Machine learning-assisted study on structure-property relationships of dyes: A review

Abstract

The structure, properties, and synthesis processes of dye molecules are governed by complex, multi-dimensional relationships. Traditional research methods face limitations: they struggle to process high-dimensional data, involve high experimental costs, and hit efficiency bottlenecks, making it hard to derive results quickly or conduct comprehensive analysis. Machine learning (ML) has emerged as a powerful solution, enabling computational models to learn structure-property relationships (SPRs) from existing data, (e.g., how molecular structure governs color, stability, or reactivity), and predict dye behavior using these patterns. This paper reviews cutting-edge ML applications in dye research, including SPR modeling for property prediction, data-driven design of novel dye molecules, and optimization of synthesis processes for improved yield or reduced reaction time. However, key challenges remain, such as inconsistent experimental data formats that hinder model training and the inherent “black-box” nature of many ML models, which limits interpretability and trust in predictions. Future progress will likely stem from integrating multimodal data (experimental, computational, and theoretical) to enhance model robustness and combining ML with fundamental chemical principles to improve reliability and interpretability. These advances promise to transform dye research from traditional trial-and-error approaches into a closed-loop, data-driven workflow where models iteratively guide experiments, accelerating discovery and optimization.