StrucGAP: a modular, streamlined and traceable data mining platform for structural and site-specific glycoproteomics

Abstract

The rapidly developing search engines for glycopeptide identification and accumulated high-resolution glycoproteomic data underscore the need for robust downstream data mining platforms towards subsequently functional and mechanistic studies. Here, we introduce StrucGAP, a Structural Glycoproteomics Analysis Platform for scalable downstream data mining of site-specific N-glycoproteomics. It integrates modules for data quality control, overall glycan structural characterization, altered glycan feature extraction, functional annotation, as well as upstream regulation and downstream networks. Its visualization and insight-tracking functionalities distill interpretation across hundreds of outputs, uniquely enabling to generate chart-based analysis reports and extract key glycosylation insights—capabilities rarely found in existing omics tools. Applying StrucGAP to an uncharacterized aging mouse uterus dataset reveals bidirectional regulation of core-fucosylation, and progressive, coordinated enrichment of glycans featuring sialylation via Neu5Ac, Lewis epitopes, and hybrid glycans along glycosylation dynamics. These changes are functionally linked to adhesion and remodeling, demonstrating StrucGAP’s ability to distill critical glycosylation insights from multi-dimensional information of structural N-glycoproteome datasets.

StrucGAP is a structural glycoproteomics analysis platform that distills multi-dimensional information of glycoproteome datasets into structural and functional insights towards mechanistic studies.

Introduction

Glycosylation refers to the enzymatic attachment of glycans to proteins, lipids, and recently reported GlycoRNAs^1,2. This critical post-translational modification on proteins significantly influences diverse biological processes such as protein folding, stability, cellular signaling, and immune system regulation. The intricate patterns of glycosylation are crucial for proper cellular functions and play significant roles in health and disease³. Mass spectrometry-based glycoproteomics has become a key technology for profiling protein glycosylation at large scale⁴, providing detailed insights into glycan structures, compositions, and attachment sites⁵. Tools like Byonic⁶, GPQuest⁷, MSFragger-Glyco⁸, O-pair Search⁹, StrucGP¹⁰, pGlyco3¹¹, Glyco-Decipher¹², and GlycanFinder¹³ help identify glycopeptides by assigning glycan compositions or structures to specific glycosites, with StrucGP uniquely enables de novo structural elucidation of N-glycans at individual glycosites, capturing diverse substructural features rather than limiting analysis to glycan compositions alone.

Despite advancements in upstream analytical tools, downstream analytical platforms for glycoproteomics data mining remain underdeveloped. In contrast, other omics fields benefit from mature analysis frameworks. For example, in single-cell transcriptomics, the toolkit Scanpy¹⁴ offers an end-to-end pipeline, from data preprocessing to the simulation of gene regulatory networks. Similarly, the tidyMass¹⁵ framework for mass spectrometry-based metabolomics integrates key analytical steps such as data preprocessing, metabolite annotation, and pathway enrichment. Even in the relatively nascent field of spatial transcriptomics, which has emerged more recently compared to other omics disciplines, a variety of specialized downstream analysis frameworks have rapidly developed to address diverse analytical needs. Tools like Squidpy¹⁶, Cell2location¹⁷, Starfysh¹⁸, and SCS¹⁹, each tailored to specific aspects of spatial transcriptomics. This underscores the need for robust downstream analysis frameworks in glycoproteomics to extract structural, quantitative, and biological insights from complex datasets.

To address this pressing need, we developed StrucGAP, a Structural and site-specific Glycoproteomics Analysis Platform designed for streamline downstream data mining towards functional elucidation of structural glycosylation features. Although glycosylation encompasses both N- and O-linked forms, the current implementation of StrucGAP is specifically tailored for N-linked glycosylation. StrucGAP integrates modules for data preprocessing, glycan structure elucidation, site-specific glycosylation interpretation, quantitative comparison, functional annotation, upstream regulation and downstream pathways with altered or pivotal glycans. In addition, StrucGAP uniquely supports two features rarely found in omics platforms: automated generation of chart-based analytical reports and extraction of key site-specific N-glycan insights, enabling rapid, interpretable summaries of complex datasets. By applying StrucGAP to an aging mouse uterus dataset and additional public datasets identified by the other five supported search engines, we demonstrated its broad applicability across multiple search engines and established it as an accessible community resource for high-resolution N-glycoproteomic analysis, enabling both exploratory research and hypothesis-driven investigations. StrucGAP is available at https://strucgap.readthedocs.io/en/latest/index.html and GitHub repository https://github.com/Sun-GlycoLab/StrucGAP.

Results

StrucGAP adopts modular architecture to supports scalable and reproducible workflow

StrucGAP is a modular computational framework for streamlined data mining of structural and site-specific N-glycoproteomic data. It enables comprehensive characterization of site-specific N-glycans across biomedical samples, encompassing their overall and altered structural features, and functional relevance. StrucGAP comprises eight interoperable modules: StrucGAP_Preprocess, StrucGAP_GlycoSite, StrucGAP_GlycanStructure, StrucGAP_GlycoPeptideQuant, StrucGAP_FunctionAnnotation, StrucGAP_GlycoNetwork, StrucGAP_DataVisualization, and StrucGAP_InsightTracker (Fig. 1), each for specific tasks.

Fig. 1. Overview of the StrucGAP platform.

Schematic illustration of StrucGAP, a modular, scalable, and traceable data mining framework for structural and site-specific glycoproteomics. The platform integrates eight interconnected modules for data preprocessing, glycosylation site profiling, glycan structure interpretation, glycopeptide quantification, functional annotation, glycosylation-related network exploration, data visualization, and insight tracking. StrucGAP ensures standardized data input, comprehensive structural resolution, functional interpretation, and reproducible analysis across diverse glycoproteomic datasets.

Central to StrucGAP’s design is its modular and object-oriented architecture, which promotes applicability, reproducibility, and customizability. Each module accepts a unified input format and can be selectively deployed or reordered based on research needs. StrucGAP currently supports input from StrucGP¹⁰, MSFragger-Glyco⁸, pGlyco3¹¹, Glyco-Decipher¹², Byonic⁶, and GlycanFinder¹³. StrucGAP provides a versatile, user-defined analytical platform, empowering researchers to systematically dissect N-glycoproteomic complexity and derive biologically meaningful insights from comprehensive datasets. Beyond its modular architecture, StrucGAP introduces methodological innovations that go beyond conventional workflows. Instead of treating glycans as indivisible entities, it decomposes them into biologically meaningful substructural motifs (e.g., Lewis epitopes, Neu5Ac/ Neu5Gc sialylation types, LacdiNAc motifs), enabling substructure-centric and systems-level network investigation of glycosylation regulation. Proportion-based quantification ensures that even rare but biologically significant motifs are detected when enriched in regulated populations, while threshold-variation analysis identifies robust, monotonic features across different cutoffs. In addition, StrucGAP not only performs functional enrichment but, more distinctively, directly maps enriched terms to site-specific N-glycan substructural features. This enables the dissection of which glycan substructures potentially drive the biological functions represented by the enriched terms, thereby extending functional interpretation from the pathway to the glycan-structural level. These strategies minimize the influence of uncertain assignments and ensure robust downstream trends. In addition, by linking substructures to upstream enzymes, glycan-binding proteins, and downstream pathways, StrucGAP provides a comprehensive framework uniquely suited for both structural and further functional N-glycoproteomics.

StrucGAP enables robust data quality control for accurate and unified inputs

StrucGAP implements stringent data quality control via StrucGAP_Preprocess module to ensure robust data mining and standardized inputs (Fig. 2a). This module currently supports inputs from StrucGP¹⁰, MSFragger-Glyco⁸, pGlyco3¹¹, Glyco-Decipher¹², Byonic⁶, and GlycanFinder¹³. In addition to StrucGP, pGlyco3 can also provide quantitative glycopeptide identifications with plausible structural annotations, therefore enabling analyses of comparable granularity. Other widely used search engines, including MSFragger-Glyco, Glyco-Decipher, Byonic, and GlycanFinder, primarily report glycan compositions rather than detailed structures. StrucGAP adapts flexibly to these varying input formats: when quantitative data are available (e.g., from MSFragger-Glyco or Glyco-Decipher), StrucGAP can perform quantitative analyses similar to those based on pGlyco3; when site information is missing (e.g., in MSFragger-Glyco), the site-specific module is simply bypassed; and when quantitative information is not provided (e.g., in Byonic and GlycanFinder), composition-level analyses remain fully supported. Information carried by search engine outputs and applicability across StrucGAP modules were presented in Supplementary Fig. 1.

Fig. 2. StrucGAP modules for data preprocessing, glycosylation site profiling, and site-specific N-glycan substructure analysis, demonstrated using a mouse uterus aging glycoproteomics dataset.

a Pipeline architecture of the StrucGAP_Preprocess module, encompassing steps for raw data cleaning, FDR filtering, outlier correction, CV filtering, PSM thresholding, and missing value imputation. b Identification statistics for unique IGPs, glycoproteins, peptides, glycosylation sites, glycan structures, and glycan compositions across the cleaned and FDR-filtered datasets. c Functional overview of the StrucGAP_GlycoSite module for profiling glycosylation at the site-specific level. d Quantification of glycosylation site distributions across glycoproteins, illustrating overall glycosite counts per glycoprotein. e Functional overview of the StrucGAP_GlycanStructure module, capturing key glycan substructural features. f Top 10 most abundant glycan structures ranked by the number of modified glycosites, shown with and without oligo-mannose glycans. g The number of glycan compositions with varying counts of structural isomers. h Distribution of core structures among all unique IGPs, with Core-I (N2H3), Core-II (core-fucosylated), Core-III (bisected), and Core-IV (dual features) quantified. i Classification of glycan types (complex, hybrid, and oligo-mannose) across all unique IGPs. j Relative abundance of 17 detected glycan branch structures and absolute counts for the eight least abundant structures, highlighting substructure diversity. k Branch count distribution among identified glycans, ranging from bi- to hexantennary forms, with frequencies shown per branch number category. l Characterization of fucosylation and sialylation types, including core and antenna fucosylation, Neu5Ac and Neu5Gc sialylation, and dual modifications. m Co-occurrence analysis from the StrucGAP_GlycanStructure module revealing preferential pairing between core and branch structures. Source data are provided as a Source Data file.

To balance clarity of workflow demonstration with breadth of validation, in the main text, we chose to apply the full StrucGAP analysis workflow to a single dataset. This design allows us to showcase the complete range of outputs, benchmark the key information extraction module, and demonstrate automated report generation in a coherent biological context. In addition, we also processed several publicly available datasets acquired with different search engines, detailed analysis reports and concise summaries have been compiled and are now included in the Supplementary Note 1 (see Supplementary Data 1–5 for details).

We applied StrucGAP to a N-glycoproteomic dataset generated from the uterine tissues of young (2-month-old, 2 M) and middle-aged (12-month-old, 12 M) female mice (n = 15 per group), representing early adulthood and reproductive senescence stages, respectively. Samples were combined into five pooled biological replicates per age group (3 mice per pool), followed by protein extraction, enzymatic digestion, and 10-plex TMT labeling. Ten TMT-labeled samples were pooled and subjected to glycopeptide enrichment and further fractionated into 24 fractions prior to LC-MS/MS. Using the StrucGP-based N-glycoproteomics pipeline, we identified 20,251 unique intact glycopeptides (IGPs) under a 1% FDR threshold (Supplementary Data 6). To ensure the reliability of downstream data mining, StrucGAP performs rigorous checks on data derived from StrucGP with TMT (or label-free) quantification for key substructural information, including core structures, glycan types, and branch structures/ numbers. StrucGAP also extracts critical glycan substructural features, including 82 LacdiNAc-related structures (56% LacdiNAc, 17% fucosylated LacdiNAc, and 26.8% sialylated LacdiNAc, based on their modified glycopeptide counts), 9,346 sialylated (36.8% Neu5Ac, 49.8% Neu5Gc, and 13.4% dual), 10,254 fucosylated (80.8% core-fucosylation, 7.9% antenna-fucosylation, and 11.2% dual), and fucosylated/sialylated glycans (19.5% F, 5% S, 8.6% G, 12% F + S, 14.4% F + G, 1.6% S + G, 4.6% F + S + G, 34.3% others; F: Fucose; S: Neu5Ac; G: Neu5Gc. Supplementary Data 6).

Following data cleaning, StrucGAP_Preprocess offers a multi-step statistical correction pipeline, including: (1) controlling false discovery rates (FDR) below 1% at peptide or glycan levels; (2) normalizing glycopeptide quantification data using global proteome scaling factors; (3) dual-stage coefficient of variation (CV) evaluation, first assessing CVs directly from raw TMT intensities to capture unbiased reproducibility, then re-evaluating CVs after preprocessed and aggregation to filter persistently variable features; (4) peptide-spectrum match (PSM)-based filtering to retain high-confidence glycopeptides; (5) glycan plausibility annotation, whereby each glycan composition and structure is mapped to GlyTouCan²⁰ identifiers, cross-checked against KEGG biosynthetic rules²¹, and further evaluated by rule-based filters derived from biosynthetic logic and chemical constraints, with user-selectable exclusion of implausible glycans (Fig. 2a). To further ensure robust differential analysis, StrucGAP recommends a data-driven fold change (FC) thresholding pipeline that integrates (i) CV-based theoretical minimal detectable fold change (MDFC), (ii) empirical null distributions from control–control comparisons FC, and (iii) empirical P-values derived from these nulls. This strategy defines FC cut-offs that are specific to dataset variability, ensuring that at least 95% of control–control comparisons are non-significant, and combines them with empirical significance testing to balance effect-size and statistical rigor (Fig. 2a).

In the mouse uterus aging dataset, for example, StrucGAP applied FDR control, reducing the dataset to 20,251 IGPs with high confidence from 73675 PSMs (FDR < 1%), with the resulting peptide- and glycan-level filtering outcomes (Fig. 2b). Normalization was performed using global proteome factors (Supplementary Data 6). CVs were first computed from raw TMT intensities and then re-computed after normalization. Median CVs decreased from 0.182 (IQR 0.145–0.232) to 0.115 (0.078–0.171) in the 2 M group, and from 0.154 (0.118–0.201) to 0.118 (0.084–0.165) in the 12 M group. CV and PSM filtering were omitted to preserve data breadth. Glycans were annotated where possible, without excluding rare or unannotated species by plausibility filtering. This choice was deliberate, as StrucGAP inherently mitigates the risk of over-interpreting any single questionable glycan by decomposing structures into reproducible substructural units (e.g., core structures, branching patterns) and performing proportion-based rather than absolute-count analyses. Specifically, 78 of 415 glycan compositions could not be matched to GlyTouCan, and among 898 glycan structures, 391 were unannotated even though these glycans were new combinations of known core and branch structures (accounting for 12.28% of 20,251 identified IGPs). Importantly, all key glycans features extracted by StrucGAP and the main conclusions described in the manuscript remained unchanged after excluding rare glycans. Only one descriptive summary statistic was altered: in the mouse uterus aging dataset, the distribution of glycan types shifted from complex > hybrid > oligo-mannose to complex > oligo-mannose > hybrid. Researchers may, depending on their goals, choose to exclude unannotated or rare glycans to perform a more conservative analysis. Altogether, the use of Python-native DataFrame maximizes interoperability with external tools, expanding accessibility and integration across glycoproteomics workflows.

StrucGAP provides an overview map of site-specific and structural N-glycoproteome

StrucGAP integrates two modules, StrucGAP_GlycoSite and StrucGAP_GlycanStructure, to dissect structural and site-specific N-glycoproteomic data. The StrucGAP_GlycoSite module focuses on mapping N-glycosite information (Fig. 2c). In the mouse uterus, most N-glycoproteins carry a single N-glycosite (71.5%), while Lrp1 exhibits up to 33 glycosites (Fig. 2d). The glycosite Asn-800 on Collagen alpha-2(VI) chain (Col6a2) shows the highest structural diversity, accommodates 141 distinct N-glycan structures. Comprehensive details on glycosite distributions across glycoproteins and glycopeptides, including glycan counts, glycan compositions, and isomers per glycosite, are provided in Supplementary Data 7.

StrucGAP_GlycanStructure delineates the overall distributions of N-glycan structural features (Fig. 2e). The top 10 glycan structures (based on their modified glycosite counts) are usually presented to provide an overview of dominant glycan structures in experimental samples. In the mouse uterus, the oligo-mannose (Man5 – Man9) and complex types equally predominate. After excluding the widely abundant oligo-mannose glycans, complex glycans dominate (7/10), with 8 sialylated and all but one hybrid glycan core-fucosylated (Fig. 2f). The 898 N-glycan structures can be sorted into 415 glycan compositions, with nearly half compositions (47.2%) exhibiting at least two structural isoforms. For example, the N4H5F1G1 were identified the highest 11 structural isomers (Fig. 2g and Supplementary Fig. 2a). The annotated spectra for all 11 isomers of N4H5F1G1, as well as for 10 isomers of N5H5F1G1 and 9 isomers of N4H5F1S1 are presented in Supplementary Data 8. In addition, representative spectra of rare glycan compositions such as N2H10, N2H11, and N2H12 (which might be the precursor forms of N-glycans during their biosynthesis), as well as two highly fucosylated glycans with 5-6 fucoses, are provided in the Supplementary Data 9 and 10.

StrucGAP_GlycanStructure further resolves site-specific N-glycans into substructural features, including core structures, glycan types, branch structures, branch counts, etc (Supplementary Data 11). Among 20,251 IGPs identified from the mouse uterus, for example, 1) the typical core (N2H3, Core-I) and core-fucosylated (Core-II) dominated (50.2% and 41.0%, respectively), while bisected (Core-III, 3.1%) and dual (Core-IV, 5.7%) are less common (Fig. 2h). 2) Complex glycans predominate (43.6%), followed by hybrid (31.3%) and oligo-mannose (25.1%) (Fig. 2i). 3) Of 17 branch structures, LacNAc leads (34.3%), trailed by LacNAc with Neu5Gc (21.1%) or Neu5Ac (16.9%), and sole GlcNAc (15.3%), with others like Lewis^x/a (4.5%) less frequent (Fig. 2j and Supplementary Fig. 2b). 4) Most glycans are bi- (42.6%) or tri-antennary (41.6%) (Fig. 2k). 5) Core fucosylation is prevalent (40.9%), while sialylation is less frequent (Neu5Gc, 23.0%; Neu5Ac, 17.0%). Notably, dual sialylation with both Neu5Ac and Neu5Gc (S + G) was least common (1.5%) (Fig. 2l and Supplementary Fig. 2c).

StrucGAP_GlycanStructure also conducts two-dimensional analyses to reveal substructural co-occurrence preferences. For example, sialyl-Lewis^x/a appeared with Core-II in 73.0% of cases, Lewis^y/b with Core-I in 70.8%, and Neu5Ac-LacNAc with Core-II in 64.5% in the mouse uterus glycoproteome (Fig. 2m and Supplementary Data 11). Overall, StrucGAP provides an overview map of site-specific and structural N-glycoproteomic in experimental samples.

StrucGAP delivers structural features of altered site-specific N-glycans

StrucGAP_GlycoPeptideQuant offers robust differential analysis, computing statistical indices including FC, area under curve (AUC), ANOVA, chi-square tests, and machine learning–based (random forest, XGBoost and PCA) feature importance (Fig. 3a), and supports customizable screening by integrating multiple statistical indices. At its core, this module adopts a substructure-centric, proportion-based strategy: instead of relying solely on absolute counts, StrucGAP evaluates the fraction of regulated features among all glycopeptides carrying the same motif, as well as their enrichment within upregulated or downregulated subsets. This dual proportional perspective ensures that rare but disproportionately regulated motifs are captured, while complementary threshold variation analysis (e.g., FC 1.2–3.0) identifies robust, cutoff-insensitive features. In practice, these principles are implemented through three complementary strategies, absolute abundance comparison, proportional enrichment among all carriers, and proportional analysis within upregulated or downregulated subsets, providing multiple, cross-validated perspectives on substructural regulation in site-specific N-glycans. These approaches generate robust differential features, minimizing the risk of biases and maximizing biological relevance.

Fig. 3. Quantitative profiling of site-specific glycans and substructural trends in aging uterus glycoproteomics via StrucGAP.

a Functional schematic of the StrucGAP_GlycoPeptideQuant module, which supports diverse statistical indices, flexible thresholding, and site-linked glycopeptide quantification. b Volcano plot of differentially expressed glycopeptides in the aging uterus dataset, showing significant up- and downregulated IGPs (FC > 1.7 or < 0.59, P-value < 0.05). Differential expression was assessed using a two-sided Student’s t test. c PCA plot based on glycopeptide expression profiles, demonstrating sample separation between age groups. d Heatmap of differentially expressed glycopeptides, revealing global expression trends and group-wise clustering. e Top 10 most frequently observed upregulated and downregulated glycan structures, excluding oligo-mannose types, highlighting distinct compositional shifts. f Absolute abundance comparison of glycan branch numbers in up- versus downregulated glycopeptides. g Proportional enrichment comparison of glycan branch numbers in up- versus downregulated glycopeptides. h Threshold variation analysis demonstrating dynamic enrichment trends for hybrid glycans, despite decreasing total counts. Source data are provided as a Source Data file.

Applied to the mouse uterus aging dataset (FC = 12 M/2 M), StrucGAP identified 1145 upregulated and 224 downregulated IGPs (P < 0.05; Fig. 3b–d). These calls were based on an FC threshold of > 1.7 or < 0.59, determined in a data-driven manner by integrating three independent estimates: the minimal detectable fold-change (MDFC, 1.75) from raw CV distributions, the global FC cut-off (1.66) from control–control empirical null distributions, and the empirical FC corresponding to P = 0.05 ( ≈ 1.66). Since the MDFC derived from raw intensities likely provides a conservative upper bound, whereas the null-based estimates from normalized data may be slightly permissive, we adopted an intermediate threshold of 1.7. This choice balances technical variability and statistical significance. Among the top 10 upregulated glycans (excluding oligo-mannose types), 90% were core-fucosylated, and 50% were sialylated, while downregulated species showed lower fucosylation (50%) and sialylation (20%). LacNAc structures were prominent (80%) in both (Fig. 3e). Notably, differential analysis of substructural features revealed that tetra-antennary glycans exhibited a balanced distribution with nearly equal numbers of up and downregulation in absolute abundance comparison, whereas most other substructures showed more upregulated IGPs (Fig. 3f and Supplementary Data 12). While bi-antennary glycans had the highest number of upregulated IGPs, proportional enrichment analysis reprioritized these features, placing six-branched glycans at the top and demoting bi-antennary structures to fifth, highlighting their relative enrichment among regulated glycopeptides (Fig. 3g). Besides, the threshold_variation_analysis() function jointly analyzes quantitative data with substructural features to identify critical substructures during statistical threshold variations. Specifically, it examines the selected metric, such as FC, across a series of thresholds (1.2, 1.5, 2, 2.5, 3), tracking shifts in the relative proportions of substructural features and the ratios of upregulated to downregulated IGPs as thresholds increase, thereby revealing dynamic trends in glycoproteomic changes. For example, the proportion of hybrid glycans rises from 29.7% to 37.5% despite decreasing counts, with the upregulated-to-downregulated ratio increasing from 1.65 to 16 (Fig. 3h and Supplementary Data 12). Meanwhile, site-specific analyses identified glycosite Asn-986 on glycoprotein Collagen alpha-4(VI) chain (Col6a4) as having the highest glycan variability (29 upregulated, 4 downregulated IGPs) despite modest protein-level change (FC = 1.31), indicating glycosylation-specific modulation (Supplementary Fig. 2d and Supplementary Data 12). In conclusion, StrucGAP effectively characterizes the dynamic alterations of glycan structures at the site-specific level.

Comprehensive functional annotation linking N-glycan features to biological functions

StrucGAP employs the StrucGAP_FunctionAnnotation module to provide functional interpretations of overall and altered glycan traits (Fig. 4a). This module supports over-representation analysis (ORA) and gene set enrichment analysis (GSEA) using either all glycoproteins from StrucGAP_Preprocess, differentially expressed glycopeptides from StrucGAP_GlycoPeptideQuant or global proteins from the StrucGAP_GlycoNetwork module. Annotation databases include Gene Ontology (GO; MF, CC, BP), Kyoto Encyclopedia of Genes and Genomes (KEGG), Reactome (REAC), WikiPathways (WP), transcription factors (TF), miRNA target interactions (MIRNA), Human Protein Atlas (HPA), CORUM complexes, and Human Phenotype Ontology (HP). Beyond standard enrichment, StrucGAP_FunctionAnnotation not only performs functional enrichment, but more distinctively maps GO and KEGG terms directly to site-specific N-glycan substructural features (e.g., core structures, glycan types, and branch structures), thereby elucidating their potential biological contributions and enabling the dissection of which substructures drive the enriched functions.

Fig. 4. Comprehensive functional annotation linking glycan substructures to biological processes via StrucGAP_FunctionAnnotation.

a Functional overview of the StrucGAP_FunctionAnnotation module, supporting ORA and GSEA-based enrichment from multiple databases, and enabling correlation analysis between enriched terms and glycan substructural features. b Subcellular localization of identified glycoproteins based on GO:CC enrichment, and statistical significance was assessed using a hypergeometric test with multiple-testing correction. c–e Distribution of core structures (c), glycan types (d), and branch structures (e) across the top 10 GO:CC-enriched terms. f Oligo-mannose glycans dominate among glycan features enriched in glycosylation-related biological processes (GO:BP), including macromolecule and protein glycosylation. g Differential GO:MF enrichment of upregulated and downregulated glycopeptides (P-value < 0.05, FC > 1.7 or < 0.59), and statistical significance was assessed using a hypergeometric test with multiple-testing correction. Source data are provided as a Source Data file.

Applying ORA to the mouse uterus aging dataset, cellular component enrichment (GO:CC) of all identified glycoproteins revealed predominant localization to extracellular and membrane-associated compartments (Fig. 4b). Substructural distribution patterns showed that the typical core (Core-I) predominated broadly, except in the cell surface, where core-fucosylation (Core-II) prevails (Fig. 4c), potentially indicating a specialized role in cell surface signaling or adhesion during aging²². Complex glycan types were also dominant across compartments (Fig. 4d), hinting at their broader functional versatility across cellular contexts. Neu5Gc-modified LacNAc emerged as a prominent branch structure in extracellular locations (Fig. 4e). Focusing on special glycans, biological process enrichment analysis (GO:BP) of fucosylated glycopeptides revealed that antenna- and core-fucosylated species shared many enriched pathways, but also showed subtype-specific functions: antenna fucosylation associated with adhesion and motility, while core fucosylation linked to developmental and multicellular processes (Supplementary Fig. 3a). Neu5Ac- and Neu5Gc-sialylated glycopeptides also exhibited shared but divergent enrichment, pointing to distinct roles in cell signaling and regulatory processes (Supplementary Fig. 3b). Interestingly, despite their limited overall abundance (25.1%), oligo-mannose glycans accounted for 71.0% of glycans involved in glycosylation-related GO:BP terms, suggesting a specialized role in glycosylation machinery²³. (Fig. 4f). Additional associations between substructural features and GO terms are detailed in Supplementary Data 13.

Leveraging differential analysis threshold from StrucGAP_GlycoPeptideQuant (FC > 1.7 or < 0.59, P-value < 0.05), StrucGAP_FunctionAnnotation further probed molecular function (GO:MF) differences between upregulated and downregulated glycopeptides. Upregulated glycopeptides were enriched in binding, adhesion, and enzymatic activity, while downregulated glycopeptides were associated with extracellular matrix structural roles (Fig. 4g and Supplementary Data 14, 15). Further GO:CC-based substructure profiling across enriched compartments revealed distinct glycan feature distributions between up- and downregulated glycopeptides, reflecting their divergent biological roles in aging (Supplementary Fig. 4a–f).

StrucGAP probes upstream regulation/ interaction and downstream signaling pathways in glyco-related networks

Glycosylation is orchestrated by intricate upstream regulatory mechanisms involving glycosylation-related enzymes and glycan-binding proteins, subsequently influencing downstream biological processes. To elucidate these regulatory relationships within a broader molecular network context, StrucGAP_GlycoNetwork integrates enzyme expression profiles, glycan-binding protein interactions, proteome-level dynamics, and downstream pathway (Fig. 5a). To enable this multi-scale investigation, StrucGAP incorporates proteomic and phosphoproteomic pipeline, both encompassing outlier correction, missing value imputation, CV filtering, and normalization across sample and feature dimensions combined with glycosylation rate calculation to classify N-glycans features, joint analysis of glycosylation sites, and comparison of glycosylation and phosphorylation sites (Fig. 5a).

Fig. 5. StrucGAP_GlycoNetwork reveals upstream regulation and interactions as well as downstream signaling pathways shaping glycosylation in uterine aging.

a Functional schematic of the StrucGAP_GlycoNetwork module, integrating proteomics-based glycosylation rate calculations and quantification of glycosylation-related enzymes, glycan-binding proteins, and signaling pathways. b Identification of N-glycan features that were altered solely at the glycopeptide level, independent of protein-level changes, suggesting glycosylation-specific regulation. c Three glycopeptides exhibited downregulation despite upregulated parent protein expression, exemplifying discordant regulation between protein abundance and glycosylation. d Glycosite-level analysis enabled by StrucGAP_GlycoNetwork, exemplified by Itbg1 (Integrin beta-1), where glycosite Asn-406 exhibited bidirectional regulation. e–h Quantitative profiling of glycosylation-associated enzymes: (e) 40 altered glycosyltransferases (P-value < 0.05). Differential expression was assessed using a two-sided Student’s t test; (f) statistical significance was assessed using a two-sided Student’s t test (Edem2, P = 3.04 × 10⁻⁴; Hpse, P = 3.69 × 10⁻⁶). Data were generated using TMT10-plex labeling with n = 5 pooled biological replicates per group (2 M and 12 M). Box plots show the median (center line), interquartile range (25th–75th percentile) (box), and whiskers extending to the minimum and maximum values; (g) heatmap of eight identified sialyltransferases, and differential expression was assessed using a two-sided Student’s t test; (h) heatmap of six identified fucosyltransferases, and differential expression was assessed using a two-sided Student’s t test. i Expression patterns of four significantly altered glycan-binding proteins. Normalized quantification values were obtained from a TMT10-plex experiment with n = 5 pooled biological replicates per group. Violin plots represent the distribution of the data, with median values indicated by the central line; individual points denote biological replicates. No error bars or error bands are shown. j KEGG-based downstream signaling pathway enrichment of altered glycoproteins and proteins. k Spearman correlation analysis among upregulated sialyltransferases, sialic acid-binding proteins, and sialylated IGPs (left), and among upregulated fucosyltransferases, fucose-associated binding proteins, and fucosylated IGPs (right), suggesting coordinated regulatory relationships. Correlations were assessed using two-sided Spearman’s rank correlation analysis. Source data are provided as a Source Data file.

To determine whether N-glycan alterations reflect glycosylation-specific regulation or protein-level shifts, StrucGAP conducted integrated glycoproteomic–proteomic analysis. Among 8860 quantified proteins in the aging uterus, 152 were upregulated and 53 downregulated (FC > 1.5 or < 0.67, P-value < 0.05). StrucGAP identified 228 upregulated and 55 downregulated glycopeptides whose changes occurred independently of protein expression (Supplementary Fig. 5a, b and Fig. 5b). Notably, three glycopeptides showed downregulation despite corresponding protein upregulation (Fig. 5c and Supplementary Data 16). Several glycosites exhibited mixed glycan regulation despite stable protein levels. For example, Asn-406 on Integrin beta-1 (Itgb1, FC = 1.05) carried glycopeptides showing predominantly increased glycans (18 species, mean FC = 1.57) alongside others with decreased tendencies (7 species, mean FC = 0.66), underscoring glycosylation-specific dynamics uncoupled from protein expression. (Fig. 5d and Supplementary Fig. 5c). These pattern suggests that glycans on this protein may involve conversion or substitution, reflecting a potential shift in glycosylation states that could influence functional outcomes in aging. GO: CC enrichment revealed that differentially expressed glycoproteins primarily localized to extracellular or membrane compartments, while proteomic changes were more intracellular, underscoring the extracellular functional relevance of glycosylation (Supplementary Fig. 5d).

To unravel the upstream regulatory axes, StrucGAP profiled alterations in glycosylation-related enzymes, including glycosyltransferases, glycosidases, sialyltransferases, and fucosyltransferases, as well as glycan-binding proteins. Among 95 identified glycosyltransferases, 40 showed modest changes (FC 0.9–1.4) (Fig. 5e). Of 38 identified glycosidases, 2 were significantly upregulated: ER degradation-enhancing alpha-mannosidase-like protein 2 (Edem2) (FC > 1.5), involved in ER-associated degradation (ERAD)²⁴, and Heparanase (Hpse) (FC > 1.5), implicated in ECM remodeling²⁵ (Fig. 5f). Among 8 sialyltransferases identified, Lactosylceramide alpha-2,3-sialyltransferase (St3gal5) and CMP-N-acetylneuraminate-beta-galactosamide-alpha-2,3-sialyltransferase 2 (St3gal2) were significantly upregulated (FC > 1.5), while St3gal6 was downregulated (FC < 0.67), implicating that elevated sialylated glycopeptides may be driven by St3gal5, which transfers sialic acid from CMP-NeuAc to galactose termini of glycosphingolipids to form gangliosides like GM3²⁶. St3gal2, which catalyzes alpha-2,3 sialylation of Galβ1-3GalNAc-terminated glycoconjugates to produce gangliosides such as GD1a and GT1b²⁷, potentially enhancing glycan-mediated signaling in uterine aging (Fig. 5g). The Alpha-(1,6)-fucosyltransferase (Fut8) showed upregulation among 6 identified fucosyltransferases (Fig. 5h and Supplementary Data 16). For 184 glycan-binding proteins identified, Endomucin (Emcn) and Fucose mutarotase (Fuom) were significantly upregulated (FC > 1.5). Emcn encodes endothelial sialomucin, which modulates cell adhesion and extracellular matrix interactions²⁸. While Fuom participates in L-fucose metabolism and glycosylation²⁹. In contrast, immune-related receptors³⁰ Killer cell lectin-like receptor 4 (Klra4) and Killer cell lectin-like receptor 7 (Klra7) were significantly downregulated (FC < 0.67) (Fig. 5i). For downstream functional interpretation, StrucGAP supports signaling pathway enrichment analyses by integrating differential glycopeptides, proteins, and phosphopeptides through its StrucGAP_FunctionAnnotation module. By integrated analyses of altered glycoproteins and global proteins, the identified significant signaling pathways related to ECM–receptor interaction, focal adhesion, lysosome function, PI3K–Akt signaling, complement and coagulation cascades, cell adhesion molecules, protein digestion and absorption, and proteoglycans in cancer. Many of these pathways, particularly those associated with extracellular matrix remodeling and growth factor signaling, are functionally relevant to uterine aging³¹ (Fig. 5j).

To further elucidate the integrated regulatory network. StrucGAP offers correlation analysis. The upregulated sialyltransferases (St3gal5 and St3gal2), sialic acid-binding protein Emcn, and sialylated IGPs exhibited strong positive correlations (Spearman’s r = 0.7–1.0), indicating a tightly linked regulatory network that may underlie enhanced sialylation during uterine aging. Likewise, the upregulated fucosyltransferase Fut8, fucose-associated glycan-binding protein Fuom, and fucosylated IGPs also showed highly correlated expression patterns (Spearman’s r = 0.8–1.0), indicating coordinated glycosylation regulation (Fig. 5j). Collectively, StrucGAP provides an integrative platform linking upstream glycosylation-related enzymes and glycan-binding proteins to downstream signaling pathways mediated by site-specific glycosylation alterations. This comprehensive approach facilitates mechanistic exploration of glycosylation networks and enhances biological interpretation of glycoproteomic data.

StrucGAP offers versatile data visualization and interpretable insight tracking

StrucGAP provides a comprehensive framework for glycoproteomic data mining, incorporating the StrucGAP_DataVisualization to offer 39 customizable plotting options, including radar charts, bar plots, heatmaps, volcano plots, for both StrucGAP outputs and user-supplied data (Fig. 6a). It also features draw_glycan(), which visually renders glycan structures from glycan coding. Importantly, the module supports automated generation of PDF reports summarizing results and visualizations, as demonstrated across this study. Two types of automated PDF analysis reports are generated: comprehensive visualization summaries and key insight reports distilled by StrucGAP_InsightTracker, streamlining review and collaborative interpretation of glycoproteomics results (Supplementary Data 17–21; see Supplementary Note 2 for details).

Fig. 6. Advanced visualization and key insight extraction powered by StrucGAP_DataVisualization and StrucGAP_InsightTracker.

a Overview of the StrucGAP_DataVisualization module, offering 38 advanced plotting options and automated report generation for multi-format results. b Schematic of the StrucGAP_InsightTracker module, which enables traceable workflows and automated extraction of key site-specific N-glycan insights across modules. c Proportional increase of Core-II glycans among both upregulated and downregulated IGPs across ascending and descending FC thresholds, respectively. d GO:BP enrichment results for upregulated and downregulated Core-II-containing IGPs. e–h Dynamic enrichment of additional substructural features across increasing FC thresholds: (e) Lewis^y/b and sialyl-Lewis^x/a branch motifs; (f) Hybrid glycan types; (g) Dual fucosylation patterns; (h) Neu5Ac-modified glycans. i Co-occurrence patterns among rising substructural features categorized as: Modest variation – hybrid glycans with Lewis^y/b; Expected structural correlation –dual fucosylation co-occurring with Lewis^y/b and sialyl-Lewis^x/a; Strongest co-occurrence – Neu5Ac with dual fucosylation, and Core-II with hybrid or dual-fucosylated glycans. j Distribution patterns of upregulated IGPs carrying specific core structures, glycan types, or fucosylation types across significantly enriched GO:BP terms (FC > 1.7, P-value < 0.05). Source data are provided as a Source Data file.

Complementing this, the StrucGAP_InsightTracker module consolidates outputs from three prior modules and distills them into biologically meaningful patterns. This high-level synthesis of structural and functional data highlights the most significant glycosylation alterations and their potential biological implications, providing an immediate summary of glycoproteomic outcomes. Its primary role is to extract key site-specific N-glycan insights by combining quantitative glycopeptide results with variation analysis thresholds, functional annotations and upstream/ downstream relationship networks associated with altered glycans. By processing data streams from modules such as StrucGAP_GlycoPeptideQuant, StrucGAP_FunctionAnnotation, and StrucGAP_GlycoNetwork, the StrucGAP_InsightTracker module identifies recurring glycosylation trends. To ensure reproducibility, this module logs all analysis parameters, tracks executed modules, and records every input-output relationship, generating a complete provenance trail. These features ensure transparent, auditable, and fully reproducible workflows, even for non-specialist users, thus supporting rigorous glycoproteomic discovery and reporting (Fig. 6b).

A. Extraction of key glycan features from StrucGAP_GlycoPeptideQuant

From StrucGAP_GlycoPeptideQuant, dynamic structural remodeling emerged in the aging uterus dataset. As the FC threshold for upregulated increased (1.2 to 3), the proportion of differential IGPs with Core-II structures rose steadily among upregulated IGPs (from 46.8% to 62.4%) (Fig. 6c). Similarly, among downregulated IGPs, Core-II frequency also increased as FC tightened (0.83 to 0.33) from 38.9% to 70.0% (Fig. 6c), potentially reflecting bidirectional regulation. Slicing the output of StrucGAP_GlycoPeptideQuant and integrating StrucGAP_FunctionAnnotation analyses revealed that upregulated Core-II glycans predominantly enrich in binding, while downregulated ones mapped to extracellular matrix-related terms, marking a shift from structural maintenance to adaptive signaling³² (Fig. 6d). This dual enrichment underscores Core-II’s bidirectional role in aging dynamics (Supplementary Data 22). Notably, at individual glycosites such as Asn-489 of Suppressor of tumorigenicity 14 protein homolog (St14), both upregulated and downregulated Core-II glycans were detected, differing subtly in antenna number, sialic acid species (Neu5Gc vs. Neu5Ac), and branch-level modifications. These structural variants, some tri-antennary with non-sialylated Lewis motifs, others bi-antennary with terminal sialylation, implicating microheterogeneities in bidirectional responses.

Also, across FC thresholds variation, multiple other substructural features also exhibited rising proportions: Lewis^y/b from 1.7% to 11.1% and sialyl-Lewis^x/a from 1.5% to 6.5% (Fig. 6e), hybrid glycan types from 29.7% to 37.5% (Fig. 6f), dual fucosylation from 13.8% to 38.9% (Fig. 6g), Neu5Ac-modified glycans from 37.4% to 47.7% (Fig. 6h). These trends likely reflect additive modifications such as Neu5Ac, in line with upregulated St3gal5 and St3gal2 expression.

StrucGAP_GlycanStructure co-occurrence analysis highlighted structural dependencies. Some pairs showed modest correlations (e.g., hybrid glycan types with Lewis^y/b from 0.7% to 5.1%, Core-II with Lewis^y/b from 2.4% to 14.0%), expected structural correlations (e.g., dual fucosylation with Lewis^y/b from 12.0% to 24.7%, and with sialyl-Lewis^x/a from 9.0% to 11.1%), and stronger associations: Neu5Ac with dual fucosylation soared from 10.9% to 56.3%, Core-II with hybrid glycan types rose bidirectionally (hybrid | Core-II: 48.4% to 66.7%, Core-II | hybrid: 30.9% to 41.0%,), and Core-II with dual fucosylation increased from 16.6% to 44.9% (Fig. 6i and Supplementary Data 22).

GO:BP enrichment of upregulated IGPs (FC > 1.7, P-value < 0.05) by StrucGAP_FunctionAnnotation containing these substructural feature pairs elucidated their roles in uterine aging: (1) Neu5Ac + dual fucosylation enriched in developmental and adhesion processes, suggesting roles in organogenesis and intercellular communication³³ (Supplementary Fig. 6a). (2) Core-II + hybrid glycan types enriched in morphogenesis and adhesion, indicating a role in structural remodeling and extracellular organization³⁴ (Supplementary Fig. 6b). (3) Core-II + dual fucosylation enriched in morphogenesis and adhesion, pointing to reinforced matrix interactions and developmental remodeling³⁵ (Supplementary Fig. 6c). No distinct site-specific bias emerged across these pairs. Collectively, these findings suggest that Neu5Ac with dual fucosylation, alongside Core-II with hybrid or dual fucosylation, synergistically enhance cell adhesion and matrix remodeling in uterine aging, likely driven by glycosyltransferase enzymes (St3gal5, St3gal2, and Fut8) and glycan-binding proteins (Emcn and Fuom).

B. Extraction of key information from StrucGAP_FunctionAnnotation

Beyond substructural and quantitative trends, the key_information_extraction() function in StrucGAP_InsightTracker also distills insights from StrucGAP_FunctionAnnotation. In the mouse uterus aging dataset, upregulated N-glycans (FC > 1.7, P-value < 0.05) showed differential distributions of substructures across pathways. When enrichment results were associated with core structures, Core-II exclusively dominated, averaging 74.6% across reproductive, collagen biosynthetic, metabolic and morphogenetic process (Fig. 6j).

Among glycan types, complex glycans were predominant (average 76.4%) in pathways related to gliogenesis, cell adhesion, and immune regulation. In contrast, oligo-mannose structures dominated acidification and lysosomal processes, averaging 84.5% (Fig. 6j). Core-fucosylated glycans topped terms associated with growth factor signaling, vascular development, and extracellular matrix organization (Fig. 6j). For sialylation, Neu5Gc modifications were enriched in immune and host-pathogen interaction processes, while Neu5Ac dominated in pathways associated with cell migration, neural development, and matrix organization, hinting at distinct sialic acid contributions in aging (Supplementary Fig. 6d and Supplementary Data 23). For downregulated N-glycans (FC < 0.59, P-value < 0.05), Core-I predominated across immune regulation, connective/cartilage development, and protein processing (average 75.8%) (Supplementary Fig. 6e). In sialylation, Neu5Ac prevailed across neurodevelopmental guidance and tissue morphogenesis (average 84.1%), while Neu5Gc dominated in processes tied to organismal downregulation, wound healing, and protein processing (average 83.3%) (Supplementary Fig. 6e and Supplementary Data 24). Other, less pronounced enrichment results have been fully retained in the comprehensive output files.

C. Extraction of key information from StrucGAP_GlycoNetwork

StrucGAP_InsightTracker utilizes the key_information_extraction() function to extract critical information from the StrucGAP_GlycoNetwork, organizing it into a relationship network. In this network, the upstream nodes represent glycosylation-related enzymes and glycan-binding proteins that regulate and interact with specific structural motifs or altered glycan features. The central nodes are the glycan substructural features, including up and downregulated LacNAc, GlcNAc, Lewis, LacdiNAc, and various forms of sialylation (Neu5Ac and Neu5Gc sialylation) and fucosylation (core and antenna fucosylation). The downstream nodes include associated signaling pathways.

For example, the network for upregulated sialylation, as shown in the Supplementary Fig. 6f, includes upstream enzymes like St3gal6 (negatively correlated) and glycan-binding proteins Siglec1 (positively correlated) and Siglec5 (negatively correlated). The corresponding downstream nodes, such as ECM-receptor interaction, are also mapped. Both networks, detailed in the supplementary materials, integrate changes in glycosylation features with their associated upstream and downstream factors, providing deeper biological insights and highlighting potential regulatory mechanisms.

Lastly, StrucGAP_InsightTracker supports reproducibility and collaboration by serializing key analysis variables into pickle files for rapid storage, sharing, and reinitialization of glycoproteomics workflows.

Discussion

To address the bottleneck of downstream data mining in glycoproteomics, we developed StrucGAP—a modular computational platform enabling in-depth mining of site-specific N-glycan data. Its design emphasizes applicability, interoperability, and reproducibility by adopting a unified Python-native input format and modular architecture. Together, these features enable researchers to dissect N-glycoproteomic complexity and derive biologically meaningful insights across diverse datasets. Importantly, no existing open-source or commercial dedicated bioinformatics tool exists that specifically addresses any single downstream analytical dimension, let alone a comprehensive, end-to-end platform. StrucGAP therefore fills a gap by delivering an integrated solution purpose-built for N-glycoproteomics, rather than a generic adaptation of proteomics pipelines.

Beyond architecture, StrucGAP embodies a substructure-oriented analytical philosophy aimed at minimizing the impact of misassigned glycopeptides, an enduring challenge in glycoproteomics. Instead of relying on glycan counts, StrucGAP decomposes glycans into reproducible substructural units and employs proportion-based analyses. This strategy reduces the influence of uncertain or rare assignments on downstream interpretations, ensuring that observed trends remain robust. At the same time, we encourage manual validation of unusual or rare glycan features, particularly highly isomeric structures guided by StrucGAP’s detailed annotation, by inspecting the corresponding MS/MS spectra for diagnostic fragment and oxonium ions, before pursuing further functional or mechanistic studies, and to avoid overinterpretation of weakly supported assignments. Complementing these principles, StrucGAP not only covers all critical aspects of downstream N-glycoproteomics data mining but also serve as a best-practices guide, thereby supporting the advancement of glycoproteomics research. Looking forward, StrucGAP’s insight-tracking and annotation capabilities offer an important interface with emerging AI tools. While large language models (LLMs) and foundation models have begun to transform genomics and transcriptomics (e.g., scGPT³⁶, nucleotide transformers), glycoproteomics remains underutilized in this space. By extracting high-confidence, site-specific features (e.g., co-modification patterns), StrucGAP generates structured, biologically relevant inputs for training intelligent models. These annotated datasets can help LLMs distinguish meaningful glycobiological features from background variation, accelerating the development of glycoproteomics.

While StrucGAP is capable of extracting key insights from N-glycoproteomics data based on the current resolution of upstream glycopeptide identification, we acknowledge that some complex glycan substructures, such as α2-3 vs α2-6 sialylation and H-type vs Lewis-type fucosylation, remain beyond the resolution of routine MS2 workflows and were therefore not addressed in the current work. These challenging substructures can increasingly be resolved through emerging technologies. Recent advances such as hybrid fragmentation strategies and higher-resolution mass spectrometers³⁷ have begun to enable stereochemical differentiation of sialic-acid linkages, and liquid chromatography-ion mobility-tandem mass spectrometry (LC-IM-MS/MS) provides an orthogonal dimension capable of separating α2-3 and α2-6 isomers based on differences in molecular shape³⁸. Glycopeptide co-elution, which can complicate isomer assignment in MS2, is also gradually addressed by improved chromatographic separation, ion mobility technology, and further optimization of upstream identification algorithms. In addition, the detailed structure features of altered glycans, including some indistinguishable structural features, could be further confirmed or validated based on the expressions of their corresponding glycosyltransferases obtained from the StrucGAP_GlycoNetwork analyses. Beyond these structural considerations, upstream tools also differ substantially in their compatibility with various quantification strategies, which further affects cross-engine comparisons. Broader compatibility across acquisition strategies and quantification modes, such as extending StrucGP to robust label-free quantification and enabling other engines to support high-precision TMT workflows, will further enable systematic comparisons of search-engine performance on the same LC–MS/MS datasets and, in turn, strengthen StrucGAP’s utility across platforms. As these technologies continue to evolve, StrucGAP is designed to adapt and integrate new capabilities, further enhancing its ability in providing comprehensive and biologically meaningful glycoproteome data mining for future functional and mechanistic studies.

In summary, StrucGAP represents an advancement in glycoproteomics and glycobiology fields by delivering an integrated and comprehensive platform for the downstream data mining of N-glycoproteomics data. Its modular design and unified input format make it a robust tool for elucidating the multifaceted roles of glycosylation in health and disease, while also paving the way for future innovations at the intersection of glycoproteomics and artificial intelligence.

Methods

Sample preparation and trypsin digestion

Uterine tissues were collected from female C57BL/6 mice at two age points (2 and 12 months). A total of 30 mice were used (n = 15 per age group). Due to limited tissue amounts, tissues from three mice were pooled to generate one biological replicate, resulting in five pooled biological replicates per group (n = 5 per group). No technical replicates were performed. Only female mice were used in this study sex was not included as a biological variable, and sex-disaggregated analysis was not applicable, as uterine tissue is sex-specific.

Tissues were cut into small fragments and washed three times with precooled phosphate-buffered saline (PBS, pH 7.4). Proteins were extracted using an 8 M urea-based lysis buffer containing ammonium bicarbonate (NH₄HCO₃) and protease inhibitor cocktail, followed by homogenization and sonication until the lysates became clear. After centrifugation at 15,000 × g for 15 min at room temperature, protein concentrations were determined using a BCA assay.

Proteins were reduced with dithiothreitol (DTT, 5 mM) at 37 °C for 1 h and alkylated with iodoacetamide (IAA, 15 mM) at room temperature in the dark for 30 min. Excess IAA was quenched by adding additional DTT (2.5 mM) for 10 min. Samples were diluted with deionized water to reduce urea concentration prior to enzymatic digestion. Proteins were digested with sequencing-grade modified trypsin at a protein-to-enzyme ratio of 100:1 (w/w) for 2 h at 37 °C, followed by a second digestion with the same enzyme ratio overnight at 37 °C.

Digestion was terminated by acidification with trifluoroacetic acid (TFA) to pH < 2. Peptides were centrifuged to remove insoluble material and desalted using C18 solid-phase extraction cartridges (cat. no. WAT054955, Waters, USA). Peptides were eluted with 60% acetonitrile (ACN)/0.1% TFA and dried under vacuum. Peptide concentrations were estimated by UV absorbance at 215 nm prior to downstream labeling.

Tandem mass tag (TMT) labeling

Equal amounts of tryptic peptides from each pooled biological replicate (five 2-month and five 12-month samples) were individually labeled with one channel of the TMT 10-plex Isobaric Label Reagent Set (cat. no. YA332834, Thermo Fisher Scientific, USA) according to the manufacturer’s instructions. Briefly, 200 μg of peptides per sample were dissolved in 100 mM HEPES (4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid) and incubated with TMT reagents reconstituted in anhydrous acetonitrile.

Young group samples were labeled with channels 126, 127 N, 127 C, 128 N, and 128 C, while middle-aged samples were labeled with channels 129 N, 129 C, 130 N, 130 C, and 131, respectively. Labeling reactions were carried out overnight at room temperature and quenched with 5 % hydroxylamine. All ten labeled samples were combined into a single multiplexed mixture and desalted again using C18 cartridges.

The pooled TMT-labeled peptide mixture was split into two portions: approximately 5% for global proteomic analysis and approximately 95% for intact glycopeptide enrichment.

Intact glycopeptide enrichment

Intact glycopeptides were enriched from the pooled TMT-labeled peptides using Oasis Mixed Anion-Exchange (MAX) cartridges (cat. no. 186000366, Waters, USA). Briefly, peptides were adjusted to 95% ACN and 1% TFA, loaded onto the MAX cartridges twice, and washed three times with the same solvent to remove non-glycosylated peptides. Glycopeptides were eluted with 50% ACN/0.1% formic acid (FA), dried under vacuum, and reconstituted in 0.1% FA for LC-MS/MS analysis.

High-pH reverse-phase fractionation

Both total peptides and enriched intact glycopeptides were fractionated by high-pH reverse-phase liquid chromatography prior to LC-MS/MS analysis to improve proteome and glycoproteome coverage. Samples were separated into 24 fractions, which were dried under vacuum and stored at − 80 °C until analysis.

LC-MS/MS analysis

LC-MS/MS analysis was performed on an Orbitrap Fusion Lumos mass spectrometer equipped with an Easy-nLCTM 1200 nano-LC system (Thermo Fisher Scientific, San Jose, CA, USA). Peptides were separated on a C18 reversed-phase analytical column at a flow rate of approximately 200 nL min⁻¹ using mobile phase A (0.1% FA in water) and mobile phase B (0.1% FA in 80% ACN).

For glycoproteomic analysis, MS1 spectra were acquired in the Orbitrap at a resolution of 60k over an m/z range of 350–1800. Automatic gain control (AGC) was set to 4 × 10⁵. Precursor ions with charge states of 2–6 were selected for fragmentation.

Two MS2 spectra were generated per precursor using higher-energy collision dissociation (HCD), one at a low collision energy (27%) for glycan structure analysis and the other at a high collision energy (40%) for peptide backbone fragmentation. MS2 spectra were acquired at a resolution of 50 k, with an isolation window of 2.0 m/z. Dynamic exclusion was enabled with a duration of 20 s to reduce repeated sequencing of the same precursor ions.

Intact glycopeptide identification and database search

Intact glycopeptides were identified using the StrucGP platform (v1.1) with its built-in 17-branch glycan structure library, in combination with the reviewed Mus musculus reference proteome database (UniProt, Proteome ID: UP000000589, downloaded on January 11, 2024). Database searches were performed assuming tryptic digestion with up to two missed cleavages.

Potential N-glycosylation sites were defined by the N-X-S/T consensus motif (X ≠ P). Precursor mass tolerance was set to 10 ppm, and fragment ion tolerance was set to 20 ppm.

Only spectra containing at least two diagnostic oxonium ions among the top 10 fragment ions were considered for intact glycopeptide identification. Identification results were filtered to achieve a 1 % false discovery rate (FDR) at both the peptide and glycan levels, using a target-decoy strategy. For each spectrum, the top-scoring peptide–glycan combination was selected as the final assignment.

Database search for proteome

All LC–MS/MS data were searched against the reviewed Mus musculus reference proteome database (UniProt, Proteome ID: UP000000589, downloaded on January 11, 2024). Peptides were identified using Proteome Discoverer v3.1 (Thermo Fisher Scientific, Germany). Search parameters included trypsin as the protease with up to two missed cleavages allowed, a precursor mass tolerance of 10 ppm, and a fragment mass tolerance of 0.02 Da. Carbamidomethylation of cysteine residues (C, + 57.021464 Da) and TMT 10-plex labeling on peptide N-termini ( + 229.162932 Da) were specified as static modifications. Oxidation of methionine residues (M, + 15.9949 Da), protein N-terminal acetylation ( + 42.010565 Da), and TMT 10-plex labeling on lysine residues (K, + 229.162932 Da) were set as variable modifications. Peptide- and protein-level identifications were filtered at a false discovery rate (FDR) of 1%.

Platform development

StrucGAP is a modular, streamlined, and traceable computational platform developed within the Python ecosystem for glycoproteomics analysis. The platform aims to provide comprehensive substructural details, quantitative insights, and functional relevance annotations for site-specific N-glycans, enabling the systematic dissection of glycosylation complexity. The design philosophy underpinning StrucGAP consists of three core principles: modularity, unified input format, and object-oriented design.

StrucGAP encompasses eight distinct modules, each corresponding to a specific analysis task in glycoproteomics. Each module is implemented as an independent Python class, encapsulating relevant analytical functions tailored to the module-specific tasks. The preprocessing module StrucGAP_Preprocess standardizes heterogeneous glycoproteomic datasets. By providing a unified, standardized data input format, this module ensures consistency and interoperability across the entire platform. To avoid reinventing the wheel and reduce the user learning curve, StrucGAP utilizes the commonly used Python data structure, Dataframe (implemented through Pandas), eliminating the need for creating new data structures. In addition, StrucGAP leverages key object-oriented concepts such as encapsulation and inheritance. Encapsulation is exemplified by the clear delineation of module responsibilities and the containment of module-specific methods within class definitions. Inheritance is implemented through shared functionalities abstracted into a base class (StrucGAP_InsightTracker Module), reducing redundancy and promoting consistency across modules, thus enabling flexibility and scalability within the analysis workflow.

Module ‘StrucGAP_Preprocess’

The StrucGAP_Preprocess module serves as the entry point of the StrucGAP platform, generating standardized, high-quality input data for all downstream modules. It accepts three primary user inputs: (1) structural and site-specific glycoproteomic identification results exported from StrucGP (TMT-labeled or label-free), (2) a sample grouping metadata file (this file defines the mapping between each TMT channel and its corresponding biological or experimental category), and (3) a user-defined glycan branch structure library (both file templates were provided in GitHub repository). Together, three inputs enable rigorous quality control and consistent formatting across datasets. In addition to StrucGP, StrucGAP_Preprocess also supports identification outputs from widely used N-glycoproteomics search engines, including MSFragger-Glyco⁸, pGlyco3¹¹, and Glyco-Decipher¹², Byonic⁶, and GlycanFinder¹³. Among them, pGlyco3¹¹ provides quantitative glycopeptide identifications with structural detail comparable to StrucGP, enabling analyses of nearly identical granularity. In contrast, MSFragger-Glyco⁸ primarily provides glycan composition information, whereas Glyco-Decipher¹², Byonic⁶, and GlycanFinder¹³ provide both glycan composition and glycosite assignment data. To facilitate input specification, users can define the source engine via the search_engine parameter (e.g., StrucGAP_Preprocess(search_engine = ‘StrucGP’)), enabling flexible integration of diverse upstream pipelines.

Beyond its initialization and output functions, StrucGAP_Preprocess implements nine core functions that constitute a complete preprocessing pipeline: data_cleaning(), cv_raw(), fdr(), outliers(), cv(), psm(), annotation(),data_imputation(), and fc_recommendation(). Each corresponds to a discrete quality control step or filtering strategy designed to enhance data reliability and interpretability.

To ensure downstream analytical reliability, the data_cleaning() function conducts structured validation of critical annotation fields, including glycan coding (the foundation for all structural features), glycan compositions, glycan types, branch numbers, and branch structures (validated against the uploaded branches library). After validation, substructural features such as core structures, LacdiNAc, sialylation type, fucosylation type, and fucosylated/sialylated patterns are extracted. In the case of label-free data, quantitative information is also retrieved. The fdr() function supports four user-defined filtering levels—no control, peptide-level, glycan-level, and both—providing customizable stringency for data confidence. The annotation() function extends this preprocessing by performing automated glycan plausibility checks. It integrates three layers of annotation: (i) mapping glycan compositions and structures to GlyTouCan identifiers, (ii) annotating potential biosynthetic pathways using the KEGG Glycan database, and (iii) applying rule-based plausibility filters derived from biosynthetic logic and chemical constraints. The filters automatically flag improbable patterns such as (a) incomplete cores (N < 2 or H < 3), (b) high-mannose structures carrying core and/or outer-arm fucoses, (c) excessive sialylation beyond antenna capacity (S > 2 estimated antennae), (d) abnormal fucosylation (F > 1 + estimated antennae + 2, with F ≥ 5), (e) rare antenna numbers ( ≥ 4), (f) oversized high-mannose glycans (H ≥ 10), (g) biologically implausible LacNAc concatenations on low-branch structures, and (h) high-mannose glycans containing bisected motifs, such as structures with three core GlcNAc residues (N3H6-10). Glycans flagged as rare or implausible can be optionally excluded by the user.

Quantitative data is handled by the cv_raw(), outliers(), cv(), psm(),data_imputation(), and fc_recommendation() functions. The cv_raw() function conducts a raw-data CV assessment, calculating coefficient of variation values for each glycopeptide directly from unnormalized TMT channel intensities across relevant replicates. This provides an unbiased measure of reproducibility prior to any normalization or aggregation, enabling users to evaluate the inherent stability of raw quantitative signals before downstream processing. The outliers() function corrects and normalizes TMT quantification data from matched reporter ions for each IGP. For an IGP with M PSMs, raw intensities across 10 TMT channels are scaled by global proteomic factors, normalized by the median intensity of control group channels per PSM, and merged across PSMs using a median_cheng() function:

$${\mathbf{I}}_{\mathrm{final},c}=\mathrm{m}\mathrm{edian}\_\mathrm{cheng}\left({\left\{\frac{{\mathbf{I}}_{\mathrm{r}\mathrm{aw},m,c}*{\mathbf{S}}_{\mathrm{g}\mathrm{lobal},c}}{\mathrm{m}\mathrm{edian}\_\mathrm{cheng}\left(\left\{{\mathbf{I}}_{\mathrm{raw},m,c^{\prime }}*{\mathbf{S}}_{\mathrm{global},c^{\prime }}\left.∣\right)c^{\prime }\in \mathrm{Control}\right\}\right)}\right\}}_{m=1}^M\right)$$ 1

Where I_raw,m,c is the raw intensity for PSM m, channel c, S_global,c is the global scaling factor, and median_cheng() function follows:

$$\mathrm{m}\mathrm{edian}\_\mathrm{cheng}\left({\left\{x_i\right\}}_{i=1}^N\right)=\left\{\begin{array}{c}\hfill \sqrt{x_{∣N/2∣}*x_{∣N/2∣+1}}\mathrm{if}N\mathrm{is}\mathrm{even}\mathrm{and}N>0,\hfill \\ \hfill x_{∣N/2∣+1}\mathrm{if}N\mathrm{is}\mathrm{odd}\mathrm{and}N>0,\hfill \\ \hfill \mathrm{N}\mathrm{aN}\mathrm{if}N=0.\hfill \end{array}\right)$$ 2

N as the number of valid (non-NaN) observations in the dataset (sample size). X_∣N/2∣ as the ∣N/2∣- th smallest value in the sorted (ascending, non-NaN) data array, indexed from 1. X_∣N/2∣+1 as the subsequent value in that sorted array.

The global scaling factor was derived from protein group quantitation results exported by Proteome Discoverer 3.0. For each TMT channel, the ratio of Normalized Abundance to Abundance was calculated across proteins, and the median of these ratios was taken as the scaling factor for that channel. This provided a robust adjustment that minimizes the influence of outliers and ensures consistency across channels.

The cv() function enables optional coefficient-of-variation filtering based on user-defined sample groupings. CV is computed independently within each group, and glycopeptides with persistently high variability after normalization will be flagged or excluded from downstream quantification. The psm() function filters IGPs by the minimum number of supporting PSMs (e.g., PSM ≥ 3), ensuring robust identification. Finally, the data_imputation() function addresses missing values using a k-nearest neighbors (KNNImputer) approach³⁹, selecting neighboring values to impute gaps and improve data usability.

The fc_recommendation() function provides a three-tier, data-driven framework for selecting FC thresholds and assessing the significance of glycopeptide regulation:

(i) CV-based theoretical minimal detectable fold-change (MDFC).

For each glycopeptide, the CV is first estimated from raw TMT channel intensities (prior to normalization). Using error-propagation principles, the log-scale variance is approximated as:

$${\sigma }_{\mathrm{l}\mathrm{n}}^2\approx \ln \left(1+{\mathrm{CV}}^2\right)$$ 3

where ln denotes the natural logarithm.

For two groups, each with n replicates, the variance of group mean difference is:

$$\mathrm{V}\mathrm{ar}\left({\mathrm{\Delta }}_{\ln }\right)\approx \frac{2{\sigma }_{\mathrm{l}\mathrm{n}}^2}{n}$$ 4

Given a chosen confidence level (e.g., z = 1.96 for 95%), the minimal detectable fold-change in the natural-log domain is:

$${\mathrm{\Delta }}_{\ln }^{\left(n\right)}=z\cdot \sqrt{\frac{2{\sigma }_{\mathrm{l}\mathrm{n}}^2}{n}}$$ 5

and the corresponding threshold in linear scale is:

$${\mathrm{M}\mathrm{DFC}}^{\left(n\right)}=e^{\left({\mathrm{\Delta }}_{\ln }^{\left(n\right)}\right)}$$ 6

This value represents the smallest FC that exceeds expected random variability given the dataset’s CV.

(ii) Empirical null from control–control comparisons.

To capture combined biological and technical variation, all pairwise log2 FCs are calculated between biological replicates within the control group.

The resulting set of absolute differences forms an empirical null distribution D₀.

The empirical FC threshold is then defined as the 95th percentile of D₀:

$${\tau }_{{\mathrm{l}\mathrm{og}}_2}={\mathrm{Q}\mathrm{uantile}}_{0.95}\left(D_0\right)$$ 7

and converted back to linear scale as:

$$\tau \mathrm{F}\mathrm{C}=2^{{\tau }_{{\mathrm{l}\mathrm{og}}_2}}$$ 8

This ensures that 95% of control–control comparisons are considered “non-significant”, providing a robust, dataset-specific cut-off.

Note: TMT quantification generally offers higher precision than label-free methods, permitting a moderately lower threshold without inflating false positives.

(iii) Coupling with empirical P-values.

For each glycopeptide, the observed group log₂FC is compared to the empirical null distribution D₀.

The empirical P-value is defined as the right-tail probability:

$$P=\frac{\mathrm{\#}\left\{d\in D_0:d\ge ∣{\log }_2\mathrm{FC}∣\right\}+1}{∣D_0∣+1}$$ 9

where |D₀| is the size of the null distribution and the symbol “#” denotes the cardinality operator, i.e., the number of elements in a set that satisfy a given condition.

This preprocessing pipeline outputs a unified, annotation-rich Dataframe object that serves as the standardized input for all downstream StrucGAP modules. By integrating validation, filtering, normalization, and imputation within a modular structure, StrucGAP_Preprocess ensures reproducible and high-confidence glycoproteomic data mining.

Module ‘StrucGAP_GlycanStructure’

The StrucGAP_GlycanStructure module enables comprehensive profiling of glycan substructural features, their co-occurrence relationships, and isomer variability across IGPs. This module can accept inputs from multiple preprocessing stages within StrucGAP_Preprocess (e.g., outliers_filtered_data, psm_filtered_data) or from differential expression outputs generated by StrucGAP_GlycoPeptideQuant, including user-specified thresholds for FC and P-value to define upregulated or downregulated glycopeptides. This flexible input design supports both global and condition-specific structural investigations. The core implementation includes four primary functions: statistics(), structure_statistics(), cor(), and isoforms().

The statistics() function provides a global overview of glycan structures in the dataset by summarizing distributions of glycan coding patterns and glycan compositions. An optional parameter allows users to exclude high-abundance oligo-mannose glycans to focus on structurally diverse and functionally relevant isomers. The structure_statistics() function performs independent quantification of all annotated glycan substructural attributes in the dataset, including glycan types, branch structures, branch counts, sialic acid and fucose counts, sialic acid and fucose structures, fucosylated types, sialylated types, and fucosylated/sialylated patterns. The cor() function computes pairwise co-occurrence frequencies of glycan substructural features across all IGPs to identify preferentially associated motifs. Such relationships can reveal structural dependencies or biosynthetic coupling events that underlie glycosylation pathway regulation. The isoforms() function investigates isomeric variants within identical glycan compositions, detecting structural diversity and potential functional implications.

Together, these functions provide multiscale characterization of glycan architecture, from individual substructure prevalence to inter-feature relationships and isomer diversity. The output is a feature-resolved structural profile that supports both exploratory analysis and hypothesis-driven investigations into glycosylation regulation.

Module ‘StrucGAP_GlycoSite’

The StrucGAP_GlycoSite module provides comprehensive analysis of N-glycosylation at the site-specific level, facilitating both protein-centric and peptide-centric evaluations of glycosylation topology and complexity. It enables users to dissect glycosylation heterogeneity across glycoproteins and to explore the micro-heterogeneity of individual glycosylation sites with high granularity. This module comprises three core analytical functions: glycoprotein_site(), glycopeptide_site(), and specific_site().

The glycoprotein_site() function summarizes glycosylation at the protein level, including the total number of glycosylation sites per glycoprotein, the number of distinct glycans observed at each site, and the diversity of glycan structures per site. It provides a macro-level overview of site occupancy and heterogeneity, allowing identification of hyper-glycosylated proteins or highly heterogeneous sites. Similarly, the glycopeptide_site() function performs a parallel analysis at the glycopeptide level, treating each unique glycopeptide as an analytical unit. It enables fine-resolution profiling of site localization and supports peptide-based comparisons. In contrast, the specific_site() function conducts in-depth analysis of individual glycosylation sites (referenced to either glycoproteins or glycopeptides, dissecting glycan counts, structures, compositions, and isomeric variants at each site.

This site-focused analysis is particularly useful for uncovering functionally relevant site-specific glycoforms, investigating site usage shifts under experimental conditions, or validating known or predicted glycosylation motifs. The outputs can be flexibly visualized or integrated with differential expression data to support downstream structural or functional interpretation.

Module ‘StrucGAP_GlycoPeptideQuant’

The StrucGAP_GlycoPeptideQuant module supports deep differential analysis of IGPs, leveraging a suite of customizable statistical indices. In addition to conventional FC and significance testing, the module integrates machine learning–derived feature importance and PCA–based indicators, enabling multi-perspective prioritization of glycopeptide features with functional or diagnostic potential. This module includes six primary analytical functions: statistics(), statistics_index(), differential_analysis(), threshold_variation_analysis(), glycopeptide_glycosite_glycan_variation(), and glycoprotein_glycosite_glycan_variation().

The statistics() function provides a comprehensive overview of glycoproteomic identification results, including the number and types of identified spectra, peptides, glycopeptides, and glycoproteins, as well as classification of glycopeptides and glycoproteins by structural features.

The statistics_index() function computes multiple indices for differential analysis, including FC with significance tests (P-values from t-tests and Mann-Whitney U tests), area under the ROC curve (AUC) with P-values, and feature importance scores from random forest and XGBoost⁴⁰. Other indices include ANOVA F-scores and chi-squared scores with corresponding P-values.

In addition, the function incorporates PCA-based feature importance, calculated from the weighted absolute loading matrix across the first two principal components. Specifically, for a matrix of glycopeptide features X, the scaled and centered input X’ is transformed via PCA:

$${\mathrm{I}\mathrm{mportance}}_i={\sum }_{k=1}^2∣L_{ik}*{\lambda }_k∣$$ 10

where L_ik is the loading of feature i on principal component k, and λ_k is the variance explained by component k. This score reflects how strongly a glycopeptide feature drives variance across conditions, independent of class labels.

The differential_analysis() function identifies significantly regulated glycopeptides under user-defined thresholds (e.g., FC > 1.5, P-value < 0.05). In addition, it supports auxiliary filtering, allowing users to combine multiple indices, such as selecting features that satisfy both FC and machine learning importance criteria, enabling downstream prioritization of glycopeptides for predictive modeling. For each threshold combination, the function also summarizes substructural regulation patterns using both absolute abundance comparison (based on the number of up- and down-regulated glycopeptides associated with each substructural feature) and proportional enrichment analysis (based on the fraction of regulated glycopeptides among all glycopeptides bearing a given feature), offering complementary perspectives on differential glycosylation.

The threshold_variation_analysis() function explores how glycan substructural regulation varies across a range of index thresholds. For FC, default thresholds span [1.2, 1.5, 2.0, 2.5, 3.0] (and reciprocals); for AUC: [0.6, 0.7, 0.8, 0.9, 1.0]; and for Random Forest, XGBoost, and PCA: the five quantiles of feature importance distributions. For statistical tests (ANOVA and chi-square), thresholds are derived from the five quantiles of F-scores and chi-squared scores, respectively. Outputs include the number and proportion of substructures showing upregulation, downregulation, and their directional ratio across thresholds.

The glycopeptide_glycosite_glycan_variation() and glycoprotein_glycosite_glycan_variation() functions quantify the number and distribution of upregulated and downregulated glycan structures associated with each glycosylation site at either the glycopeptide or glycoprotein level. This supports the dissection of site-specific glycan remodeling events within regulated proteins.

Together, this module enables robust, multi-dimensional glycopeptide differential analysis with customizable analytical granularity. It also facilitates the identification of substructures and glycosylation patterns most affected by experimental perturbations, offering a foundation for mechanistic interpretation or machine learning–based predictive modeling.

Module ‘StrucGAP_FunctionAnnotation’

The StrucGAP_FunctionAnnotation module enables multi-source functional enrichment analysis and explores the relationship between glycan substructures and enriched biological processes. It supports annotation of glycoproteins derived from StrucGAP_Preprocess, differentially expressed glycopeptides from StrucGAP_GlycoPeptideQuant, and global proteomic data from StrucGAP_GlycoNetwork. This module includes four core functions: ora(), gsea(), go_function_structure(), and kegg_function_structure().The ora() function performs over-representation analysis using the g:Profiler API⁴¹ and supports comprehensive customization, including: selection of species-specific background, customizable enrichment databases (e.g., GO:MF, GO:BP, GO:CC, KEGG, REAC, WP, TF, MIRNA, HPA, CORUM, HP), choice of significance thresholds based on upregulated or downregulated features from StrucGAP_GlycoPeptideQuant. This design enables both broad exploratory analyses and targeted hypothesis-driven investigations of glycoprotein function.

The gsea() function conducts GSEA using the GSEApy⁴², allowing users to analyze ranked gene or glycopeptide lists. Like ora(), it supports multiple enrichment libraries and species-specific settings, and is particularly useful when glycoproteomic signal changes are subtle but coordinated across functionally related genes.

The go_function_structure() function performs integrated analysis between enriched GO terms and glycan substructural features (e.g., core structures, glycan types, branching patterns, sialylation/fucosylation motifs). By calculating enrichment ratios and co-occurrence patterns, it reveals substructure features that preferentially associate with specific biological functions—such as immune regulation, cell adhesion, or signal transduction. Analogous to go_function_structure(), the kegg_function_structure() function maps glycan substructures to enriched KEGG pathways, uncovering connections between glycosylation patterns and metabolic, signaling, or disease-related pathways. Outputs include a matrix linking pathway terms to specific substructural motifs, enabling targeted interpretation of glycan functional roles.

Together, this tools enables dual-layered annotation: (1) classical functional enrichment at the protein or peptide level, and (2) glycan-aware functional interpretation, where key substructures are prioritized based on their biological relevance. This integrated approach expands the interpretability of glycoproteomic data, highlighting structure-function relationships that may reflect regulatory logic, biosynthetic constraints, or disease relevance.

Module ‘StrucGAP_GlycoNetwork’

The StrucGAP_GlycoNetwork module facilitates analysis of upstream and downstream interactions involved in the glycosylation process, integrating global proteomics data, enzyme expression profiling, and glycan-related site classification. It addresses whether observed changes in site-specific N-glycan features reflect alterations in glycosylation itself or are secondary to shifts in global protein abundance. This module includes two primary functional components.

The proteomics() function implements a comprehensive preprocessing pipeline for global proteomic datasets, natively accepts global proteome tables exported by the widely used Proteome Discoverer, enabling seamless ingestion and glycosylation network analysis whenever such data are available. Raw TMT-labeled protein-level quantification data undergo multiple quality control steps, including FDR and PSM count thresholds filtering, missing value imputation, outlier correction, and flexible sample- and feature-level normalization strategies (options include robust normalization, quantile normalization, z-score, range normalization, pareto normalization, level normalization, or no normalization). Following preprocessing, the module calculates differential expression values (FC and P- value) for each protein. These results are cross-referenced with glycopeptide-level expression data from glycoproteomic analyses to classify each IGP into one of nine glycosylation regulatory categories (e.g., downregulated glycopeptide on unchanged protein, or upregulated glycopeptide on upregulated protein). This stratification distinguishes direct glycosylation regulation from abundance-driven shifts and aids interpretation of regulatory versus structural modulation of glycan expression. In addition, this function tracks glycosylation dynamics at the site level, integrating glycosylation status across matched glycoproteins and corresponding IGPs. Furthermore, the module includes five curated functional submodules for profiling glycosylation-relevant protein families: (1) glycosyltransferases(): quantifies 161 glycosyltransferases curated from previous literature⁴³, (2) glycosidases(): profiles 63 glycosidases annotated in UniProt⁴⁴, (3) sialyltransferases(): covers 21 sialyltransferases from UniProt⁴⁴, (4) fucosyltransferases(): tracks 51 fucosyltransferases from UniProt⁴⁴, (5) glycan_binding_proteins(): annotates 276 glycan-binding proteins curated from the GlyCosmos database⁴⁵. Users may either employ the built-in reference lists or upload customized sets of interest. For all sets, quantitative expression levels are extracted from global proteomics datasets, enabling correlation analyses between enzyme or glycan-binding protein expression and observed glycosylation patterns. To support integrative annotation workflows, the utility function convert_accession_to_gene() leverages the STRING database API⁴⁶ to convert protein accession numbers into gene symbols, facilitating alignment of proteomic and transcriptomic data layers.

Together, StrucGAP_GlycoNetwork bridges site-specific glycoproteomic changes with upstream regulatory enzymes and downstream protein-level trends, providing a mechanistic framework for dissecting the molecular networks governing glycosylation.

Module ‘StrucGAP_DataVisualization’

The StrucGAP_DataVisualization module enhances the StrucGAP platform by providing advanced visualization and automated report generation, ensuring comprehensive data interpretation. It processes Dataframes from upstream modules and offers 38 visualization schemes, including functions like radar(), bar(), and boxplot(), utilizing a suite of Python libraries such as pyecharts (https://pyecharts.org/#/), matplotlib⁴⁷, PyComplexHeatmap⁴⁸, plotly⁴⁹, and networkx⁵⁰ for flexible and publication-ready rendering. Three functions, correlation_heatmap(), dimension_reduction(), and network_plot(), integrate analytical capabilities. The correlation_heatmap() function computes correlation coefficients across features, dimension_reduction() supports PCA, t-SNE, and UMAP⁴⁰ for dimensionality reduction, and network_plot() leverages the STRING API⁴⁶ for protein-protein interaction (PPI) analysis. Notably, draw_glycans() also enables batch rendering of glycan structures based on user-defined glycan coding lists, an efficiency enhancement over existing tools limited to one-by-one visualization.

A distinctive capability of this module lies in its support for automated report composition. Beyond plot rendering, StrucGAP allows users to batch-generate layout-optimized figure panels and annotated reports in PDF format. This is enabled by three key functions. The function add_figure() collects figure metadata and assigns them to designated figure groups. The function_calculate_layout() dynamically computes layout grids based on the number of figures, dividing an A4 page into five vertical regions—four for images (up to 12 total) and one for textual captions. The function compose_figures() integrates all elements into a structured PDF report. It supports custom region mapping (custom_sizes) and includes automatic figure numbering and legend rendering. This report generation workflow enables automated, customizable, and visually coherent packaging of analytical results, streamlining the preparation of figure panels for publications, presentations, or internal data reviews.

Together, this module enhances the interpretability, accessibility, and dissemination of glycoproteomics results by tightly coupling data visualization with a modular reporting infrastructure.

Module ‘StrucGAP_InsightTracker’

The StrucGAP_InsightTracker module provides systematic tracking of all analytical processes within StrucGAP, ensuring full reproducibility, transparency, and traceability of the glycoproteomics analysis workflow. It also offers automated extraction of biologically meaningful glycan substructural patterns and enables serialized output of session-wide results for rapid sharing and reuse. To ensure complete traceability across the StrucGAP platform, StrucGAP_InsightTracker records all executed modules, parameter settings, and output content using the following core functions: register_module() logs each module execution, log_params() records all input parameters, log_output() stores intermediate and final module outputs. The output_analysis_params() function generates a comprehensive tabular summary of all parameters used across modules, facilitating both auditability and method reporting.

The key_information_extraction() function mines outputs from both StrucGAP_GlycoPeptideQuant, StrucGAP_GlycoNetwork, and StrucGAP_FunctionAnnotation to identify structurally and functionally relevant glycan substructural features. From StrucGAP_GlycoPeptideQuant, the function examines results generated by threshold_variation_analysis() function and flags substructures whose regulation ratios progressively increase across thresholds (e.g., FC) variation, with a minimum occurrence threshold (default: ≥ 3) to ensure robustness. From StrucGAP_GlycoNetwork, the function extracts and organizes key information into a relationship network. This function operates by querying the UniProt API to retrieve the function description text of each protein in the global proteome dataset. Using regular expressions, it identifies glycosylation-related enzymes and glycan-binding proteins potentially associated with 10 key substructural features: LacNAc, GlcNAc, Lewis, LacdiNAc, sialylation, Neu5Ac sialylation, Neu5Gc sialylation, fucosylation, Core fucosylation, and Antenna fucosylation. For each feature, regular expressions are created to match enzymes and glycan-binding proteins, and semantic analysis of the function description is employed to assess the relevance of these proteins to the specific glycan substructure (due to the inherent limitations of regular expression matching and semantic analysis, this process may result in occasional false-positive associations, and users are therefore advised to further inspect or validate the matched results in practical applications). Inclusion and exclusion rules are also defined to refine these associations. Once these associations are identified, networkx⁵⁰ is used to build a relationship network. In this network, the upstream nodes represent enzymes or glycan-binding proteins, with node size proportional to the FC of the respective enzyme or protein. The connections between upstream nodes and central nodes (glycan substructural features) are represented by lines weighted with the Spearman correlation coefficient, indicating the strength of association between the enzyme/protein and glycan feature. Downstream nodes represent glycoprotein-associated signaling pathways, with node size proportional to the number of glycoproteins associated with that pathway. From StrucGAP_FunctionAnnotation, the function inspects enrichment results from GO and KEGG databases and computes the preference of each glycan substructure (e.g., core structure, glycan type, sialylation/fucosylation patterns) for specific functional terms. This is quantified as the proportion of enriched terms in which a substructure is dominant (e.g., ratio ≥ 0.6). When the number of dominant terms exceeds 5% of the total number of terms for a given database-substructure pair, a structure-function preference is inferred. For example, if 6 of 133 GO:BP terms show > 60% dominance of a particular core structure, this about the 5% threshold, indicating potential biological specificity.

To facilitate reproducible workflows and collaborative analysis, StrucGAP_InsightTracker includes two utility functions for session-wide data storage. The output_pickle() function serializes all non-Flask global variables into a.pkl file, excluding modules or proxy objects that cannot be pickled, and the read_pickle() function restores these variables into the current namespace, ensuring fast reinstatement of an entire analysis session.

Together, StrucGAP_InsightTracker ensures that the entire glycoproteomics pipeline remains fully auditable and reproducible, while its automated key insight extraction and serialization features make it a valuable tool for downstream interpretation, figure preparation, and collaborative research.

Ethical statement

All procedures related to mouse studies complied with ethical regulations and were approved by the Ethics Committee of Northwest University, China (Approval No. NWU-AWC-20240901M).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

M.Y., Y.W., Z.Z., and S.S. designed the study; Y.W. performed sample preparation and MS analysis with help from Z.Z. and Y.C.; Y.X., X.W., and K.H. completed a manual verification of analysis results. M.Y. developed the methods, packages, and prepared the figures with help from Z.J. and T.L.; M.Y. and S.S. wrote and edited the manuscript with help from all the authors.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

Unless otherwise stated, all data supporting the results of this study can be found in the article, supplementary, and source data files. The mass spectrometry data have been deposited to the ProteomeXchange Consortium⁵¹ [http://proteomecentral.proteomexchange.org] via the iProX partner repository⁵² [https://www.iprox.cn] under the dataset identifier PXD063260. Source data are provided in this paper.

Code availability

StrucGAP is a pip-installable Python package. The source code is available at GitHub [https://github.com/Sun-GlycoLab/StrucGAP], with documentation at [https://strucgap.readthedocs.io/en/latest/index.html]. A citable version of the code used in this study has been archived on Zenodo [10.5281/zenodo.18523067].

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-70560-7.