Machine learning integration of bulk and single-cell RNA-seq data reveals glycolytic heterogeneity in colorectal cancer.

Abstract

As one of the most prevalent malignancies worldwide, colorectal cancer (CRC) exhibits a strong metabolic dependency on glycolysis, which fuels tumor expansion and shapes an immunosuppressive microenvironment. Despite its clinical significance, the regulatory landscape and cellular diversity of glycolytic metabolism in CRC require systematic exploration. Multi-omics datasets (bulk/scRNA-seq and spatial transcriptomics) were analyzed to quantify glycolytic signatures. Core regulatory genes were selected via integrated pathway mapping and a machine learning framework incorporating five-feature selection algorithms. Cellular subpopulations were delineated by metabolic profiles, with niche interactions modeled through ligand-receptor network analysis. Findings were validated across multicenter cohorts. Our analyses identified a tumor subpopulation characterized by a High Glycolytic State (HGS), displaying elevated glycolytic signature alongside stem-like properties. Spatial profiling demonstrated relative enrichment of HGS cells in central tumor regions, potentially reflecting adaptation to nutrient-limited conditions. Among the molecular features associated with HGS maintenance, five candidate regulators (PFKP, ERO1A, FKBP4, HDLBP, HSPA5) showed correlation with unfavorable clinical outcomes. Our study characterizes the metabolic heterogeneity of CRC and suggests a potential role for HGS cells in shaping the tumor microenvironment. The molecular features identified here may offer insights into metabolic dependencies that could be explored for future therapeutic targeting.