Background: Colon adenocarcinoma (COAD) is a heterogeneous tumor and senescence is crucial in the occurrence of cancer. This study aimed to identify senescence-based subtypes and construct a prognostic signature to predict the prognosis and guide immunotherapy or chemotherapy decisions for COAD patients.Methods: Based on the single-cell RNA sequencing (scRNA-seq) data of 13 samples from the Gene Expression Omnibus (GEO) database, we assessed cellular senescence characteristics. Transcriptome data, copy number variations (CNVs) and single nucleotide variations (SNVs) data were obtained from The Cancer Genome Atlas (TCGA) database. GSE39582 and GSE17537 were used for validation. Senescence subtypes were identified using unsupervised consensus clustering analysis, and a prognostic signature was developed using univariate Cox analysis and least absolute shrinkage and selection operator (LASSO). Response of risk groups to chemotherapy was predicted using the half-maximal inhibitory concentration (IC50) values. We further analyzed the relationship between risk gene expression and methylation level. The prediction performance was assessed by nomogram.Results: Senescence-related pathways were highly enriched in malignant cells and bulk RNA-seq verified cellular senescence. Three senescence subtypes were identified, in which patients in clust3 had poorest prognosis and higher T stage, accompanied with higher tumor mutation burden (TMB) and mutations, activated inflammatory response, more immune cell infiltration, and higher immune escape tendency. A senescence-based signature using 11 genes (MFNG, GPRC5B, TNNT1, CCL22, NOXA1, PABPC1L, PCOLCE2, MID2, CPA3, HSPA1A, and CALB1) was established, and accurately predicted a lower prognosis in high risk patients. Its robustness was validated by external cohort. Low risk patients were more sensitive to small molecule drugs including Erlotinib, Sunitinib, MG-132, CGP-082996, AZ628, Sorafenib, VX-680, and Z-LLNle-CHO. Risk score was an independent prognostic factor and nomogram confirmed its reliability. Four risk genes (CALB1, CPA3, NOXA1, and TNNT1) had significant positive correlation with their methylation level, while six genes (CCL22, GPRC5B, HSPA1A, MFNG, PABPC1L, and PCOLCE2) were negatively correlated with their methylation level.Conclusion: This study provides novel understanding of heterogeneity in COAD from the perspective of senescence, and develops signatures for prognosis prediction in COAD.