Streamlining\nPhenotype Classification and Highlighting\nFeature Candidates: A Screening Method for Non-Targeted Ion Mobility\nSpectrometry-Mass Spectrometry (IMS-MS) Data

Nontargeted\nanalysis (NTA) is increasingly utilized for its ability\nto identify key molecular features beyond known targets in complex\nsamples. NTA is particularly advantageous in exploratory studies aimed\nat identifying phenotype-associated features or molecules able to\nclassify various sample types. However, implementing NTA involves\nextensive data analyses and labor-intensive annotations. To address\nthese limitations, we developed a rapid data screening capability\ncompatible with NTA data collected on a liquid chromatography, ion\nmobility spectrometry, and mass spectrometry (LC-IMS-MS) platform\nthat allows for sample classification while highlighting potential\nfeatures of interest. Specifically, this method aggregates the thousands\nof IMS-MS spectra collected across the LC space for each sample and\ncollapses the LC dimension, resulting in a single summed IMS-MS spectrum\nfor screening. The summed IMS-MS spectra are then analyzed with a\nbootstrapped Lasso technique to identify key regions or coordinates\nfor phenotype classification via support vector machines. Molecular\nannotations are then performed by examining the features present in\nthe selected coordinates, highlighting potential molecular candidates.\nTo demonstrate this summed IMS-MS screening approach, we applied it\nto clinical plasma lipidomic NTA data and exposomic NTA data from\nwater sites with varying contaminant levels. Distinguishing coordinates\nwere observed in both studies, enabling the evaluation of phenotypic\nmolecular annotations and resulting in screening models capable of\nclassifying samples with up to a 25% increase in accuracy compared\nto models using annotated data.