Usage Information
Abstract
BACKGROUND. Hepatocellular carcinoma (HCC) exhibits molecular heterogeneity that challenges histopathologic classification and biomarker discovery. We assessed whether spatially resolved N-glycan imaging with machine learning could classify tumor regions and infer glutamine synthetase (GS) status. METHODS. In this retrospective study, MALDI mass spectrometry imaging of N-glycans was performed on formalin-fixed, paraffin-embedded sections from two independent cohorts (discovery, n = 88; validation, n = 60) with pathologist annotation. An XGBoost classifier was trained on 90 discriminative N-glycan features using patient-grouped cross-validation. Performance was assessed by AUC for pixel- and biopsy-level discrimination of tumor from adjacent non-tumor tissue, and for GS status classification. RESULTS. Pixel-level AUCs were 0.95 (cross-validation) and 0.89 (external validation); biopsy-level AUCs were 1.0 and 0.97, correctly identifying 97% of tumor-containing biopsies. Probability maps recapitulated pathologist-defined boundaries; UMAP embeddings captured inter- and intratumoral heterogeneity. Discriminative species (m/z 2393.846, 1905.634, 1743.579, 1809.639) reflected complex, fucosylated, branched remodeling. N-glycans bearing six GlcNAc residues were enriched in GS+ (n = 45) versus GS− (n = 17) tumors (P = 0.001) and discriminated GS status (AUC = 0.75), consistent with GLUL and MGAT5 upregulation in TCGA-LIHC. CONCLUSION. MALDI N-glycan imaging with machine learning enables spatially resolved, objective classification of HCC and links glycan phenotypes to tumor-associated metabolic programs. TRIAL REGISTRATION. Not applicable; retrospective analysis of archival, de-identified tissue. FUNDING. NIH/NCI R01CA285370, 1R01CA289381, R33CA267226, R01CA282022, R21CA263464, R21CA286287, R01CA253460, S10OD030212, R01CA251155, R01CA250227, U01CA271887, P50CA295495, P30CA138313, P20GM130457, P30DK123704, P30DK120531,R24DK139775; NIH/NIA R01AG078702; Smart State Endowment, State of South Carolina; LeDucq Foundation.
Authors
Muhammed F. Bayram, Jade K. Macdonald, Andrew DelaCourt, Peggi M. Angel, Richard R. Drake, Aatur Singhi, David Geller, Satdarshan P. Monga, Amit Singal, Anand Mehta
Usage data is cumulative from September 2026 through October 2026.
| Usage | JCI | PMC |
|---|---|---|
| Text version | 133 | 0 |
| 67 | 0 | |
| Supplemental data | 20 | 0 |
| Citation downloads | 50 | 0 |
| Totals | 270 | 0 |
| Total Views | 270 | |
Usage information is collected from two different sources: this site (JCI) and Pubmed Central (PMC). JCI information (compiled daily) shows human readership based on methods we employ to screen out robotic usage. PMC information (aggregated monthly) is also similarly screened of robotic usage.
Various methods are used to distinguish robotic usage. For example, Google automatically scans articles to add to its search index and identifies itself as robotic; other services might not clearly identify themselves as robotic, or they are new or unknown as robotic. Because this activity can be misinterpreted as human readership, data may be re-processed periodically to reflect an improved understanding of robotic activity. Because of these factors, readers should consider usage information illustrative but subject to change.
