eng Oriental Scientific Publishing Company Biomedical and Pharmacology Journal 0974-6242 2025-10-21 18 October Spl Edition 68462 article Muneeba Afzal Mukhdoomi 1 Manzoor Ahmad Chachoo 1 Department of Computer Sciences, University of Kashmir, Srinagar, India, Decoding the molecular heterogeneity of cancer is fundamental to the application of precision oncology, but multi-omics data are inherently noisy, high-dimensional and incomplete, making it difficult to find robust subtypes. We propose an integrative computational framework that integrates four complementary modules: robust correlation estimation to noise these patient similarity networks, UMAP for nonlinear dimensionality reduction, p-Laplacian Hypergraph construction to capture higher-order relations, and Mapper-based topological data analysis to identify shape-driven patient subgroups. This design is modular, scalable and resilient to the absence of particular omics modalities, allowing both local and global structures to contribute to clinically relevant subtyping. We validated the framework in five TCGA cohorts, GBM, BRCA, LUAD, KIRC, and COAD, based on log-rank p-values, Restricted Life Expectancy Difference RLED and silhouette measures. The method was consistently more effective than established techniques such as SNF, NEMO and RSC-OTRI. In GBM, it resulted in a survival separation of 221 days with the log-rank p-value of 0.0006 and a silhouette score of 0.58. Strong stratification was also evident in the BRCA and LUAD cohorts, where RLED gains were 174 and 146 days, respectively, and the silhouette scores were >0.52. Ablation studies confirmed the need for each module, as excluding robust correlation decreased GBM RLED from 221 to 167 days, replacing UMAP severely decreased clustering quality and excluding Mapper decreased survival stratification. Pathway enrichment analyses supported the biological significance of the subtypes and associated them with PI3K-Akt signalling, hypoxia response, and ER+/HER2 pathways. Overall, this framework provides a powerful, interpretable and clinically versatile approach to multi-omics cancer subtyping, with potential to drive the advancement of patient stratification and/or guide precision oncology interventions. https://biomedpharmajournal.org/vol18octoberspledition/multi-omics-cancer-subtyping-with-robust-correlation-umap-and-topological-hypergraph-learning/ Cancer subtyping Multi-omics integration p-Laplacian Hypergraph Robust correlation Survival analysis TCGA Topological data analysis UMAP