This resource provides an open-source computational workflow for multi-instrumental nontargeted chemical fingerprinting. It helps researchers track microplastic pollution sources by classifying new and environmentally weathered plastics using machine learning, particularly useful for environmental litigation and regulatory monitoring efforts.

About ML Workflow: Weathered Plastic Chemical Source Tracking

The ‘Chemical Fingerprints of New vs Weathered Plastics: A Machine Learning Approach’ resource presents an open-source computational workflow to identify and track microplastic pollution sources. Developed by Huy Nguyen and colleagues, it addresses the challenge of chemically characterizing plastics and linking environmental samples to potential origins. Environmental scientists, regulators, and forensic investigators can use this workflow for localized source attribution and monitoring.

This workflow integrates multi-instrumental, nontargeted (organic and inorganic) data from three distinct mass spectrometry methods: automatic thermal desorption-gas chromatography–mass spectrometry (ATD-GC-MS), high-performance liquid chromatography quadrupole-time-of-flight mass spectrometry (HPLC-QToF-MS), and inductively coupled plasma-tandem mass spectrometry (ICP-MS/MS). It includes novel data imputation, normalization, and feature selection strategies with a Random Forest classifier. The workflow was evaluated using 42 pristine (store-bought) and 21 environmentally collected (weathered) plastic items from U.S. retail products and California beaches.

Key Features of ML Workflow: Weathered Plastic Chemical Source Tracking

• **Multi-instrumental Data Integration:** Combines chemical profiles from ATD-GC-MS, HPLC-QToF-MS, and ICP-MS/MS for comprehensive chemical fingerprinting.
• **Nontargeted Chemical Analysis:** Identifies a broad range of organic and inorganic chemical features, including plastic additives and sorbed environmental contaminants.
• **Machine Learning Classification:** Utilizes a Random Forest classifier alongside custom data imputation, normalization, and feature selection to distinguish plastic types and sources.
• **Validated Source Tracking Approach:** Achieved high classification accuracy (up to 1.00 MCC for ICP-MS/MS and 0.79 for HPLC-QToF-MS) with a “Suspect Source Comparison” method, simulating matching environmental plastics to a suspect source.
• **Open-Source Code:** All R scripts for data processing, statistical analyses, and machine learning are openly available for reproducibility and adaptation.
• **Real-World Sample Analysis:** Developed and tested with 63 real plastic items, including weathered samples collected from California beaches.

Development and Validation

This computational workflow was developed by a research team including Huy Nguyen, Eric Fries, David Jones, David Megson, Alexa Zonderman, Lars Hildebrandt, Scott Coffin, Daniel Pröfrock, and Roxana Suhring. They published their findings in the peer-reviewed journal *Environmental Science & Technology*: Nguyen et al., 2026, Environ. Sci. Technol.

The study validated two approaches for plastic source tracking. While a “Universal Fingerprint” approach showed low accuracy due to formulation variability, the “Suspect Source Comparison” approach demonstrated high classification accuracy, successfully linking environmentally sampled plastics to potential local sources. The methodology accounts for environmental weathering and surface-sorbed contaminants by comparing features shared between store-bought and environmental plastics. The detailed sample preparation (e.g., cutting samples to microplastic size, no surface cleaning for environmental samples) and multi-instrumental analysis protocols offer a robust framework, complementing discussions on best practices such as those outlined in the SCCWRP Workshop on Sample Collection Method for Microplastics in Drinking Water.

Access and Data Availability

The full scientific article, “Chemical Fingerprints of New vs Weathered Plastics: A Machine Learning Approach,” is available via its Digital Object Identifier (DOI) at pubs.acs.org/est. The article is licensed under CC-BY-NC-ND 4.0.

All R scripts for the computational workflow are openly accessible on GitHub at github.com/huymanhnguyen0811/Microplastic-Fingerprinting. This repository provides researchers with the code necessary to replicate and adapt the workflow for their own studies.

Community and Support

The availability of the R scripts on GitHub encourages community engagement, allowing researchers to explore the methodology and report issues.