De novo identification of differentially methylated regions in the human genome

Table 1 Usage of the four candidate methods in this study

Method	Implementationlanguage	Relevant parameters used in call ^a	Computational time (to nearest second) ^b
DMRcate v1.2.0	R	lambda = 1000, C=2, p.adjust.method = ~BH~, pcutoff = α	129
Bumphunter from minfi v1.10.2	R	cluster = NULL, maxGap = 1000, smooth = TRUE, smoothFunction = loessByCluster, cutoff = α	58
comb-p v0.35	Python	–dist 1000 –seed α	96
Probe Lasso from ChAMP v1.2.7	R	filterXY = F, mafPol.lower = 0, mafPol.upper = 1, lassoRadius = 1000, minDmrSep = 1000, adjPVal = 1, minSigProbesLasso = 1, DMRpval = α	6,867

^aArgument α indicates the tuned parameter for which sets of called DMRs were produced. Any omissions imply function defaults. ^bComputational time is for the VA vs SA comparison, which is the time needed to fit the whole 450K methylome (using parameters in the previous column) with a single core of a clean Dual Xeon eight-core E5-2650 compute node, using 64 GB of virtual memory.

ISSN: 1756-8935