Machine learning-derived cellular senescence index for predicting prognosis and drug sensitivity in patients with...

Cellular senescence, an inevitable phase in the cellular lifecycle, is increasingly implicated in cancer development. Clear cell renal cell carcinoma (ccRCC), a lethal malignancy of the urinary system, underscores the need for senescence-based risk models. Through single-cell analysis, we identified senescent cells within ccRCC tumors and delineated their distinct biological features. We then integrated ten machine learning algorithms-plsRcox, Ridge, Enet, CoxBoost, Lasso, StepCox, RSF, SuperPC, GBM, and survivalSVM-generating 101 combinatorial models via pairwise integration. The optimal Lasso-StepCox model was selected based on the highest mean concordance index (C-index), yielding a minimized senescence-related gene signature of only 9 genes (significantly below the typical 15-30-gene range). This signature formed the basis of a senescence-related scoring model (SRSM) for ccRCC patient survival assessment. Patients with high SRSM exhibited significantly poorer survival (P < 0.001), enhanced oxidative phosphorylation, and an immunosuppressive tumor microenvironment (TME) characterized by elevated regulatory T cell (Treg) infiltration. In vitro validation confirmed that NME2 knockdown suppressed ccRCC proliferation and invasion. Collectively, the SRSM framework provides a precise tool for prognostic stratification and therapeutic targeting in ccRCC. Therefore, to bridge the gap between the intricate landscape of cellular senescence revealed by scRNA-seq and the pressing clinical need for robust prognostic tools, we aimed to develop a quantifiable senescence-related scoring model (SRSM). By integrating multiple machine learning algorithms, we sought not only to achieve superior predictive accuracy but also to derive a minimized gene signature for enhanced clinical applicability, ultimately translating the biological insights of senescence into a precise prognostic framework for ccRCC.