Make An Appointment
Make An Appointment
ttScreening
ttScreening is a feature selection method designed to identify important CpG sites across multiple iterations using t-tests. The method evaluates the stability and reproducibility of selected features by applying statistical thresholds (e.g., p < 0.05 or p < 0.1) across repeated subsampling.
The plots compare different significance thresholds and correction methods (TT, TTsame, FDR, and Bonferroni) under various scenarios where the number of truly important CpG sites ranges from 10 to 400 out of 2000. The Y-axis shows the number of falsely identified CpG sites, while the X-axis represents the cutoff percentage—how often a site must be selected across iterations to be considered important.
Results suggest that:
- Mid-range iteration cutoffs (e.g., 50–70%) help reduce false positives.
- Bonferroni is the most conservative method but may miss important features.
- ttScreening with looser thresholds (TT: 0.05/0.1) performs well when the number of relevant CpG sites is moderate to high.
- Method effectiveness depends on the true signal density, and tuning is important for balancing sensitivity and specificity.
Ray MA, Tong X, Lockett GA, Zhang H, Karmaus WJ. An Efficient Approach to Screening Epigenome-Wide Data. Biomed Res Int. 2016;2016:2615348. doi: 10.1155/2016/2615348. Epub 2016 Mar 13. PMID: 27034928; PMCID: PMC4808532.