Joint Variable Selection for Omic Biomarkers in Time-to-Event Data

VampW, a Bayesian method, reduces false discoveries and improves disease onset prediction by jointly modeling correlated omics features in time-to-event data.

• Why it matters: Accurate identification of genetic and biological pathways involved in disease onset is crucial for understanding complex diseases, but current methods often ignore correlations and increase false positives.
• What they did: The authors developed vampW, a fully parametric Bayesian approach based on Vector Approximate Message Passing applied to a Weibull model, and tested it on large-scale biobank data with over 53,000 individuals.
• The result: VampW identified 219 protein associations with disease outcomes, many not found by marginal testing, and reduced prediction error by over 30% compared to CoxPH and DeepSurv, demonstrating its effectiveness for complex disease analysis.