Filtering High-Dimensional Methylation Marks With Extremely Small Sample Size: An Application to Gastric Cancer Data

DNA methylations in critical regions are highly involved cancer pathogenesis and drug response. However, to identify causal out of a large number potential polymorphic methylation sites is challenging. This high-dimensional data brings two obstacles: first, many established statistical models not scalable so features; second, multiple-test overfitting become serious. To this end, method quickly filter candidate narrow down targets for downstream analyses urgently needed. BACkPAy pre-screening Bayesian approach detect biological meaningful patterns differential levels with small sample size. prioritizes potentially important biomarkers by the false discovery rate (FDR) approach. It filters non-informative (i.e., non-differential) flat pattern across experimental conditions. In work, we applied genome-wide dataset three tissue types each type contains gastric samples. We also LIMMA (Linear Models Microarray RNA-Seq Data) compare its results what achieved BACkPAy. Then, Cox proportional hazards regression were utilized visualize prognostics significant markers The Cancer Genome Atlas (TCGA) survival analysis. Using BACkPAy, identified eight patterns/groups probes from dataset. TCGA data, five prognostic genes predictive progression cancer) that contain some probes, whereas no was using Benjamin-Hochberg FDR LIMMA. showed importance analysis extremely size cancer. revealed RDH13, CLDN11, TMTC1, UCHL1, FOXP2 can serve as treatment promoter level these serum could have diagnostic functions patients.