Identifying a prognostic signature for clear cell renal cell carcinoma: the convergence of single-cell and bulk sequencing with machine learning.

Background: Clear cell renal cell carcinoma (ccRCC) is a highly aggressive renal cancer subtype and lacks highly precise individualized treatment options. Thus, we used a novel computational framework to construct a consensus machine learning-related signature (MLRS) to predict prognosis and screen patients effectively for immunotherapy. Methods: An integrative machine learning procedure involving 10 methods was used to contract MLRS. Various methods were used to evaluate immune cell infiltration and biological characteristics. Moreover, we explored the response to immunotherapy and drug sensitivity. Single-cell RNA sequencing analysis, qRT-PCR, and a CCK-8 assay were used to clarify the biological functions of the hub gene. Results: MLRS demonstrated outstanding performance in predicting prognosis compared with the other published signatures, and the high-MLRS group had a favorable outcome in four independent datasets. Furthermore, the low-MLRS group displayed a greater possibility of responding to immunotherapy and had a "hot" tumor immunophenotype. The high-MLRS group was characterized by a phenotype of immune suppression and was less likely to benefit from immunotherapy, while some small molecule inhibitors might serve as promising treatment options. Single-cell analysis revealed that MLRS was highly enriched in endothelial cells. We also identified DLL4/Notch and JAG/Notch signaling as the key ligand-receptor pairs in ccRCC. EMCN was downregulated in ccRCC, and further functional experiments demonstrated that EMCN knockdown inhibited cell proliferation. Conclusion: The MLRS can predict patient prognosis, may be utilized to screen potential populations that may benefit from immunotherapy, and predict potential drug targets, with broad significance for the clinical treatment of ccRCC.