MS-GDF: A Framework for Automated Identification of Stable Isotope-Derivatized Glycans through Deep Denoising and Two-Level Evidence Integration

Stable isotope derivatization improves the sensitivity and quantitative accuracy of glycomic analysis but also introduces additional spectral complexity, including isotope-pair signatures, multiple charge states, and redundant natural isotope peaks, which complicates automated glycan analysis. Here, we developed MS-GDF (mass spectrometry−glycan derivatization filter), an automated data-processing framework for stable isotope derivatization-based glycomics. MS-GDF integrates deep denoising with two-level evidence evaluation. The deep denoising module combines charge-aware MS-point filtering and isotope-cluster filtering to reduce redundant MS1 signals and generate candidate features. MS-GDF then supports glycan composition assignment by integrating MS1 stable isotope pairing with MS/MS glycone mass pool matching, without relying solely on database annotation. In a benchmark using human plasma N-glycans, MS-GDF reduced the number of MS1 features by 77.5% and 73.7% in the light- and heavy-labeled datasets, respectively. During manual review, 95.1% of Level 1 candidates were confirmed as valid light/heavy isotope pairs. Application to a clinical cohort comprising 124 human plasma samples enabled the assignment of 99 N-glycan compositions and revealed lung cancer-associated differences in N-glycan profiles. MS-GDF provides an automated, evidence-supported computational framework for stable isotope derivatization-based glycomics. The tool is available at https://github.com/XMU-Wulab/MS-GDF.