A privacy-preserving approach for cloud-based protein fold recognition.

Area of Science:

• Computational biology
• Machine learning
• Cryptography

Background:

• Cloud-based machine learning as a service (MLaaS) offers accessibility but raises privacy concerns, particularly in sensitive medical applications like protein fold recognition.
• Protecting both data (protein sequences) and models is crucial for secure MLaaS in healthcare.

Purpose of the Study:

• To propose a novel MLaaS solution for protein fold recognition that ensures privacy using secure three-party computation.
• To develop efficient private computational building blocks for complex machine learning operations.

Main Methods:

• Implemented a secure three-party computation framework for MLaaS.
• Developed private building blocks for essential operations (addition, multiplication, etc.).
• Demonstrated the approach using a privacy-preserving recurrent kernel network (RKN).

Main Results:

• The privacy-preserving RKN achieved performance comparable to non-private models.
• The solution demonstrated linear scalability with RKN parameters, indicating practical viability.
• The developed private building blocks efficiently support complex computations.

Conclusions:

• The proposed MLaaS solution effectively preserves privacy in protein fold recognition.
• This framework is adaptable for other medical machine learning tasks, enhancing secure MLaaS adoption.
• The solution offers a viable path for privacy-preserving computational biology research and deployment.