Researchers pioneer quantum-embedded graph neural networks for breakthrough drug molecule prediction

HEFEI, Aug. 11 (Xinhua) -- Researchers have recently achieved drug-molecule property prediction based on quantum-embedded graph neural network architecture, pioneering a new path for molecular analysis and drug development, the Science and Technology Daily reported on Monday.

In drug development, accurately predicting molecular properties is key to efficiently screening candidate drugs. Graph neural networks study drug molecules by treating atoms as "dots" and chemical bonds as "lines." While existing quantum algorithms can better process these dots, they struggle with the lines.

The research team innovatively designed a quantum-embedded graph neural network architecture, integrating quantum edge and quantum node embedding methods, enabling the quantum-level simultaneous processing of both atoms and chemical bonds for the first time.

This breakthrough significantly enhances the accuracy of molecular behavior predictions, thereby improving drug discovery efficiency. The team has validated the reliability of the quantum embedding approach on the Origin Wukong quantum computer, showing that its models maintain stable performance even under the constraints of current noisy quantum hardware.

The study was carried out by Hefei-based startup Origin Quantum in collaboration with the University of Science and Technology of China and the Institute of Artificial Intelligence of the Hefei Comprehensive National Science Center. The findings have been published in the Journal of Chemical Information and Modeling.

"If traditional graph neural networks are a telescope, our quantum-embedded architecture with quantum edge encoding is a microscope -- not only locating atoms but also clearly capturing chemical bond interactions, pushing drug development toward precision design," said Dou Menghan, the lead developer of the Origin Wukong software team.