Machine learning-based definition of cellular senescence reveals pro-senescence potential implications in lung adenocarcinoma

Despite growing evidence implicating cellular senescence in tumor progression, methodological challenges in objectively quantifying senescent cell burden across cancer types continue to limit mechanistic studies and clinical translation. To establish a robust quantification framework, we curated a comprehensive compendium of 888 transcriptomic profiles spanning diverse cell types and senescence-inducing conditions. Following rigorous normalization and batch-effect correction, the Boruta algorithm was employed to derive a stable consensus cellular senescence-related gene signature (CSGS). By benchmarking ten machine learning algorithms within a rigorous cross-validation framework, we developed the Predictive Cellular Senescence Model (PreCSenM), which integrates the CSGS to compute a continuous cellular senescence (CS) score as a standardized metric for senescence level. Benchmarking analyses demonstrated that PreCSenM achieves robust and consistent performance, providing more reliable prediction of senescent states than existing methods across both normal and cancer transcriptomic datasets. In lung adenocarcinoma (LUAD), PreCSenM identified the CS score as a robust predictor of clinical outcomes. Multi-omics analyses further revealed that higher CS levels correlate with greater genomic stability and enhanced immune-related features, aligning with clinical observations. Notably, drug discovery using PreCSenM identified histone deacetylase inhibitors (HDACis) as potent inducers of CS in LUAD. Transcriptional and epigenetic profiling of HDACi-treated cells pinpointed FOSB as a core transcription factor driving CS, and FOSB knockdown reduced HDACi-induced senescence. Our study establishes PreCSenM as a multidimensional senescence quantification tool, bridging computational prediction with clinical relevance and mechanistic validation to enable senescence-directed therapies in precision oncology. A user-accessible web portal for PreCSenM is available at http://precsenm.bmicc.org/ .