|
Abstract:
|
CpG island regions of multiple cancer-specific genes are methylated in CIMP+ patients. An Arizona colorectal cancer methylation study yielded methylation data for up to 4 genes for over 800 patients. Raw methylation scores for each gene consisted of zeros, percentages and missing data, with several different platforms used to measure methylation for some genes. In order to assess overall CIMP status, summarizing propensity for methylation, a single score for each patient is needed. The naïve aggregation methods used in existing literature ignore gene-gene and platform-platform variation in methylation patterns, and fail to appropriately handle missing data. We propose a model-based CIMP score constructed based on a two-step, logistic regression-linear mixed model method. These model-based CIMP scores were more strongly correlated with known clinical factors than naïve CIMP scores.
|
