Synthetic data constructed from real microarray data can be used to assess a marker identification methodology, with class labels, markers, and interaction models handpicked and treated as ground truth.
When assessing the prognostic significance of the markers identified, our analysis revealed several marker dependencies and interactions.
From each analysis, the number of markers and their interactions explaining the variation in gene expression was determined and used in further analysis.
The procedure selects per gene the markers and their interactions for the linear model 1 that would best explain the variation in expression.
Although association-based approaches may ultimately be found superior for identifying cancer markers and their interactions, these methods also have limitations.
Bayesian epistasis association mapping (BEAM) [ 32] treats the disease-associated markers and their interactions via a Bayesian partitioning model and computes, via Markov chain Monte Carlo (MCMC), the posterior probability that each SNP set is associated with the disease.
