Promoter DNA Methylation Metrics and Gene Expression in Human CD3+ T Cells: Sensitivity to Analytical Choices

Background/Objectives: Promoter DNA methylation is conventionally summarised as a regional mean. Two alternatives have been proposed to capture information the mean discards: the concurrence ratio derived from partially methylated bisulfite reads (CAMDA) and single-CpG Shannon (β-) entropy. Benchmarks of such metrics typically report one coefficient per metric on a single gene set at a single aggregation scale. We asked how far that comparison depends on analytical choices that are rarely reported. Methods: Matched whole-genome bisulfite sequencing (WGBS) and RNA-seq from one donor of human CD3+ T cells were analysed. Promoter windows spanning −1000 to +500 bp around the transcription start site (TSS) were divided into fifteen 100 bp bins; all three metrics were computed on an identical CpG set and correlated with log2(FPKM) by Spearman’s ρ. The comparison was repeated across annotation source, quantification level, coverage threshold, aggregation scheme, the treatment of zero-expression transcripts and CpG island status, with metric differences assessed by paired bootstrap over promoters and over chromosomes. Results: Across a grid of five analytical choices, the mean methylation coefficient ranged from −0.145 to −0.593 and the ordering of the metrics reversed between configurations. At the 100 bp bin scale, the largest difference between metrics was 0.037; aggregating the same CpGs to the whole promoter opened differences of up to 0.31, so the apparent superiority of a metric is itself a function of the aggregation scale. CAMDA exceeded the regional mean consistently but modestly, with the advantage largest in distal upstream bins and indistinguishable from zero adjacent to the TSS. β-Entropy is a strictly monotone function of the folded measure 1 − 2|β − 0.5|; at the 100 bp scale, the mean of per-CpG entropy tracked the folding transform of regional mean methylation to within 0.002 in every bin, while at the whole-promoter scale, the two separated by 0.041. Within CpG islands, the regional mean weakened (ρ = −0.246) while CAMDA did not (−0.494). Conclusions: In this single-donor dataset, analytical design affected the estimated coupling more than the choice of summary metric. The observed range is dataset-specific and should not be interpreted as a transferable numerical estimate. Rather, these results show that evaluating plausible analytical configurations can provide important context when comparing promoter methylation summary metrics.