We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Dataset for the statistical confidence of microplastic particle identification via infrared and Raman spectra
Summary
This dataset accompanies a study developing a statistically rigorous method (multiview conformal prediction) for identifying microplastic particles from two simultaneously collected spectra (infrared and Raman), improving the accuracy and efficiency of plastic identification in environmental samples. Better identification methods are essential for reliable monitoring of microplastic pollution across different environments and regulatory contexts.
Research overview: Database matching, in which an unknown spectrum is compared to a reference library of known spectra, is commonly used for targeted analysis of microplastics (MPs, particles between 1-5000 micrometers in size) in environmental samples. A score of likeness between the unknown particle and reference library can be used to tabulate the potential chemical matches and determine if the highest-scoring potential matches are the correct identity of the unknown particle. However, the methods used to determine a correct match are often arbitrary or based on precedent, despite the scores of likeness being highly dependent on a number of variables involved in a data analysis routine. Moreover, there is debate on how to identify MP particles when more than one chemical method is used to obtain spectra. This study utilized multiview conformal prediction (MVCP) to determine the probability of the correct match being returned for an unknown particle using simultaneously-collected photothermal infrared (PTIR) and Raman spectra. These scores of likeness between the two spectra were treated as coordinates in a two-dimensional score space. The MVCP method was used to calculate a boundary, surrounding an area of the score space called the envelope, that only returned potential matches if they met a user-defined confidence guarantee. We find that MVCP is more efficient, meaning fewer potential matches were returned, while maintaining accuracy than its counterpart single-view CP methods (i.e., only one of the two types of spectra were used). We additionally find that MVCP is more robust than the single-view CP methods when one of the two spectra for an unknown particle are either difficult or unable to be identified. Finally, we demonstrate the application of MVCP to an environmental sample that had MP particles impacted on it. MP particles were successfully identified in the sample, and MVCP was used to identify areas of improvement in the overall data analysis routine.