Liquid crystal-driven interfacial ordering of microplastics: Advancing microplastics characterization below the macro-scale
Data files
Oct 29, 2025 version files 444.95 MB
-
deepPolyNet_(1).ipynb
169.90 KB
-
GradCAM___High_Throughput_(1).ipynb
13.04 KB
-
Morphology_ML_Analysis_(1).ipynb
386.86 KB
-
Multi-label_classificaiton_(1).ipynb
4.72 KB
-
PS-PMMA_dataset.zip
444.37 MB
-
README.md
1.88 KB
Abstract
Global research efforts have developed methods for the detection and characterization of millimeter and larger plastic particles in the environment, but detecting and characterizing micrometer-scale and smaller plastic particles (microplastics or MPs) remains an unresolved challenge. MPs are difficult to characterize because they are small, have chemically heterogeneous surfaces that are transformed by environmental weathering, and are often accompanied by colloidal organic matter. Here, we advance the characterization of mixtures of MPs in the micrometer size range by leveraging their spontaneous adsorption and self-organization at liquid crystal (LC)-aqueous interfaces. We show that surface-sensitive interparticle interactions mediated by the LC can drive mixtures of colloidal MPs into unique assembly patterns that are accurately recognized using computer vision approaches. In particular, we show that we can identify MP composition (polystyrene and polymethyl methacrylate) in complex samples that contain colloidal natural organic matter and have been weathered using UV light. Additionally, we explore the basis by which the computer vision methods are able to classify MP samples, generating fresh insights into the physical processes by which colloidal dynamics and non-equilibrium interfacial phenomena influence the assembly of colloids at fluid interfaces. Overall, our results advance efforts 30 to develop characterization methods for colloidal-scale MPs that are broadly accessible (e.g., to citizen scientists).
Dryad DOI: https://doi.org/10.5061/dryad.qfttdz0vv
This repository contains code examples and four image/video datasets for microplastic assembly activity analysis.
- deepPolyNet_(1).ipynb
- GradCAM___High_Throughput_(1).ipynb
- Morphology_ML_Analysis_(1).ipynb
- Multi-label_classificaiton_(1).ipynb
- PS-PMMA_dataset.zip
PS-PMMA Project includes essential image datasets and code for analyzing PS-PMMA microplastic assembly patterns. Due to file size constraints, the complete image dataset is available via a separate Google Drive link. We have shown the example images in the repo. The dataset comprises 8,400 images distributed across 42 distinct classes, with 200 images per class. Each class is defined by a unique combination of four experimental conditions:
- Concentration: 0 (control, no microplastic), 20, 200, 400, and 800 mg/L
- Composition: PS, PS75%/PMMA25%, PS50%/PMMA50%, PS25%/PMMA75%, PMMA, and No MP
- UV Treatment Time: 0, 3, 5, and 10 days
- NOM Presence: Yes (NOM added) or No (no NOM)
The repository provides training code for both single-label and multi-label classification using the deepPolyNet model.
Additionally, it includes:
a. Grad-CAM++ feature importance analysis for model interpretation
b. Morphology-property feature analysis scripts, as illustrated and discussed in Figure 2 of the accompanying documentation or manuscript.
Citation: Please use the following citation: Mukherjee, Fiona; Shi, Anye; You, Fengqi; Abbott, Nicholas (2025). Liquid Crystal-Driven Interfacial Ordering of Microplastics: Advancing Microplastics Characterization Below the Macro-scale [Dataset]. Dryad. https://doi.org/10.5061/dryad.qfttdz0vv