A machine learning-assisted hypoxia-associated prognostic model reveals PFKP-mediated glycolytic regulation in uveal melanoma
Abstract
PurposeUveal melanoma (UM) is a highly aggressive intraocular malignancy with strong metastatic propensity and poor response to immunotherapy. Hypoxia profoundly reshapes the tumor microenvironment and drives immune evasion. This study aimed to develop a machine learning-assisted hypoxia-related prognostic model and identify relevant metabolic candidates associated with UM progression.MethodsRNA-sequencing and clinical data from 80 UM patients were obtained from TCGA. Thirty highly connected hypoxia-related genes were prioritized through protein-protein interaction analysis and evaluated using five complementary survival-based machine learning approaches: LASSO-Cox, CoxBoost, random survival forest, XGBoost-Cox, and Survival-SVM. Genes supported by at least three of the five methods were defined as core candidates and further evaluated by multivariable Cox regression to construct the prognostic model. The prognostic relevance of the model was evaluated in the dependent GSE22138 cohort using metastasis-free survival (MFS) as the clinical endpoint. Survival analysis, ROC, PCA, GSEA, and immune microenvironment analyses were performed. PFKP was further examined for its associations with immune features and experimentally validated through in vitro and in vivo assays.ResultsMulti-algorithm consensus identified PFKP, ENO3, and FOXO3 as core candidate genes. PFKP showed the strongest cross-model consistency and was supported by all five methods, whereas ENO3 and FOXO3 were supported by four methods. Multivariable Cox regression retained ENO3 and PFKP as the final prognostic genes, yielding a parsimonious two-gene risk model that was further evaluated in the independent GSE22138 cohort. Higher risk scores were associated with worse overall survival in TCGA-UM, shorter MFS in GSE22138, and more advanced disease. Hypoxia- and glycolysis-related pathways were enriched in the high-risk group, which also exhibited higher immune and stromal scores, lower tumor purity, and increased expression of immune checkpoint genes. PFKP expression was positively associated with CD8+ T-cell infiltration and immune checkpoint-related gene expression. Functionally, PFKP knockdown reduced UM cell proliferation, migration, and glycolytic activity in vitro and suppressed xenograft tumor growth in vivo.ConclusionWe established a cross-algorithm hypoxia-related prognostic framework for UM. ENO3 and PFKP constituted the final prognostic signature, while PFKP, supported by all five algorithms and experimental validation, emerged as a functionally relevant glycolysis-related candidate associated with UM progression.
Related
- Conserved Brain-Associated CD8+ T Cell Programs and Their Regulation by the Brain Microenvironment
- Microbiome-Associated Regulation of Isoquinoline Alkaloids in Sinomenium acutum Reveals Functional Endophytes for Sinomenine Accumulation
- Oncogenic CLDN1 promotes thyroid cancer progression through TAp63-mediated glycolytic switching
- Machine learning-assisted Repli-Histo labeling reveals distinct transcription-dependent constraints on chromatin motion in living cells
- Machine learning-assisted genomic characterization of population differentiation between Chinese Kirghiz and Kyrgyzstan local sheep
BibTeX
@article{lin2026machine,
title = {{A machine learning-assisted hypoxia-associated prognostic model reveals PFKP-mediated glycolytic regulation in uveal melanoma}},
author = {Kunxia Lin and Yi Cui and Yuzhe Jiang and Mi Zheng and Yuyang Yang and Jing Chen and Wenjie Wu and Nuo Xu},
journal = {Frontiers in Immunology},
year = {2026},
month = nov,
doi = {10.3389/fimmu.2026.1993731},
url = {https://www.frontiersin.org/articles/10.3389/fimmu.2026.1993731},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - A machine learning-assisted hypoxia-associated prognostic model reveals PFKP-mediated glycolytic regulation in uveal melanoma
AU - Kunxia Lin
AU - Yi Cui
AU - Yuzhe Jiang
AU - Mi Zheng
AU - Yuyang Yang
AU - Jing Chen
AU - Wenjie Wu
AU - Nuo Xu
PY - 2026
DA - 2026/11/05
JO - Frontiers in Immunology
DO - 10.3389/fimmu.2026.1993731
UR - https://www.frontiersin.org/articles/10.3389/fimmu.2026.1993731
AB - PurposeUveal melanoma (UM) is a highly aggressive intraocular malignancy with strong metastatic propensity and poor response to immunotherapy. Hypoxia profoundly reshapes the tumor microenvironment and drives immune evasion. This study aimed to develop a machine learning-assisted hypoxia-related prognostic model and identify relevant metabolic candidates associated with UM progression.MethodsRNA-sequencing and clinical data from 80 UM patients were obtained from TCGA. Thirty highly connected hypoxia-related genes were prioritized through protein-protein interaction analysis and evaluated using five complementary survival-based machine learning approaches: LASSO-Cox, CoxBoost, random survival forest, XGBoost-Cox, and Survival-SVM. Genes supported by at least three of the five methods were defined as core candidates and further evaluated by multivariable Cox regression to construct the prognostic model. The prognostic relevance of the model was evaluated in the dependent GSE22138 cohort using metastasis-free survival (MFS) as the clinical endpoint. Survival analysis, ROC, PCA, GSEA, and immune microenvironment analyses were performed. PFKP was further examined for its associations with immune features and experimentally validated through in vitro and in vivo assays.ResultsMulti-algorithm consensus identified PFKP, ENO3, and FOXO3 as core candidate genes. PFKP showed the strongest cross-model consistency and was supported by all five methods, whereas ENO3 and FOXO3 were supported by four methods. Multivariable Cox regression retained ENO3 and PFKP as the final prognostic genes, yielding a parsimonious two-gene risk model that was further evaluated in the independent GSE22138 cohort. Higher risk scores were associated with worse overall survival in TCGA-UM, shorter MFS in GSE22138, and more advanced disease. Hypoxia- and glycolysis-related pathways were enriched in the high-risk group, which also exhibited higher immune and stromal scores, lower tumor purity, and increased expression of immune checkpoint genes. PFKP expression was positively associated with CD8+ T-cell infiltration and immune checkpoint-related gene expression. Functionally, PFKP knockdown reduced UM cell proliferation, migration, and glycolytic activity in vitro and suppressed xenograft tumor growth in vivo.ConclusionWe established a cross-algorithm hypoxia-related prognostic framework for UM. ENO3 and PFKP constituted the final prognostic signature, while PFKP, supported by all five algorithms and experimental validation, emerged as a functionally relevant glycolysis-related candidate associated with UM progression.
SN - 1664-3224
ER -