Optimizing chemical information retrieval from in-field proton transfer reaction time-of-flight mass spectrometry breath analysis to discover compounds correlated to tetrachloroethylene concentration.

内容摘要

Proton transfer reaction time-of-flight mass spectrometry (PTR-TOFMS) is a powerful tool for real-time analysis of volatile organic compounds (VOCs), including tetrachloroethylene (PCE). However, data analysis of large datasets of PTR-TOFMS data can be challenging to optimally extract chemical information. To address this challenge, we present a MATLAB-based workflow that processes PTR-TOFMS data directly from raw HDF5 files and analyzes all measurement cycles (i.e., spectra) in an untargeted manner. The workflow integrates mass spectrum preprocessing, alignment, and feature selection to maximize signal-to-noise ratio (S/N) and improve analyte discovery. Preprocessing included averaging, smoothing, and baseline correction, followed by correlation optimized warping (COW) alignment adapted to one-dimensional mass spectral data to correct m/z peak shifting across all spectra and samples. We implemented a tile-based F-ratio analysis to the one-dimensional (1D) PTR-TOFMS spectra to discover analytes correlated with PCE concentration. Samples with the five highest and five lowest PCE concentrations were separated into two classes, and 1D tile-based Fisher-ratio (F-ratio) analysis was followed by a 95% confidence interval t-test to generate a statistically filtered hit list. Principal component analysis (PCA) was used to visualize class separation, and performance was quantitatively assessed using the degree of class separation (DCS). PCE was ranked as the top hit (F-ratio = 129), and 20 of 487 discovered analyte hits passed the statistical significance threshold. Receiver operating characteristic analysis yielded an area under the curve (AUC) of 0.98, indicating effective discrimination of PCE-correlated analytes. The DCS between high- and low-PCE classes increased from 1.9 to 5.1, representing nearly a three-fold improvement when PCA was applied to only the top 20 F-ratio hits. Overall, this study demonstrates a workflow incorporating preprocessing and alignment across all mass spectra and samples and applies a tile-based automated F-ratio calculation framework to one-dimensional PTR-TOFMS data, enabling robust untargeted detection and quantification of analytes correlated with PCE.