Bulk and single-cell RNA sequencing analyses coupled with multiple machine learning to develop a glycosyltransferase associated signature in colorectal cancer.

Background: This study aims to identify key glycosyltransferases (GTs) in colorectal cancer (CRC) and establish a robust prognostic signature derived from GTs.

Methods: Utilizing the AUCell, UCell, singscore, ssgsea, and AddModuleScore algorithms, along with correlation analysis, we redefined genes related to GTs in CRC at the single-cell RNA level. To improve risk model accuracy, univariate Cox and lasso regression were employed to discover a more clinically subset of GTs in CRC. Subsequently, the efficacy of seven machine learning algorithms for CRC prognosis was assessed, focusing on survival outcomes through nested cross-validation. The model was then validated across four independent external cohorts, exploring variations in the tumor microenvironment (TME), response to immunotherapy, mutational profiles, and pathways of each risk group. Importantly, we identified potential therapeutic agents targeting patients categorized into the high-GARS group.

Results: In our research, we classified CRC patients into distinct subgroups, each exhibiting variations in prognosis, clinical characteristics, pathway enrichments, immune infiltration, and immune checkpoint genes expression. Additionally, we established a Glycosyltransferase-Associated Risk Signature (GARS) based on machine learning. GARS surpasses traditional clinicopathological features in both prognostic power and survival prediction accuracy, and it correlates with higher malignancy levels, providing valuable insights into CRC patients. Furthermore, we explored the association between the risk score and the efficacy of immunotherapy.

Conclusion: A prognostic model based on GTs was developed to forecast the response to immunotherapy, offering a novel approach to CRC management.