Related Experiment Video
Updated: Jun 9, 2025
Author Spotlight: Advancing Early Detection and Treatment of Gastrointestinal Tumors
Published on: February 16, 2024
Identification of Fatty Acid Metabolism-Related Subtypes in Gastric Cancer Aided by Machine Learning
Maolin Hou1, Jinghua Chen2, Le Yang3
1Department of Internal Medicine, Siziwangqi People's Hospital, Wulancabu, 011800, People's Republic of China.
Introduction:
Gastric cancer, the fifth most common malignant tumor in the world, poses a serious threat to human health. However, the role of fatty acid metabolism (FAM) in gastric cancer remains incompletely understood. We aim to provide guidance for clinical decisions by utilizing public database of gastric adenocarcinoma to establish an FAM-related gene subtypes via machine learning algorithm.
Methods:
The intersection of FMGs from KEGG, Hallmark, and Reactome bioinformatics databases and the DEGs of the TCGA-STAD cohort was used to decompose the gene matrix related to establish FAM-related gene subtypes by NMF. Comparison of immune infiltrating differences between subtypes using ESTIMATE and Cibersort algorithms. The multifactor Cox regression to identify independent risk genes for patient prognosis based on the subtypes. A prognostic model including independent risk genes was built using random survival forest and Cox regression. IHC validation in gastric cancer and adjacent tissues confirmed the above gene expression level.
Results:
71 DEGs related to FMGs of STAD were identified, which was used to established the FAM-related gene subtypes, C1 and C2. The immune infiltrating analysis showed that most immune features of C2 were significantly upregulated compared to C1. The independent risk genes were CGβ8, UPK1B, and OR51G based on the subtypes. A gastric cancer prognostic model consisting of independent risk genes was constructed and patients were classified into high-risk and low-risk groups with survival differential analysis. Finally, IHC showed that CGβ8 and UPK1B expression were upregulated in gastric cancer, while OR51G2 did not detect differences in expression.
Conclusion:
The study developed a machine learning-based gastric cancer prognosis risk model using FMGs. This model effectively stratifies patients according to their risk levels and provides valuable insights for clinical decision-making, enabling accurate evaluation of patient prognosis.
