Image-to-molecule benchmarking dataset with fractal pattern and hierarchical morphology recognition

Area of Science:

• Materials Science
• Chemistry
• Data Science

Background:

• Quaternary phosphonium salts (QPSs) exhibit significant morphological diversity.
• This diversity is linked to subtle variations in their molecular structure, specifically differing by a single methylene group.
• Understanding this molecule-morphology relationship is crucial for materials design.

Purpose of the Study:

• To document and analyze the morphological diversity of 19 homologous QPSs using advanced microscopy techniques.
• To create open-access datasets of microscopy images for QPSs.
• To establish a benchmark for machine learning applications in bridging the molecule-morphology gap.

Main Methods:

• Utilized scanning electron microscopy (SEM) and optical microscopy at various magnifications.
• Analyzed microstructures of crystallized QPS droplets.
• Collected and curated datasets of microscopy images for 19 homologous QPSs.

Main Results:

• Observed and documented high morphological diversity in QPS microstructures.
• Established a correlation between QPS molecular structure and observed microstructural patterns.
• Identified hierarchical patterns and fractal elements within the QPS microstructures.
• Datasets include microscopy images and corresponding molecular information.

Conclusions:

• The molecular structure of QPSs directly influences their crystallized droplet microstructures.
• The generated datasets are valuable for bidirectional machine learning: predicting molecular formulas from images and vice versa.
• This work provides a foundation for data-driven materials design by bridging the molecule-morphology gap.