Anal Chim Acta. 2026 Oct 22;1420:345982. doi: 10.1016/j.aca.2026.345982. Epub 2026 Jul 16.
ABSTRACT
BACKGROUND: Untargeted LC-MS metabolomics converts chromatographic ion signals into feature tables used for downstream comparison, annotation, and biomarker discovery. However, widely used preprocessing workflows often return discordant feature lists and missingness patterns from the same raw data, limiting reproducible quantitative interpretation. Continuous wavelet transform (CWT)-based peak detection is central to several workflows, yet its scale-normalization convention was inherited from energy-preserving signal analysis rather than area-oriented chromatographic integration. The problem addressed here is whether CWT normalization itself creates scale-selection and integration-boundary bias in LC-MS feature-table construction.
RESULTS: Under a Gaussian reference peak model with a Mexican-hat wavelet, amplitude-preserving 1/a normalization produced a defined optimum at a = √2σ and model-derived integration boundaries at ±3σ, supporting area-oriented peak integration. We implemented this correction in MetaboQuality with shape-driven grouping and anchor-guided recovery, then evaluated it using authentic standards, pooled-QC replicates, public QC data, component ablation, comparator sensitivity, decoy controls, non-Gaussian simulations, and chemical-reference validation. In the primary pooled-QC benchmark before post-detection filling, MetaboQuality produced 4029 complete groups with RSD <30%, compared with 2461 for XCMS and 1672 for MZmine 4. At the stricter RSD <10% threshold, MetaboQuality yielded 931 complete groups before filling versus 526 for the XCMS reference branch; after filling, MetaboQuality yielded 973 groups, compared with 678 for XCMS and 748 in the XCMS fitgauss sensitivity run.
SIGNIFICANCE AND NOVELTY: The novelty lies in defining CWT normalization as an LC-MS-specific peak-integration determinant rather than a generic signal-processing detail. MetaboQuality links amplitude-preserving integration, shape-driven grouping, and anchor-guided recovery into a coordinated workflow, providing a principled route to more complete and reproducible feature tables without relying on unconstrained statistical imputation. This clarifies where algorithmic design can improve measurement consistency.
PMID:42648851 | DOI:10.1016/j.aca.2026.345982
