Machine Learning-Based Glycolipid Metabolism Gene Signature Predicts Prognosis and Immune Landscape in Oesophageal Squamous Cell Carcinoma

Abstract

Using machine learning approaches, we developed and validated a novel prognostic model for oesophageal squamous cell carcinoma (ESCC) based on glycolipid metabolism-related genes. Through integrated analysis of TCGA and GEO datasets, we established a robust 15-gene signature that effectively stratified patients into distinct risk groups. This signature demonstrated superior prognostic value and revealed significant associations with immune infiltration patterns. High-risk patients exhibited reduced immune cell infiltration, particularly in B cells and NK cells, alongside increased tumour purity. Single-cell RNA sequencing analysis uncovered unique cellular composition patterns and enhanced interaction intensities in the high-risk group, especially within epithelial and smooth muscle cells. Functional validation confirmed MECP2 as a promising therapeutic target, with its knockdown significantly inhibiting tumour progression both in vitro and in vivo. Drug sensitivity analysis identified specific therapeutic agents showing potential efficacy for high-risk patients. Our study provides both a practical prognostic tool and novel insights into the relationship between glycolipid metabolism and tumour immunity in ESCC, offering potential strategies for personalised treatment.