Background: Head and neck squamous cell carcinoma (HNSCC) responds poorly (<20%) to immune checkpoint blockade. Since sialylation suppresses anti-tumour immunity through Siglec signalling independently of the PD-L1/PD-1 axis, we mapped its prognostic landscape in HNSCC by integrating prognostic modelling, molecular subtyping, and AI-guided drug discovery. Methods: Transcriptomic data from seven GEO cohorts and TCGA-HNSC (n = 501) were analysed with a 1204-gene sialylation compendium. The Sialylation Niche Index (SNI) was built in TCGA-HNSC from 117 machine-learning algorithm combinations (nominally 101), with survival-based feature selection and model fitting confined to the training set; model selection used TCGA-HNSC and the external model-selection cohort GSE42743 (n = 74 with overall survival), and the three independent test cohorts were scored without re-fitting (E-MTAB-8588, n = 83; GSE65858, n = 270; and GSE41613, n = 97). Key model genes were characterised by single-cell RNA sequencing (168,742 cells), spatial transcriptomics, and AI-guided screening of 249,455 compounds. Results: Thirty-one prognostic sialylation-associated DEGs defined an immunosuppressive subtype (C1) and an immune-active subtype (C2) with divergent survival. The SNI achieved a C-index of 0.88 in TCGA-HNSC, 0.68 in GSE42743, and 0.58–0.62 in the three independent test cohorts and remained independently prognostic after adjustment for age and other clinicopathological variables in multivariable analysis. SHAP analysis identified HSPH1 (risk-associated) and ST6GALNAC1 (protective) as principal contributors with opposing microenvironmental associations. Conclusions: Sialylation constitutes a distinct immunosuppressive axis in HNSCC, complementary to PD-1 blockade. The SNI provides a biologically anchored prognostic framework whose cross-platform transfer requires recalibration; putative candidate compounds require experimental validation.
Background: Allergic rhinitis is common and clinically heterogeneous, but the genes whose genetically regulated immune-state expression influences susceptibility remain incompletely resolved. Methods: We integrated dynamic CD4+ T cell eQTL data (46 activation-state profiles) with allergic rhinitis GWAS summary statistics, applying transcriptome-wide Mendelian randomization (MR), colocalization, SMR/HEIDI, metabolite screening, paired single-cell RNA-seq and TCR analysis, and computational follow-up. Colocalization support was prespecified as conditional ratio PP.H4/(PP.H3 + PP.H4) > 0.7; HEIDI at p > 0.05. Results: After MHC-region exclusion, 50,654 gene-profile pairs were tested and 129 MR-significant genes were identified. Colocalization supported 418 pairs (111 genes); 70 genes passed both SMR and HEIDI; and cross-database comparison yielded a nine-gene overlap set. In GSE200107, baseline good-versus-poor responder differences were larger than treatment-induced transcriptional change; patient-level reanalysis found no pair surviving multiple-testing correction (n = 4 vs. 3); these findings are hypothesis-generating. Paired TCR analysis suggested greater clonotype renewal in good responders (descriptive). Structure-guided screening prioritized unvalidated binder candidates for SLC25A46 and TLR1 (TLR1 TIR docking −5.86 kcal/mol; MD RMSD 0.20 nm). Conclusions: An exploratory metabolite screen yielded nominally significant indirect-effect signals, but the candidate metabolite did not colocalize with the disease locus. This framework prioritizes candidate therapeutic targets for allergic rhinitis.