Quantum neural networks meet federated learning for DNA mutation prediction

Area of Science:

• Computational Biology
• Quantum Computing
• Bioinformatics

Background:

• DNA mutation prediction is crucial for understanding diseases.
• Genomic data analysis faces privacy and noise challenges.
• Quantum biology offers insights into mutation mechanisms.

Purpose of the Study:

• To introduce QuanGAT, a hybrid framework for privacy-preserving DNA mutation prediction.
• To simulate decentralized genomic environments while protecting sensitive data.
• To integrate quantum mechanics and graph attention networks for enhanced prediction.

Main Methods:

• Developed QuanGAT, combining quantum neural networks (QNNs), graph attention networks (GATs), and federated learning.
• Utilized parameterized quantum circuits with quantum noise channels (depolarizing, amplitude damping).
• Implemented differential privacy using an attention-based architecture with Laplace noise.

Main Results:

• QuanGAT demonstrated superior performance in DNA mutation prediction on PPI, STRING, and OBGN-Proteins datasets.
• Outperformed state-of-the-art graph neural networks by up to 4.5% in accuracy and 6.3% in macro F1 score in federated settings.
• Successfully simulated decentralized genomic environments with noise-aware, privacy-preserving capabilities.

Conclusions:

• Integrating QNNs with attention-based graph learning improves DNA mutation prediction.
• QuanGAT offers a robust solution for decentralized and privacy-aware genomic data analysis.
• The framework holds potential for advancing biomedical research in quantum-inspired computational biology.