WT 2.0: Unveiling the “dark matter” in the metabolome using all-ion stepwise fragmentation acquisition mode (ASFAM) and deep learning-based feature extraction

Current metabolomics methods often miss low-abundance compounds and yield incomplete or ambiguous MS2 spectra, resulting in the presence of “dark matter” within the metabolome. Here, we introduce WT 2.0, which employs an all-ion stepwise fragmentation acquisition mode (ASFAM) to acquire comprehensive MS1 data, providing stepping 1 m/z windows for all precursor ions, and thereby capturing (theoretically) all deconvolution-free high-resolution MS2 data, as well as MS2 for low-abundance precursors that evade MS1 scans. A tailored algorithm extracts these low-response features, and a deep learning model predicts their high-resolution precursor ion m/z values. A Siamese network subsequently aligns features across samples. When applied to rice leaf LC-MS profiles, WT 2.0 increases detected features (with MS2) by 150% versus data-dependent acquisition (DDA) and 70% versus data-independent acquisition (DIA), and doubles high-confidence identifications compared to DDA while achieving a 360% increase over DIA. Under various stress conditions in rice leaves, WT 2.0 identifies 3-formylindole, β-carboline, and feruloyl putrescine as novel stress-responsive metabolites missed by conventional methods. In plasma LC-MS profiles, WT 2.0 reveals novel metabolic alterations in stage I lung cancer: increased free bile acids (cholic acid and deoxycholic acid) and decreased conjugated bile acids (glycocholic acid, taurocholic acid, and tauro-ursodeoxycholic acid) relative to benign lung disease. A panel of 20 plasma metabolites, featuring these bile acids, diagnoses LC with an AUC of 0.996. By integrating ASFAM, specialized feature extraction, predictive modeling, and advanced alignment, WT 2.0 significantly enhances detection and identification coverage, providing a powerful tool to expand metabolome analyses across diverse research applications.