Interpretable Multimodal Graph Learning Platform for Rational Design of AIEgens: From Molecular Structure and...

Area of Science:

• Materials Science
• Photophysics
• Computational Chemistry

Background:

• Aggregation-induced emission luminogens (AIEgens) have vast potential in materials science.
• Elucidating structure-property relationships in AIEgens is hindered by data dispersion and complex correlations.
• Traditional machine learning models lack interpretability for AIEgen data.

Purpose of the Study:

• To develop a data-driven, interpretable deep learning model (GATM) for predicting AIEgen properties.
• To decipher intricate relationships between molecular structures, solvent environments, and photophysical properties.
• To enable rational design and inverse design of functional AIEgens.

Main Methods:

• Constructed a multimodal predictive framework (GATM) integrating graph neural networks and machine learning.
• Utilized multisource data including molecular structures, photophysical parameters, and solvent environments.
• Employed graph attention networks (GAT) for visualizing solvent-solute interactions and analyzing feature importance.

Main Results:

• GATM achieved high predictive accuracy (mean R² > 0.90) for key AIEgen parameters.
• The model accurately predicted fluorescence lifetime, quantum yield, and spectral properties.
• Synthesized AIEgens demonstrated high accuracy (100%) in pesticide detection and discrimination with a low detection limit (0.4 nM).

Conclusions:

• The GATM model offers a novel paradigm for the rational design of AIEgens.
• The interpretable deep learning approach facilitates understanding of AIEgen luminescence mechanisms.
• This platform accelerates the development of new functional materials through intelligent prediction and inverse design.