Scholarlink

Background <p>Oral squamous cell carcinoma (OSCC) carries a persistently poor prognosis, with five-year survival rates of 50–60%, driven in part by its marked biological heterogeneity. Here, we integrated multi-omics data and machine learning to develop and characterise a candidate prognostic signature and to explore its potential correspondence with routine clinical pathological markers.</p> Methods <p>Candidate prognostic genes were identified by integrating bulk transcriptomic data (TCGA and GEO cohorts) through differential expression analysis, Mfuzz temporal clustering, and weighted gene co-expression network analysis (WGCNA). An optimal prognostic model was selected by benchmarking 101 combinations of machine-learning algorithms. Single-cell RNA sequencing (scRNA-seq) was used to characterise the cellular origins and signalling interactions of the signature. Clinical relevance was explored in an independent cohort of 37 OSCC patients.</p> Results <p>This approach identified a three-gene candidate prognostic signature comprising PTGFRN, MICB, and IFI16. The optimal model—StepCox[forward] combined with Random Survival Forest—demonstrated consistent discriminative performance (1-year AUC: 0.871 in training, 0.795 [95% CI 0.627–0.928] in internal validation, and 0.683 in cross-dataset validation). SHAP analysis identified PTGFRN as the most influential contributor to individual risk predictions. At the single-cell level, a proliferating cell subpopulation co-expressing these three genes served as a signalling hub, preferentially communicating with malignant epithelial cells via the cyclophilin A (CypA) pathway. Qualitative multiplex immunofluorescence (mIF) staining of paired tumour and adjacent normal tissues indicated apparently elevated protein-level expression of all three signature genes in OSCC; this observation is descriptive only, as formal image-based quantification was not performed and remains to be performed. In an independent retrospective clinical cohort (n = 37), the aggressive phenotype “P16 − /p53mut/High Ki-67” was identified in 4 of 37 patients (10.8%) and showed an exploratory, fragile association with advanced T stage (4/4 [100.0%] vs. 11/33 [33.3%]; Fisher’s exact <i>p</i> = 0.010); a single-patient reclassification sensitivity analysis abolished this signal (<i>p</i> = 0.142), and the very small subgroup size (n = 4) precludes any confirmatory interpretation. Trends toward higher lymph node metastasis and advanced clinical stage did not reach statistical significance. All clinical-cohort findings are therefore hypothesis-generating and not a validation of the signature. Exploratory molecular docking suggested potential binding of PTGFRN–dasatinib (− 7.3&#xa0;kcal/mol) and IFI16–AZD7762 (− 8.8&#xa0;kcal/mol), warranting experimental validation.</p> Conclusion <p>The PTGFRN/MICB/IFI16 signature represents a candidate prognostic biomarker for OSCC, whose preliminary molecular correspondence with a clinically identifiable pathological phenotype (P16-/p53mut/Ki-67high) and hypothesis-generating drug-sensitivity predictions provide a basis for prospective validation, which is required before any clinical application can be considered.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-omics and machine-learning framework identifies a candidate three-gene prognostic signature (PTGFRN, MICB, and IFI16) for oral squamous cell carcinoma with exploratory associations with a P16-negative, p53-mutant, and high Ki-67 pathological phenotype

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月30日

查看全文
  • Tian-Feng Xu,
  • Xian Wang,
  • Hua Li,
  • Shi-Xi He,
  • Yu-Liang He,
  • Xuan-Ping Huang

摘要

Background

Oral squamous cell carcinoma (OSCC) carries a persistently poor prognosis, with five-year survival rates of 50–60%, driven in part by its marked biological heterogeneity. Here, we integrated multi-omics data and machine learning to develop and characterise a candidate prognostic signature and to explore its potential correspondence with routine clinical pathological markers.

Methods

Candidate prognostic genes were identified by integrating bulk transcriptomic data (TCGA and GEO cohorts) through differential expression analysis, Mfuzz temporal clustering, and weighted gene co-expression network analysis (WGCNA). An optimal prognostic model was selected by benchmarking 101 combinations of machine-learning algorithms. Single-cell RNA sequencing (scRNA-seq) was used to characterise the cellular origins and signalling interactions of the signature. Clinical relevance was explored in an independent cohort of 37 OSCC patients.

Results

This approach identified a three-gene candidate prognostic signature comprising PTGFRN, MICB, and IFI16. The optimal model—StepCox[forward] combined with Random Survival Forest—demonstrated consistent discriminative performance (1-year AUC: 0.871 in training, 0.795 [95% CI 0.627–0.928] in internal validation, and 0.683 in cross-dataset validation). SHAP analysis identified PTGFRN as the most influential contributor to individual risk predictions. At the single-cell level, a proliferating cell subpopulation co-expressing these three genes served as a signalling hub, preferentially communicating with malignant epithelial cells via the cyclophilin A (CypA) pathway. Qualitative multiplex immunofluorescence (mIF) staining of paired tumour and adjacent normal tissues indicated apparently elevated protein-level expression of all three signature genes in OSCC; this observation is descriptive only, as formal image-based quantification was not performed and remains to be performed. In an independent retrospective clinical cohort (n = 37), the aggressive phenotype “P16 − /p53mut/High Ki-67” was identified in 4 of 37 patients (10.8%) and showed an exploratory, fragile association with advanced T stage (4/4 [100.0%] vs. 11/33 [33.3%]; Fisher’s exact p = 0.010); a single-patient reclassification sensitivity analysis abolished this signal (p = 0.142), and the very small subgroup size (n = 4) precludes any confirmatory interpretation. Trends toward higher lymph node metastasis and advanced clinical stage did not reach statistical significance. All clinical-cohort findings are therefore hypothesis-generating and not a validation of the signature. Exploratory molecular docking suggested potential binding of PTGFRN–dasatinib (− 7.3 kcal/mol) and IFI16–AZD7762 (− 8.8 kcal/mol), warranting experimental validation.

Conclusion

The PTGFRN/MICB/IFI16 signature represents a candidate prognostic biomarker for OSCC, whose preliminary molecular correspondence with a clinically identifiable pathological phenotype (P16-/p53mut/Ki-67high) and hypothesis-generating drug-sensitivity predictions provide a basis for prospective validation, which is required before any clinical application can be considered.