Non-Negative Matrix Factorization of SWATH DIA Data Improves Global Metabolite Identification

Liquid chromatography coupled to High-Resolution Mass Spectrometry (LC-HRMS) is the most widely used approach for the global detection of small molecules in biological samples (metabolomics). In complement to such MS1 data, structural identification of metabolites implies the acquisition of fragmentation spectra by performing tandem mass spectrometry (MS2) experiments. To achieve both global detection and identification in a single run, the recently introduced acquisition mode called Sequential Window Acquisition of all THeoretical fragment ions (SWATH-type) Data Independent Acquisition (DIA) alternates MS1 detection and MS2 analysis of large and continuous m/z windows. The resulting MS2 data, however, contain a mixture of fragment ions originating from different precursor ions. To deconvolve these data and reconstruct pure individual MS2 spectra, the few existing software rely on determining a peak shape for each precursor ion. Such a strategy, however, may fail to separate co-eluting compounds. Here, we show how sparse non-negative matrix factorization (NMF) can separate pure spectral components successfully. We developed an end-to-end workflow called DIA-NMF to process SWATH DIA files, identify the detected compounds, and showed that it out-performs the reference algorithms MS-DIAL and DecoMetDIA, especially in the case of low-intensity or co-eluting compounds. Importantly, the reconstructed spectra include all the MS1 and MS2 ions related to the sought compounds and thus provide enriched chemical information that facilitates interpretation and identification.