Interpretable prediction of aggregation-induced emission molecules based on graph neural networks

Area of Science:

• Materials Science
• Chemistry
• Artificial Intelligence

Background:

• Aggregation-induced emission (AIE) materials offer unique photophysical properties.
• Developing efficient methods for AIEgen identification and rational design remains a challenge.

Purpose of the Study:

• To develop an interpretable graph neural network (GNN) for accurate AIEgen identification.
• To establish a rational design framework for novel AIE materials based on identified characteristic functional groups.

Main Methods:

• Developed an interpretable graph neural network model.
• Achieved 96.4% accuracy in AIEgen identification.
• Utilized insights to propose virtual library strategies: self-fragment and donor-acceptor docking.

Main Results:

• Identified 24 characteristic functional groups crucial for AIE properties.
• Successfully predicted four novel AIEgens using the developed virtual library strategies.
• Experimental confirmation validated the predicted AIEgens.

Conclusions:

• The interpretable GNN provides a powerful tool for AIEgen discovery.
• The proposed virtual library strategies enable rational design of AIE materials.
• This work establishes a robust framework for advancing AIE material development.