Enhancing Confidence in Microplastic Spectral Identification via Conformal Prediction

Area of Science:

• Environmental Science
• Analytical Chemistry
• Materials Science

Background:

• Microplastics are a global environmental concern, posing challenges for identification due to spectral similarities among common polymers.
• Current methods rely on the hit quality index (HQI) for spectral matching, but often lack associated confidence levels.

Purpose of the Study:

• To address the lack of confidence in microplastic spectral identification.
• To introduce a machine-learning framework, conformal prediction, for robust microplastic identification.

Main Methods:

• Applied conformal prediction to a machine-learning framework for spectral analysis.
• Utilized microplastic reference libraries (aged and pristine) and environmental spectra.
• Employed two similarity metrics to compute HQI and demonstrate the framework's benefits.

Main Results:

• Conformal prediction outputs a set of possible labels with user-defined probability, ensuring the true identity is included.
• The approach enhances confidence in spectral matching, reducing the need for manual inspection.
• Demonstrated an adaptable workflow with open-access code for the microplastic community.

Conclusions:

• Conformal prediction offers a robust method for confident microplastic identification.
• This framework improves the reliability of spectral matching and quantification in microplastic research.
• The open-access code facilitates wider adoption and enhances the robustness of microplastic analysis in environmental studies.