Background: Colorectal cancers (CRCs) continue to be the leading cause of cancer-related deaths worldwide. The exact landscape of the molecular features of TGF-β pathway-inducing CRCs remains uncharacterized.
Methods: Unsupervised hierarchical clustering was performed to stratify samples into two clusters based on the differences in TGF-β pathways. Weighted gene co-expression network analysis was applied to identify the key gene modules mediating the different characteristics between two subtypes. An algorithm integrating the least absolute shrinkage and selection operator (LASSO), XGBoost, and random forest regression was performed to narrow down the candidate genes. Further bioinformatic analyses were performed focusing on COMP-related immune infiltration and functions.
Results: The integrated machine learning algorithm identified COMP as the hub gene, which exhibited a significant predictive value for two subtypes with an area under the curve (AUC) value equaling 0.91. Further bioinformatic analysis revealed that COMP was significantly upregulated in various cancers, especially in advanced CRCs, and regulated the immune infiltration, especially M2 macrophages and cancer-associated fibroblasts in CRCs.
Conclusions: Comprehensive immune analysis and experimental validation demonstrate that COMP is a reliable signature for subtype prediction. Our results could provide a new point for TGFβ-targeted anticancer drugs and contribute to guiding clinical decision making for CRC patients.
Keywords: COMP; TGF-β; colorectal cancer.