Simulation based validation

simulation-based-validation

Summary

Generate synthetic methylation data using dataSim() to benchmark and validate statistical methods such as overdispersion-corrected differential methylation analysis. This skill enables controlled testing of analysis parameters and their effects on stringency and accuracy before applying methods to real data.

When to use

When you need to verify that a statistical correction (e.g., overdispersion adjustment in calculateDiffMeth with overdispersion='MN') produces expected changes in test stringency, or when you want to validate that a new analysis workflow produces correct q-value distributions and variance adjustments under known ground-truth conditions.

When NOT to use

• When you have sufficient real experimental replicates and real biological effect sizes — simulation is no substitute for validation on actual data with true biological variance.
• When your primary goal is differential methylation discovery on real samples — simulation-based validation is a method-vetting step, not a substitute for direct analysis.
• When computational resources are severely constrained and you must prioritize analysis speed over methodological validation.

Inputs

• methylKit dataSim() parameters (number of replicates, number of methylation sites)
• methylBase object (simulated or real methylation data with coverage and methylation percentage columns)

Outputs

• methylDiff object with q-values and test statistics from uncorrected run
• methylDiff object with q-values and test statistics from corrected run (overdispersion='MN')
• q-value distribution comparison (mean, median, and percentile values)

How to apply

Use methylKit's dataSim() function to generate a synthetic methylBase object with controlled parameters (e.g., 6 replicates and 1000 methylation sites). Run calculateDiffMeth() on this object twice: once with your method of interest (e.g., overdispersion='MN' with test='Chisq') and once with a baseline uncorrected configuration (overdispersion=FALSE or default). Extract and compare the resulting q-value distributions between runs to verify that the corrected method produces higher average q-values (more stringent multiple-testing correction) and that the variance adjustment factor φ = X²/(N-P) is correctly applied, confirming the expected shift from Chi-square to F-test behavior.

Related tools

• methylKit (Core R package providing dataSim() function for synthetic methylation data generation and calculateDiffMeth() for differential methylation analysis with overdispersion correction) — https://github.com/al2na/methylKit
• R (Runtime environment for executing methylKit functions and statistical comparisons)

Examples

library(methylKit); sim <- dataSim(replicates=6, sites=1000); diff_corrected <- calculateDiffMeth(sim, overdispersion='MN', test='Chisq'); diff_uncorrected <- calculateDiffMeth(sim, overdispersion=FALSE, test='Chisq'); mean(diff_corrected@data$qvalue); mean(diff_uncorrected@data$qvalue)

Evaluation signals

• Mean q-value from overdispersion-corrected run (overdispersion='MN') is higher than uncorrected baseline, confirming more stringent multiple-testing correction
• Statistical test automatically switches from Chi-square (uncorrected) to F-test (corrected with overdispersion='MN'), verifiable via test statistic type in methylDiff object
• Variance scaling factor φ = X²/(N-P) is applied only in the corrected run, evidenced by larger denominator in test statistic calculation
• Q-value distribution shift is consistent across the full range of methylation percentages (not an artifact of a few extreme sites)
• Simulated data with known effect sizes confirms that corrected method maintains or improves specificity while adjusting for overdispersion relative to binomial expectations

Limitations

• Simulated data from dataSim() may not capture all real-world sources of variance and batch effects present in actual bisulfite-seq experiments.
• Validation on simulation alone does not guarantee equivalent performance on real data with complex sample heterogeneity, sequencing errors, or technical biases.
• The overdispersion='MN' correction assumes methylation-specific negative binomial distribution; if your data violates this assumption, validation results may not transfer to real analysis.

Evidence

• [other] Generate a simulated methylBase object using methylKit's dataSim() function with parameters set to 6 replicates and 1000 methylation sites.: "Generate a simulated methylBase object using methylKit's dataSim() function with parameters set to 6 replicates and 1000 methylation sites."
• [other] Extract and compare the q-value distributions from the corrected run (overdispersion='MN') against a parallel uncorrected run (overdispersion=FALSE or default) to verify that the corrected method produces higher average q-values: "Extract and compare the q-value distributions from the corrected run (overdispersion='MN') against a parallel uncorrected run (overdispersion=FALSE or default) to verify that the corrected method"
• [other] The calculateDiffMeth() function with overdispersion='MN' parameter calculates a scaling parameter φ = X²/(N-P) to adjust variance as φ·n_i·π̂_i·(1-π̂_i), which makes statistical tests more stringent by correcting for variance in excess of binomial expectations and automatically switches from Chisq to F-test.: "The calculateDiffMeth() function with overdispersion='MN' parameter calculates a scaling parameter φ = X²/(N-P) to adjust variance as φ·n_i·π̂_i·(1-π̂_i), which makes statistical tests more stringent"
• [readme] methylKit is an R package for DNA methylation analysis and annotation from high-throughput bisulfite sequencing: "methylKit is an R package for DNA methylation analysis and annotation from high-throughput bisulfite sequencing."