC offer practical solutions for privacy-preserving collaborative drug discovery.
• These tools facilitate advancements in drug discovery by overcoming data privacy barriers.