Abstract
BACKGROUND: Multi-omics integration may provide additional information about the development of tumors and improve the performance of predictive models. The key challenge lies in integrating several omics sources, especially to capture their biological relationships. Previous studies proposed a structural equation model framework to combine two data platforms for predicting survival; however, several limitations remain. RESULTS: In this study, we introduce an extended Bayesian survival model combined with a structural equation model for adaptation to broader applications. The No U-turn Sampling (NUTS) algorithm was utilized to efficiently sample the posterior distribution of model parameters. Through a series of simulation studies, our model showed excellent goodness-of-fit and predictive performance. To validate the efficiency of our model, we utilized a gastric cancer dataset with three omics types (mRNA, microRNA, and methylation) obtained from The Cancer Genome Atlas. After bioinformatic processing, we included six mRNA, microRNA, and methylation loci datasets into the framework and discovered that our model exhibited greater predictive performance compared to non-integrated and Integrative Bayesian Analysis of Genomics (iBAG) models. CONCLUSIONS: In conclusion, our extended Bayesian structural equation model for multi-omics survival analysis provides a robust framework that significantly enhances predictive accuracy by effectively capturing complex biological relationships across diverse omics data sources, demonstrating clear advantages over both non-integrated approaches and existing integrative methods like iBAG.