Korea University Develops AI That Explains Why New Materials Produce Particular Colors

A new Korea University AI system can do more than predict the color a molecule will produce in a particular liquid. It can also indicate which parts of the molecule caused the color to change.

The research team, led by Professor Park Seong-nam of the university’s Department of Chemistry, developed the technology to reduce the time needed to create new materials. Such materials could be used in OLED smartphone displays, fluorescent substances for cancer diagnosis and solar cells. The findings appeared online Aug. 27 in the international journal Advanced Science.

OLED materials and fluorescent diagnostic substances emit light in solvents, but their color and brightness can change depending on the liquid around them. Earlier AI models could predict the result, but they could not explain why it occurred. Researchers therefore had to repeatedly synthesize materials and run experiments without knowing which molecular features were responsible.

“

This research directly incorporated chemical and physical insights into complex solvent effects into a deep-learning model, achieving both high prediction accuracy and explainability.

The new model, named Solvatochromic Subgroup Contribution, or SSC, breaks a molecule into functional groups and calculates how strongly each group affects the color. In the team’s example, an amine group produced major color changes as the solvent changed, while a methyl group had little effect. The model can express the contribution of each group numerically.

The researchers said SSC achieved higher prediction accuracy than existing AI models, and its explanations matched chemical principles that scientists have long verified. Before synthesizing a material, researchers could use the model to alter functional groups and search for a molecular structure that produces a desired color. The approach currently applies to ordinary organic molecules, but remains difficult to use with complex substances such as polymers and nanoparticles.

“

In the future, AI is expected to move beyond simply predicting molecular properties to understanding the chemical causes that determine those properties, and to use that understanding to design new molecules and materials with desired characteristics.