Multi-Omics Prognostic Marker Discovery and Survival Modeling: A Case Study on Pan-Cancer Survival Analyses in Women’s Cancers
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Survival analysis is essential for predicting patient outcomes and guiding personalized cancer treatments. While multi-omics data offers valuable insights, its high dimensionality complicates analysis and clinical application. Many studies still rely on the traditional Cox proportional hazards model, with limited exploration of alternative survival algorithms or robust feature selection methods. Few frameworks effectively integrate features across multiple omics modalities. To address these issues, we developed PRISM (PRognostic marker Identification and Survival Modelling through Multi-omics Integration), a comprehensive framework designed to improve survival predictions and identify key prognostic markers. PRISM systematically compares various feature selection methods and survival models and employs a robust pipeline that selects features from single-omics data, integrating them through feature-level fusion and multi-stage refinement. Applied to TCGA data for Breast Invasive Carcinoma (BRCA), Cervical Squamous Cell Carcinoma and Endocervical Adenocarcinoma (CESC), Ovarian Serous Cystadenocarcinoma (OV) and Uterine Corpus Endometrial Carcinoma (UCEC), PRISM demonstrated that integrating DNA methylation, miRNA, and copy number variation data provided complementary information, significantly outperforming other modality combinations in three cancers (BRCA: C-index 0.77; CESC: 0.80; UCEC: 0.76). Pan-cancer analysis further revealed shared oncogenic pathways and therapeutic targets. PRISM provides a scalable, generalizable solution for multi-omics integration, advancing cancer research and precision medicine.
Key Points
-
Advancing Multi-Omics Survival Modelling – PRISM introduces a robust computational framework that enhances survival predictions and identifies key prognostic markers by systematically integrating multi-omics data.
-
Rigorous Feature Selection and Benchmarking – PRISM systematically evaluates multiple feature selection techniques and survival models using a voting-based approach to improve methodological robustness.
-
Optimized and Cost-Effective Signature Panel – By employing recursive feature elimination (RFE), PRISM minimizes the number of selected biomarkers, improving cost-effectiveness while maintaining high predictive accuracy for clinical applications.
-
Comprehensive Multi-Omics Integration – PRISM enables the fusion of features across diverse omics modalities using single-stage and two-tier refinement strategies, optimizing the final prognostic signature for maximum predictive power.
-
Scalability and Clinical Relevance – PRISM is generalizable beyond TCGA datasets, making it applicable to large-scale multi-omics studies with broad clinical and biomedical implications for personalized medicine and cancer research.
