A novel gene signature for predicting outcome in colorectal cancer patients based on tumor cell-endothelial cell interaction via single-cell sequencing and machine learning.

Background: The intricate interactions between malignant cells and endothelial cells (ECs) are crucial in the progress of colorectal cancer (CRC). Identifying molecular signatures associated with this interaction could yield critical prognostic insights and inform personalized therapeutic approaches. Methods: We conducted an in silico study integrating single-cell RNA sequencing and bulk transcriptome data to characterize the cellular heterogeneity of CRC. Through computational cell interaction analysis facilitated the elucidation of signaling dynamics among cell subpopulations linked to CRC prognosis. Prognostic signatures were developed using various machine learning algorithms based on marker genes linked to the identified cell subpopulations. Immune cell infiltration assessment and gene enrichment analysis were performed to characterize CRC patients stratified by the signature. Results: Our analysis revealed two distinct cell subgroups, Malignant Cluster01 tumor cells, and Tip-like endothelial cells, showing significant interaction and closely associated with colorectal cancer prognosis. Specifically, Malignant Cluster01 subpopulations primarily served as signal senders, while Tip-like endothelial cells acted as receivers in PARs signaling. The Malignant Cluster01 and Tip-like endothelial cells related machine learning-derived prognostic signature (MTMLDPS), demonstrated potent prognostic capability, effectively predicting colorectal cancer patient outcomes across diverse databases. The colorectal cancer group with a high Malignant Cluster01 and Tip-like endothelial cells related machine learning-derived prognostic signature score exhibited significant associations with invasion, epithelial-mesenchymal transition, and angiogenesis pathways, along with immune cell infiltration. Conclusion: The Malignant Cluster01 and Tip-like endothelial cells related machine learning-derived prognostic signature holds promise for improving prognostic precision and guiding individual therapeutic strategies in colorectal cancer patients. Moreover, our findings emphasize the importance of considering tumor-endothelial cell interactions in cancer prognosis, providing insights for future therapeutic interventions targeting these interactions.