CoxMKF

CoxMKF implements a multiple-knockoff filter that performs high-dimensional mediation analysis for survival outcomes to identify DNA methylation CpG mediators and control the false discovery rate in finite samples.

Key Features:

• Innovative Methodology: Integrates multiple knockoffs with a Cox proportional hazards model to perform mediation analysis on survival outcomes and control FDR in finite samples.
• Stability and Reliability: Aggregates multiple knockoffs to mitigate the randomness of model-X knockoffs and enhance the stability of mediator selection.
• Empirical Validation: Demonstrated by simulation studies to control the false discovery rate in finite-sample settings.
• Application in Epigenetic Research: Applied to a TCGA lung cancer dataset with 754 subjects and over 365,000 DNA methylation CpG sites, identifying four CpG sites mediating the effect of smoking on overall survival.

Scientific Applications:

• Epigenetic mediation analysis: Identify DNA methylation CpG sites that mediate effects of environmental exposures such as smoking on survival outcomes.
• Survival analysis with high-dimensional mediators: Perform mediation inference in Cox models when the number of mediators is large relative to sample size.
• Analysis of cancer cohort datasets: Investigate mediators of exposure effects on overall survival in datasets such as TCGA lung cancer.

Methodology:

Generate knockoff variables that mimic the statistical properties of original mediators without association to the outcome, integrate these knockoffs into a Cox proportional hazards framework, aggregate multiple knockoffs for stability, and control the false discovery rate.