Customizable Raman ML Models for Rapid Identification

Scholars Portal (Borealis Dataverse), American Chemical Society
Graphical abstract for paper (https://doi.org/10.1021/acs.analchem.2c02451)

Description

The Customizable Machine-Learning Models for Rapid Microplastic Identification Python code and associated data for training and deploying machine learning classifiers to identify microplastics from Raman spectra is a resource for environmental scientists and analytical chemists who use Raman microscopy for polymer identification in microplastic research.

About Raman ML Models for Microplastic Identification

This resource offers a set of Python-based machine learning models specifically designed for identifying microplastics via Raman microscopy. Researchers Benjamin Lei, Justine R. Bissonnette, Úna E. Hogan, Avery E. Bec, Xinyi Feng, and Rodney D. L. Smith developed these models to streamline and customize the spectral analysis workflow. The models help researchers accurately classify various polymer types from complex Raman spectra, addressing a critical challenge in microplastic characterization.

The models employ machine learning algorithms, including random forest and perceptron, to analyze Raman spectroscopic data. They accept raw Raman spectra as input and output identified polymer types. The associated code allows users to customize models based on their specific analytical needs and spectral libraries, enhancing flexibility in diverse research contexts. This approach aims to reduce manual interpretation time and improve consistency in microplastic identification.

Key Features of Raman ML Models

  • **Customizable Model Training:** Users can train and adapt machine learning models with their own spectral libraries and data, enhancing applicability across different instruments and sample types.
  • **Python Implementation:** The models are provided as Python code, allowing for integration into existing analytical workflows and scripting environments.
  • **Algorithm Variety:** Supports various machine learning algorithms, including random forest, for robust pattern recognition in Raman spectra.
  • **Rapid Identification:** Designed to accelerate the processing and identification of microplastic particles compared to manual spectral comparison.
  • **Microplastic-Specific Application:** Tailored for the unique challenges of polymer identification in environmental microplastic samples.

Development and Validation

Benjamin Lei and colleagues developed these machine learning models, detailing their methodology and validation in the peer-reviewed publication, Lei et al., 2022, *Customizable Machine-Learning Models for Rapid Microplastic Identification Using Raman Microscopy*, Analytical Chemistry. This research established the models’ capability to provide rapid and customizable identification of microplastics using Raman microscopy. The publication outlines the computational framework and experimental validation, demonstrating the efficacy of the machine learning approach in spectral analysis.

Access and Data Availability

The customizable machine learning models and associated data are accessible through the Borealis Dataverse repository. Researchers can access and download the Python code and supplementary files directly from the dataset’s persistent identifier page: borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/LSN0R0. The resource includes the necessary files for implementing and training the models for microplastic identification. Specific licensing information for the code is not explicitly detailed on the Borealis page but is typically covered by the repository’s data usage terms or embedded within the downloaded files.

Why This Resource Belongs on Plastiverse

This resource offers a critical tool for advancing analytical methods in microplastics research, aligning perfectly with Plastiverse’s mission to curate cutting-edge solutions. By providing machine learning algorithms specifically for Raman spectroscopy, it directly supports researchers in efficient and reproducible microplastic identification. The focus on customizable Python-based models enables flexible deployment across diverse laboratory settings, enhancing the accuracy and throughput of polymer characterization in environmental samples.